Prune the uv cache at the end of every CI job

The uv cache lives on a persistent volume shared by every job on the
runner (/srv/act-runner-cache/uv), so nothing trimmed it and it grew
without bound. `uv cache prune --ci` drops the entries that are not worth
keeping between runs (pre-built wheels for local sources) while leaving
the downloaded-wheel cache that makes `uv sync` fast.

In bump-version the step carries the same is_merge guard as the rest of
that job, since uv is only set up on a merge push.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.3.15"
+current_version = "0.3.10"
 parse = "(?P<major>\\d+)\\.(?P<minor>\\d+)\\.(?P<patch>\\d+)"
 serialize = ["{major}.{minor}.{patch}"]
 search = "{current_version}"
@@ -8,7 +8,7 @@ regex = false
 allow_dirty = false
 commit = true
 tag = false
-message = "chore: bump version {current_version} -> {new_version}"
+message = "chore: bump version {current_version} -> {new_version} [skip ci]"
 pre_commit_hooks = ["uv lock", "git add uv.lock"]
 
 [[tool.bumpversion.files]]

.gitea/workflows/ci.yml

@@ -2,9 +2,10 @@ name: CI
 
 "on":
   push:
-    branches: ["master"]
+    branches: ["**"]
     tags: ["**"]
-  pull_request: {}
+  pull_request:
+    branches: [master]
 
 env:
   UV_CACHE_DIR: /uv-cache
@@ -27,6 +28,9 @@ jobs:
           echo "UV_LINK_MODE=copy" >> "$GITHUB_ENV"
       - run: uv sync --extra cpu --extra dev
       - run: uv run ruff check .
+      # The uv cache is a persistent volume shared by every job on this
+      # runner, so each job trims what it no longer needs before exiting.
+      - run: uv cache prune --ci
 
   ruff-format:
     name: Format (ruff format)
@@ -45,6 +49,7 @@ jobs:
           echo "UV_LINK_MODE=copy" >> "$GITHUB_ENV"
       - run: uv sync --extra cpu --extra dev
       - run: uv run ruff format --check .
+      - run: uv cache prune --ci
 
   type-check:
     name: Type check (ty)
@@ -63,6 +68,7 @@ jobs:
           echo "UV_LINK_MODE=copy" >> "$GITHUB_ENV"
       - run: uv sync --extra cpu --extra dev
       - run: uv run ty check .
+      - run: uv cache prune --ci
 
   test:
     name: Tests
@@ -86,6 +92,7 @@ jobs:
         with:
           name: coverage-report
           path: coverage.xml
+      - run: uv cache prune --ci
 
   bump-version:
     name: Bump version, tag, and update changelog on merge to master
@@ -155,7 +162,7 @@ jobs:
             uv run git-cliff --tag "$TAG" --unreleased --prepend CHANGELOG.md
             git add CHANGELOG.md
             if ! git diff --cached --quiet -- CHANGELOG.md; then
-              git commit -m "chore: update changelog for $TAG"
+              git commit -m "chore: update changelog for $TAG [skip ci]"
             else
               git restore --staged CHANGELOG.md
             fi
@@ -172,6 +179,10 @@ jobs:
             git tag -a "$TAG" -m "$TAG"
             git push origin "refs/tags/$TAG"
           fi
+      # Same guard as every other step here: on a non-merge push uv was never
+      # set up, so there is nothing to prune.
+      - run: uv cache prune --ci
+        if: steps.merge_check.outputs.is_merge == 'true'
 
   sync-version-on-tag:
     name: Sync project version with tag
@@ -192,7 +203,7 @@ jobs:
             git config user.name "gitea-actions"
             git config user.email "actions@git.larsbogner.de"
             git add pyproject.toml uv.lock
-            git commit -m "chore: sync project version to tag ${GITHUB_REF_NAME}"
+            git commit -m "chore: sync project version to tag ${GITHUB_REF_NAME} [skip ci]"
             git push origin HEAD:master
             git push origin ":refs/tags/${GITHUB_REF_NAME}"
             git tag -f "${GITHUB_REF_NAME}" HEAD
@@ -200,33 +211,4 @@ jobs:
           else
             echo "Tag version matches project version ($CURRENT_VERSION)"
           fi
-
-  publish-package:
-    name: Publish package to Gitea package registry
-    needs: [ruff-check, ruff-format, type-check, test, sync-version-on-tag]
-    if: startsWith(github.ref, 'refs/tags/')
-    runs-on: ubuntu-latest
-    container:
-      image: docker.gitea.com/runner-images:ubuntu-latest
-      volumes:
-        - /srv/act-runner-cache/uv:/uv-cache
-    steps:
-      # Check out by tag name (not the triggering SHA) since sync-version-on-tag
-      # may have force-moved the tag to a version-corrected commit.
-      - uses: actions/checkout@v4
-        with:
-          ref: ${{ github.ref_name }}
-      - uses: astral-sh/setup-uv@v5
-        with:
-          enable-cache: false
-      - run: |
-          echo "UV_CACHE_DIR=/uv-cache" >> "$GITHUB_ENV"
-          echo "UV_LINK_MODE=copy" >> "$GITHUB_ENV"
-      - run: uv build
-      # CI_TOKEN needs write:package scope (in addition to write:repository,
-      # used elsewhere) for this upload to authenticate.
-      - run: |
-          uv publish \
-            --publish-url "https://git.larsbogner.de/api/packages/lars/pypi" \
-            --username gitea-actions \
-            --password "${{ secrets.CI_TOKEN }}"
+      - run: uv cache prune --ci

CHANGELOG.md

@@ -1,37 +1,5 @@
 # Changelog
 
-## [0.3.15] - 2026-08-28
-
-### Changed
-
-- Perf: defer heavy imports in giant/dwarf CLIs until commands run
-
-## [0.3.14] - 2026-08-28
-
-### Changed
-
-- Ci: give automated commits visible checks, scope CI triggers, publish releases
-
-- Ci: fix pull_request trigger not registering
-
-## [0.3.13] - 2026-08-28
-
-### Added
-
-- Add inference-time model_config overrides with a sampling-key allowlist [gitea #87](https://git.larsbogner.de/lars/giant/issues/87)
-
-## [0.3.12] - 2026-08-28
-
-### Added
-
-- Add sampled n_sec under n_sec.mode = 'head' [gitea #86](https://git.larsbogner.de/lars/giant/issues/86)
-
-## [0.3.11] - 2026-08-26
-
-### Changed
-
-- Feat(analysis): per-step secondary multiplicity plots
-
 ## [0.3.10] - 2026-08-26
 
 ### Changed

CLAUDE.md

@@ -9,6 +9,7 @@ uv sync --extra cpu                # install dependencies with CPU-only torch (s
 uv sync --extra cuda               # install dependencies with CUDA 11.8 torch
 uv sync --extra cpu --extra dev    # add dev extras (pytest, ruff, ty, bump-my-version, git-cliff, + all runtime extras)
 uv sync --extra cpu --extra geometry  # add scikit-learn for the geometry oracle (giant rollout)
+uv sync --extra cpu --extra workflow  # add b2luigi for `giant workflow` pipeline orchestration
 pytest               # run tests
 giant new-run --hidden-dim 512 --lr 3e-4        # scaffold a config.toml + run dir ahead of training
 giant train path/to/steps.parquet               # train (defaults: stage 1 flow, stage 2 wgan + autoregressive)
@@ -18,8 +19,9 @@ giant train path/to/steps.parquet --router --router-type energy  # MoE routing t
 giant model summary --config config.toml         # build-only: parameter counts + which config keys actually bite
 giant predict path/to/steps.parquet --checkpoint ckpt/best.pt   # per-step predictions
 giant rollout path/to/steps.parquet --checkpoint ckpt/best.pt --geometry oracle.pkl  # full showers
-giant analyze submit rollout.yaml --accounting-group cms   # parallel rollout-vs-reference analysis on HTCondor
-giant analyze render <run_dir> --gallery                   # render PDFs + HTML gallery (run_dir from prep/submit)
+giant workflow run spec.toml --batch --workers 20          # whole pipeline (cache-warm -> train -> rollout -> analysis)
+giant analyze prep rollout.yaml --chunks 32                # lay out an analysis run dir (compute jobs come from the workflow)
+giant analyze render <run_dir> --gallery                   # render PDFs + HTML gallery (run_dir from prep)
 giant analyze metrics <train_run_dir>                      # training-progress plots from metrics.csv
 dwarf --help                                     # dataset/tooling CLI: convert, migrate, bump-gen,
                                                   # bump-schema, status, update-manifest, create-manifest,
@@ -51,7 +53,7 @@ Work on this repo happens across three kinds of machine:
 
 - **Local dev machines** (laptop + desktop, identical): repo at `~/Programming/giant`, no access to `/ceph` — datasets, training results, and models aren't reachable here.
 - **Portal machines** (`portal1`, `deepthought`, `deepthought2`, `bms1`, `bms2`, `bms3`): repo lives under `/work`, and `/ceph` holds ROOT/parquet files and trained models. **These are shared with other users** — stay strictly within `/work/lbogner` and `/ceph/lbogner`, and keep resource usage to roughly a quarter of CPU/RAM and a single GPU so as not to disturb other users' jobs.
-- **HTCondor worker nodes**: never run or SSH onto these directly — the only sanctioned path is submitting jobs through condor (`giant analyze submit`, and the in-progress remote-GPU train/rollout submission on `condor-gpu-train-rollout`). `/ceph` is available there; `/work` is only sometimes mounted, depending on the node.
+- **HTCondor worker nodes**: never run or SSH onto these directly — the only sanctioned path is `giant workflow run <spec.toml> --batch` (b2luigi, see the Workflow section), which submits and polls every job. `/ceph` is available there; `/work` is only sometimes mounted, depending on the node.
 
 ## Architecture
 
@@ -97,9 +99,11 @@ Secondary energies are a **stick-breaking partition of the `e_sec` budget** from
 
 **Validation** (`giant/validate.py`): step-level marginal + KL-divergence comparisons during training (`--validate-every`).
 
-**Analysis** (`giant/analysis/`, `giant analyze` CLI): a lean, streaming rollout-vs-reference plotting pipeline that compares one or more autoregressive `giant rollout` runs against a single held-out miniCaloSim reference steps file shared by all of them, and produces publication-styled PDFs assembled into an HTML gallery — one distinctly colored series per rollout, one reference line/panel. It exploits the fact that rollout output and a raw reference file share a world-frame physical column subset under identical names (`pre_*`/`post_*`/`edep`/`step_length`/`pdg`/`material`/`event_id`), so no ALR/local-frame decode is needed — everything is world-frame mm/MeV. Structure: `sources.py` (canonical LazyFrames + `RolloutSpec`/`Side` — a rollout's opened frames + per-checkpoint diagnostic inputs — + synthetic-termination-row filtering + the secondary view, which is `generation>0 & step_no==0` rollout tracks vs exploded `sec_*_list` reference columns), `variables.py` (the per-step value expressions shared by range sizing and the plot registry), `reduce.py` (the streaming primitives — a single `hist1d` `group_by([group,bin]).len()` pass, per-event scalars, edep-weighted depth/transverse profiles, species share, leakage), `grouping.py`/`context.py` (fixed bin edges + energy-quantile/pdg/material group sets resolved once by `prep` into `shared.json` over the union of the reference and every rollout, so every compute job is one pass with no range scan), `reduced.py` (`Partial`/`Reduced` — the compact self-describing JSON a compute job emits), `catalog.py` (the declarative `PlotSpec` registry — marginals × {overall,energy,pdg,material}, per-event totals, shower profiles/containment, species/leakage, secondaries, distance summaries, router and type-embedding diagnostics; `giant analyze list` prints every id), `runtime_estimate.py` (per-(plot, chunk) walltime estimates for the submit description), and `render.py` (the only module importing ETPlot's `plotstyle`/LaTeX; dispatches on `Reduced.kind`, writes PDFs + `metadata.yaml`; each rollout gets a stable `ps.get_color(i)` slot by its position in `series`, the reference always draws in one fixed dashed-ink style). `Bundle.rollouts` is a name-keyed dict of `Side`, and every `compute_partial`/`finalize` builds a `Reduced.payload["series"]` dict keyed the same way, with `payload["reference"]` as the one distinguished non-rollout entry. The heatmap-shaped specs (`marginal_distance_summary`, `sec_count_per_step_by_species` — the latter also drawing the reference as its own panel) and the checkpoint-bound diagnostics (`router_gating.py`, `type_embedding_distance.py`) are inherently one-matrix/one-checkpoint per rollout, so they render as one panel per rollout instead of one line/bar per rollout.
+**Analysis** (`giant/analysis/`, `giant analyze` CLI): a lean, streaming rollout-vs-reference plotting pipeline that compares one or more autoregressive `giant rollout` runs against a single held-out miniCaloSim reference steps file shared by all of them, and produces publication-styled PDFs assembled into an HTML gallery — one distinctly colored series per rollout, one reference line/panel. It exploits the fact that rollout output and a raw reference file share a world-frame physical column subset under identical names (`pre_*`/`post_*`/`edep`/`step_length`/`pdg`/`material`/`event_id`), so no ALR/local-frame decode is needed — everything is world-frame mm/MeV. Structure: `sources.py` (canonical LazyFrames + `RolloutSpec`/`Side` — a rollout's opened frames + per-checkpoint diagnostic inputs — + synthetic-termination-row filtering + the secondary view, which is `generation>0 & step_no==0` rollout tracks vs exploded `sec_*_list` reference columns), `variables.py` (the per-step value expressions shared by range sizing and the plot registry), `reduce.py` (the streaming primitives — a single `hist1d` `group_by([group,bin]).len()` pass, per-event scalars, edep-weighted depth/transverse profiles, species share, leakage), `grouping.py`/`context.py` (fixed bin edges + energy-quantile/pdg/material group sets resolved once by `prep` into `shared.json` over the union of the reference and every rollout, so every compute job is one pass with no range scan), `reduced.py` (`Partial`/`Reduced` — the compact self-describing JSON a compute job emits), `catalog.py` (the declarative `PlotSpec` registry — marginals × {overall,energy,pdg,material}, per-event totals, shower profiles/containment, species/leakage, secondaries, distance/confusion summaries, router and type-embedding diagnostics; `giant analyze list` prints every id), `runtime_estimate.py` (per-(plot, chunk) walltime estimates for the submit description), and `render.py` (the only module importing ETPlot's `plotstyle`/LaTeX; dispatches on `Reduced.kind`, writes PDFs + `metadata.yaml`; each rollout gets a stable `ps.get_color(i)` slot by its position in `series`, the reference always draws in one fixed dashed-ink style). `Bundle.rollouts` is a name-keyed dict of `Side`, and every `compute_partial`/`finalize` builds a `Reduced.payload["series"]` dict keyed the same way, with `payload["reference"]` as the one distinguished non-rollout entry. The heatmap-shaped specs (`marginal_distance_summary`, `n_sec_confusion`) and the checkpoint-bound diagnostics (`router_gating.py`, `type_embedding_distance.py`) are inherently one-matrix/one-checkpoint per rollout, so they render as one panel per rollout instead of one line/bar per rollout.
 
-**Input is one or more `giant rollout` YAML sidecars** (`condor.py:load_rollout_yamls`): each YAML's `output`/`dataset` keys name its rollout parquet and seed file (= the reference truth); every supplied YAML must resolve to the same `dataset`, checked up front with a clear error otherwise (the premise is "N candidates vs one ground truth"). Each rollout's series name comes from a repeated `--label` CLI flag, else the YAML stem (N>1), else `"rollout"` (a single YAML). `prep` creates a **run directory** (`<cwd>/analysis_runs/analysis_<id>/` by default, `--run-dir` to override) holding `shared.json`, `run_meta.json` (`RunMeta.rollouts: list[{name,path,plot_meta}]`, insertion order = CLI order = every plot's series order), `reduced_partial/`, `reduced/`, `plots/`. **Compute/merge/render split:** `giant analyze submit a.yaml [b.yaml ...] --chunks N` runs `prep` (recording `N` in `run_meta.json`) then submits one HTCondor job per (plot, chunk) pair (`compute-one --id --chunk --run-dir`, polars/numpy only — no LaTeX on workers), each streaming over an `event_id`-disjoint slice (`event_id % N == chunk`) of the reference **and every rollout** and writing a small `reduced_partial/<id>__<chunk>.json`; every `PlotSpec` splits into a `compute_partial`/`finalize` pair so chunks can be summed/concatenated back per rollout (`chunkable=False` specs — the checkpoint-bound diagnostics, already bounded/subsampled — always run as a single chunk). The local `giant analyze render <run_dir>` first joins every plot's chunk partials into `reduced/<id>.json` (`merge_all`, a no-op join when `N=1`; `merge-one` does a single plot for debugging), then turns those into the styled PDF/gallery tree. `giant analyze metrics <train_run_dir>` is a separate, unrelated entry point: training-progress plots straight from a run's `metrics.csv`.
+**Input is one or more `giant rollout` YAML sidecars** (`run.py:load_rollout_yamls`): each YAML's `output`/`dataset` keys name its rollout parquet and seed file (= the reference truth); every supplied YAML must resolve to the same `dataset`, checked up front with a clear error otherwise (the premise is "N candidates vs one ground truth"). Each rollout's series name comes from a repeated `--label` CLI flag, else the YAML stem (N>1), else `"rollout"` (a single YAML). `prep` creates a **run directory** (`<cwd>/analysis_runs/analysis_<id>/` by default, `--run-dir` to override) holding `shared.json`, `run_meta.json` (`RunMeta.rollouts: list[{name,path,plot_meta}]`, insertion order = CLI order = every plot's series order), `reduced_partial/`, `reduced/`, `plots/`. **Compute/merge/render split:** `giant analyze prep a.yaml [b.yaml ...] --chunks N` records `N` in `run_meta.json`, and the workflow's `AnalysisComputeTask` runs one HTCondor job per (plot, chunk) pair (`compute-one --id --chunk --run-dir`, polars/numpy only — no LaTeX on workers), each streaming over an `event_id`-disjoint slice (`event_id % N == chunk`) of the reference **and every rollout** and writing a small `reduced_partial/<id>__<chunk>.json`; every `PlotSpec` splits into a `compute_partial`/`finalize` pair so chunks can be summed/concatenated back per rollout (`chunkable=False` specs — the checkpoint-bound diagnostics, already bounded/subsampled — always run as a single chunk). The local `giant analyze render <run_dir>` first joins every plot's chunk partials into `reduced/<id>.json` (`merge_all`, a no-op join when `N=1`; `merge-one` does a single plot for debugging), then turns those into the styled PDF/gallery tree. `giant analyze metrics <train_run_dir>` is a separate, unrelated entry point: training-progress plots straight from a run's `metrics.csv`.
+
+**Workflow orchestration** (`giant/workflow/`, `giant workflow run` CLI): b2luigi is the **only sanctioned way to run a multi-step pipeline**; `giant`/`dwarf` are single-step primitives the tasks invoke. One workflow TOML (`configs/workflow_example.toml`) parameterises a whole experiment — `[workflow]`/`[condor]`/`[dataset]`/`[geometry]` plus repeated `[[train]]`/`[[rollout]]`/`[[analysis]]` tables, each cross-referenced by name — and `spec.py` parses it into frozen dataclasses, rejecting unknown keys and dangling references. Every task's output directory is `<result_dir>/<kind>/name=<name>/spec_hash=<hash>/…`, where the 8-hex `spec_hash` covers that task's resolved sub-spec **and its transitive parents**, so an edited spec re-runs exactly the affected subtree instead of silently reusing stale outputs. The DAG (`tasks.py`): `DatasetTask` (external, fails fast if `/ceph` isn't mounted) → `WarmCacheTask` / `GeometryOracleTask` → `TrainEpochTask(name, milestone)` → `TrainTask` → `RolloutTask` → `AnalysisPrepTask` → `AnalysisComputeTask(name, plot_id, chunk)` → `AnalysisRenderTask` → `WorkflowTask`. Training is fanned out into **one short GPU job per epoch** (`epochs_per_job` trades queue waits back), chained by `--resume` on the previous job's `last.pt` — the loop already handles that unchanged — and `TrainTask` republishes `best.pt`/`last.pt`/a concatenated `metrics.csv` so nothing downstream sees the fan-out. `StreamingStepsDataset.set_epoch` and `config.epoch_seed` (both applied per epoch by `training/loop.py`) derive the batch order and the global RNG state from `(seed, epoch)`, so epoch *k* is bit-identical either way — verified by diffing a chained run's `metrics.csv` against a single 3-epoch `giant train`. `AnalysisRenderTask` is always local (the only step importing plotstyle/LaTeX); `htcondor.py` holds the CPU/GPU submit settings, with the GPU requirement strings (`TARGET.ProvidesEtpCeph` + device/memory pins) ported from the `condor-gpu-train-rollout` branch. `run.py` is the script b2luigi re-executes on workers (`--spec` forwarded via `task_cmd_additional_args`, so a worker resolves the identical graph); `giant workflow run` is a thin exec of it. Needs `uv sync --extra cpu --extra workflow`.
 
 **Shower rollout** (`giant/rollout.py`, `giant rollout` CLI): autoregressively steps the two-stage model into a full shower, advancing tracks breadth-first (every sweep steps all active tracks once, in `batch_size` chunks, so many tracks share each forward pass). Each primary post-step becomes the next pre-step, secondaries are pushed as new tracks, and per-step `material`/`layer_id` come from a `GeometryOracle` (`giant/geometry.py`, built via `dwarf build-geometry-oracle`) that learns position → (material, layer_id) from data and flags detector escape by nearest-neighbour distance. Tracks terminate on one of the `TERM_*` reasons in `constants.py` (energy cutoff, max steps, escape, natural end, unknown pdg, max tracks); energy is deposited locally on every stop except escape (leakage), so showers conserve energy by construction. `giant/checkpoint_io.py` is the shared checkpoint → ready-to-run-models path used by both `predict` and `rollout`.
 
@@ -113,8 +117,6 @@ Secondary energies are a **stick-breaking partition of the `e_sec` budget** from
 
 **v0.3.0 — Stage-2 autoregressive redesign (implemented, released; on `master` since 2026-08-13):** motivated by the 2026-08-03 WGAN rollout benchmark, which failed specifically at the secondary-species level (zero photon secondaries, ~4M hallucinated `-14` muon antineutrinos). Stage 2 became autoregressive in descending-energy order with teacher forcing, and the particle-type representation went back to **categorical** (`particle_type.target = "onehot"`), reversing the 2026-07-17 continuous `(log-mass, charge)` target. The config break (`[conditioning]`/`[stage1_model]`/`[stage2_model]`/`[train]` replacing the flat `train.mode` + `[model]`) makes per-stage generators, stage-2-only training, and one-shot-vs-autoregressive comparison all expressible, and the `network.py` refactor into composable parts (encoder × trunk × objective) also makes routed WGAN work for the first time.
 
-**Baseline benchmark (done, 2026-08-26):** `configs/baseline.toml`'s first full rollout-vs-Geant4 validation (`analysis_341dfb14`, checkpoint `20260814_1743_s2-flow_h512_s2h512_bs36864_ep50/best.pt`, epoch 50/50). Confirms the v0.3.0 pivot fixed the species collapse — zero photon secondaries / hallucinated `-14` muon antineutrinos are both gone (γ at 95% of truth, no `-14` in the top species) — and rules out `conditioning.*.type = "physical"` as the cause, since this checkpoint pairs it with `flow`/no-router and still doesn't collapse. Bulk shower observables are close to Geant4 (total deposited energy +1.9%, containment depth-90%/95% both 0.986×), but steps/event now *over*-shoots by 1.32× (the opposite sign from every pre-v0.3.0 checkpoint), no hadronic/nuclear secondaries are produced at all, and event-to-event energy variance is ~16× too narrow. Writeup: `/home/lars/knowledge-base/experiments/giant-baseline-flow-ar-rollout-validation.md`.
-
 v0.2 configs and checkpoints are auto-migrated (`config.migrate_config`, `model._legacy._migrate_legacy_model_config`, both drawing on shared facts in `giant/_migration.py`). **v0.2 checkpoint-loading support has no expiry decided yet**: `/ceph` still holds pre-v0.3.0 checkpoints and analysis runs referencing them, so don't delete or substantially alter either migration function or `tests/legacy/network_v02_snapshot.py` (the frozen v0.2 snapshot they're tested against) without an explicit decision to do so first.
 
 **Faster-eval architectures — both implemented, neither validated.** Target is a ~10× native-Geant4 eval budget; no eval-latency number exists for any configuration yet, so that budget is unverified across the board.
@@ -123,4 +125,4 @@ v0.2 configs and checkpoints are auto-migrated (`config.migrate_config`, `model.
 
 A sampling-calorimeter (multi-material) dataset track is still open and unblocked, not yet started. See the knowledge base (`/home/lars/knowledge-base/meta/roadmap.md`).
 
-**Condor-submitted GPU training/rollout (in progress, `condor-gpu-train-rollout` branch, not yet merged):** moves `giant train`/`giant rollout` off the shared portal GPU dev machines (see Compute environment) onto remote-GPU HTCondor submission on TOpAS/NEMO2. Partway between "needs major features" and feature-complete — not ready to merge yet.
+**Condor-submitted GPU training/rollout (`condor-gpu-train-rollout` branch, superseded):** its goal — moving `giant train`/`giant rollout` off the shared portal GPU dev machines onto remote-GPU HTCondor submission — is now met by the b2luigi workflow above. Its `train-submit`/`rollout-submit` commands are deliberately **not** ported and must not be revived when that branch is eventually merged; the only part that survived is `_gpu_requirements`, which moved into `giant/workflow/htcondor.py`.

README.md

@@ -110,10 +110,15 @@ giant/
 │   │   ├── reduced.py      #   Partial/Reduced — the compact JSON a compute job emits
 │   │   ├── catalog.py      #   declarative PlotSpec registry (`giant analyze list`)
 │   │   ├── router_gating.py / type_embedding_distance.py   # checkpoint-bound diagnostics
-│   │   ├── runtime_estimate.py  #   per-(plot, chunk) walltime estimates for submit
-│   │   ├── condor.py       #   prep / compute-one / merge / submit-description plumbing
+│   │   ├── runtime_estimate.py  #   per-(plot, chunk) walltime estimates for the job requests
+│   │   ├── run.py          #   prep / compute-one / merge plumbing
 │   │   └── render.py       #   PDFs + HTML gallery (only module importing plotstyle/LaTeX)
-│   └── cli.py              # `giant train` / `new-run` / `model summary` / `predict` / `rollout` / `analyze`
+│   ├── workflow/           # b2luigi pipeline orchestration (`giant workflow run spec.toml`)
+│   │   ├── spec.py         #   workflow TOML -> frozen dataclasses, validation, per-task spec hashes
+│   │   ├── htcondor.py     #   CPU/GPU submit settings (docker image, +RemoteJob, GPU requirements)
+│   │   ├── tasks.py        #   the task graph: cache-warm -> train (one job/epoch) -> rollout -> analysis
+│   │   └── run.py          #   the script b2luigi re-executes on every worker
+│   └── cli.py              # `giant train` / `new-run` / `model summary` / `predict` / `rollout` / `analyze` / `workflow`
 ├── giant/tools/                        # dataset/tooling logic, unified under the `dwarf` CLI (`dwarf --help`)
 │   ├── dwarf.py                        # Typer app: convert, migrate, bump-gen, bump-schema, status,
 │   │                                    #   update-manifest, create-manifest, make-root,
@@ -140,9 +145,10 @@ uv sync --extra cpu --extra geometry  # add scikit-learn, for `dwarf build-geome
 uv sync --extra cpu --extra analysis  # matplotlib/polars/plotstyle, for `giant analyze render`
 uv sync --extra cpu --extra convert   # uproot/awkward/polars, for `dwarf convert`
 uv sync --extra cpu --extra wandb     # W&B logging (`giant train --wandb`)
+uv sync --extra cpu --extra workflow  # b2luigi, for `giant workflow run`
 ```
 
-The `dev` extra pulls in `convert`, `analysis`, `geometry` and `wandb` as well.
+The `dev` extra pulls in `convert`, `analysis`, `geometry`, `wandb` and `workflow` as well.
 
 `cpu` and `cuda` are mutually exclusive — pick one to select the torch build (pinned to 2.3.x). Plain `uv sync` installs no torch at all. See `CLAUDE.md` for details.
 
@@ -179,17 +185,29 @@ Config-file-only knobs (no CLI flag — use `--config config.toml`): `stage2_mod
 - `giant analyze` — deeper rollout-vs-reference diagnostics (marginals by energy/pdg/material, per-event totals, shower profiles, species share, leakage, secondaries):
 
 ```bash
-giant analyze submit rollout.yaml --accounting-group cms   # prep + one HTCondor job per plot × chunk (compute only)
-giant analyze submit a.yaml b.yaml --accounting-group cms --label flow --label wgan  # N rollouts vs one shared reference
+giant analyze prep rollout.yaml --chunks 8                 # lay out the run directory
+giant analyze prep a.yaml b.yaml --label flow --label wgan  # N rollouts vs one shared reference
 giant analyze render <run_dir> --gallery                   # local: merge chunks, then styled PDFs + HTML gallery (needs LaTeX)
 
 giant analyze list                                         # every catalog plot id
-giant analyze prep rollout.yaml --chunks 8                 # just the run directory, no submission
 giant analyze compute-one --id marginal_edep --run-dir <run_dir> --chunk 0   # what a condor job runs
 giant analyze merge-one --id marginal_edep --run-dir <run_dir>               # merge one plot's chunks (debugging)
 ```
 
-`<run_dir>` defaults to `<cwd>/analysis_runs/analysis_<id>` (`--run-dir` overrides it; `prep`/`submit` print it). Multiple rollout YAMLs must all name the same reference (`dataset`) file; each renders as its own colored series against one reference line/panel. Compute jobs are polars/numpy only; only `render` needs LaTeX, so it always runs locally.
+The per-(plot, chunk) compute jobs themselves are submitted by the workflow (below), not by `giant analyze` — these commands are the single-step primitives it calls. `<run_dir>` defaults to `<cwd>/analysis_runs/analysis_<id>` (`--run-dir` overrides it; `prep` prints it). Multiple rollout YAMLs must all name the same reference (`dataset`) file; each renders as its own colored series against one reference line/panel. Compute jobs are polars/numpy only; only `render` needs LaTeX, so it always runs locally.
+
+## Workflow orchestration
+
+Multi-step pipelines run through [b2luigi](https://github.com/belle2/b2luigi) — one spec file describes a whole experiment, and every step's outputs are files on `/ceph` that are only recomputed when their spec (or an upstream one) changes:
+
+```bash
+uv sync --extra cpu --extra workflow
+giant workflow run configs/workflow_example.toml --mode dry-run   # what would run
+giant workflow run configs/workflow_example.toml --mode show-output  # where every output goes
+giant workflow run configs/workflow_example.toml --batch --workers 20  # submit to HTCondor and wait
+```
+
+The spec holds `[workflow]`/`[condor]`/`[dataset]`/`[geometry]` plus repeated `[[train]]`, `[[rollout]]` and `[[analysis]]` tables cross-referenced by name (see `configs/workflow_example.toml`). The task graph is `DatasetTask → WarmCacheTask/GeometryOracleTask → TrainEpochTask… → TrainTask → RolloutTask → AnalysisPrepTask → AnalysisComputeTask(plot, chunk) → AnalysisRenderTask`. Training is split into one short GPU job per epoch (chained by `--resume`), which schedules better on a busy farm and survives preemption; `TrainTask` then publishes one `best.pt`/`last.pt`/`metrics.csv` for everything downstream. Rendering always runs locally, since it is the only step that needs LaTeX.
 
 Separately, `giant analyze metrics <train_run_dir>` renders training-progress plots (loss/lr/accuracy/grad-norm/router/wgan/throughput) straight from a training run's `metrics.csv`.
 

cliff.toml

@@ -37,7 +37,7 @@ commit_preprocessors = [
 protect_breaking_commits = false
 commit_parsers = [
     { message = "^Merge ", skip = true },
-    { message = "^chore: (bump version|update changelog|sync project version)", skip = true },
+    { message = "\\[skip ci\\]", skip = true },
     { message = "^Add", group = "<!-- 0 -->Added" },
     { message = "^(Fix|Clamp|Clip)", group = "<!-- 1 -->Fixed" },
     { message = "^(Remove|Drop|Deprecate)", group = "<!-- 2 -->Removed" },

configs/baseline.toml

@@ -29,21 +29,11 @@
 #     capacity overfitting is not the binding constraint, and every recent
 #     run used 0.0.
 #
-# Known weak spots, now measured against this exact config rather than
-# extrapolated from the pre-v0.3 field (analysis_341dfb14, best.pt @ epoch
-# 50/50, full writeup: knowledge-base/experiments/
-# giant-baseline-flow-ar-rollout-validation.md). Unlike every pre-v0.3
-# checkpoint (which under-produced steps/event by 1.6-5x), this baseline
-# OVER-produces steps/event by 1.32x (1.86e5 vs Geant4 1.41e5) and
-# under-produces secondaries/event by 0.84x (5.97e4 vs 7.14e4) — the sign on
-# steps flipped with the v0.3 autoregressive pivot, so don't assume it still
-# undershoots. Secondary-species hallucination (zero photons, hallucinated
-# `-14` muon antineutrinos) that broke every prior checkpoint is gone; the
-# remaining species gap is a total absence of hadronic/nuclear secondaries
-# (protons, neutrons, ion recoils), not miscalibration of the ones produced.
-# Total deposited energy/event is +1.9% high but its event-to-event spread is
-# ~16x too narrow (31 MeV vs Geant4's 491 MeV). Per-step deposited energy is
-# the worst per-step marginal (KS 0.179 vs 0.004-0.071 for the others).
+# Known weak spots this baseline is expected to *exhibit* (they are the
+# reason for the comparisons, not a reason to retune this file): every model
+# on record under-produces steps per event by ~2x (rollout ~7e4 vs Geant4
+# ~1.4e5) and secondaries per event by 2-3.5x (~2-3e4 vs 7.2e4), and n_sec
+# head accuracy sits at 0.863-0.867 regardless of size or objective.
 
 [meta]
 # REQUIRED. Without it config.migrate_config reads this file as v0.2 and

configs/workflow_example.toml

@@ -0,0 +1,68 @@
+# Example GIANT workflow spec — `giant workflow run configs/workflow_example.toml`.
+#
+# One file parameterises a whole experiment: the datasets, the geometry oracle,
+# N trainings, N rollouts, and the analyses comparing them. Every task's output
+# directory carries a hash of its resolved sub-spec (plus its parents), so
+# editing anything here re-runs exactly the affected subtree and nothing else.
+#
+# result_dir/log_dir must be visible from both the submit host and the workers
+# (i.e. on /ceph) — there is deliberately no HTCondor file transfer.
+
+[workflow]
+name       = "baseline-vs-router"
+result_dir = "/ceph/lbogner/workflows/baseline-vs-router"
+log_dir    = "/ceph/lbogner/workflows/baseline-vs-router/logs"
+
+[condor]
+accounting_group = "cms"
+repo_dir         = "/work/lbogner/giant"   # also b2luigi's working_dir
+env_script       = "/work/lbogner/giant/condor_env.sh"
+docker_image_cpu = "cverstege/alma9-gridjob"
+docker_image_gpu = "mschnepf/slc7-condocker"
+remote           = true
+
+[dataset]
+steps     = "/ceph/lbogner/geant_steps/train/"     # training data
+reference = "/ceph/lbogner/geant_steps/holdout/"   # rollout seeds + analysis truth
+
+[geometry]
+method    = "slab"
+subsample = 500_000
+
+[[train]]
+name           = "baseline"
+config         = "configs/baseline.toml"
+epochs         = 200
+epochs_per_job = 1            # one short GPU job per epoch, chained
+request_gpus   = 1
+gpu_memory_mb  = 20000
+overrides      = { lr = 3e-4 }   # `giant train` flag names
+
+[[train]]
+name           = "router-balanced"
+config         = "configs/router.toml"
+epochs         = 200
+epochs_per_job = 1
+request_gpus   = 1
+gpu_memory_mb  = 20000
+
+[[rollout]]
+name          = "baseline"
+train         = "baseline"     # -> [[train]].name
+n_events      = 2000
+energy_cutoff = 0.1
+
+[[rollout]]
+name          = "router-balanced"
+train         = "router-balanced"
+n_events      = 2000
+energy_cutoff = 0.1
+
+[[analysis]]
+name        = "baseline-vs-router"
+rollouts    = ["baseline", "router-balanced"]
+chunks      = 32
+energy_bins = 4
+bins        = 50
+top_pdg     = 6
+gallery     = true

giant/analysis/__init__.py

@@ -4,7 +4,7 @@ Compares one or more autoregressive ``giant rollout`` runs against a single
 held-out miniCaloSim reference file shared by all of them, producing
 publication-styled comparison plots (one colored series per rollout, one
 reference line) generated in parallel on HTCondor (one job per plot x data
-chunk, compute/merge/render split).
+chunk, compute/merge/render split) — orchestrated by ``giant/workflow``.
 
 Only ``render`` (and the ``render`` CLI path) imports plotstyle/LaTeX; everything
 re-exported here is plotstyle-free so it runs on a compute worker. Import
@@ -12,10 +12,9 @@ re-exported here is plotstyle-free so it runs on a compute worker. Import
 """
 
 from giant.analysis.catalog import build_catalog, catalog_ids, get_spec
-from giant.analysis.condor import (
+from giant.analysis.run import (
     LoadedRollout,
     RunMeta,
-    SubmitConfig,
     compute_one,
     compute_reduced,
     derive_run_dir,
@@ -24,7 +23,6 @@ from giant.analysis.condor import (
     merge_all,
     merge_one,
     prep,
-    write_submit,
 )
 from giant.analysis.context import Context, build_context
 from giant.analysis.reduced import Partial, Reduced
@@ -37,7 +35,6 @@ __all__ = [
     "get_spec",
     "LoadedRollout",
     "RunMeta",
-    "SubmitConfig",
     "compute_one",
     "compute_reduced",
     "derive_run_dir",
@@ -46,7 +43,6 @@ __all__ = [
     "merge_all",
     "merge_one",
     "prep",
-    "write_submit",
     "Context",
     "build_context",
     "Partial",

giant/analysis/catalog.py

@@ -22,9 +22,9 @@ which is the order rollouts were given on the CLI) plus the single reference.
 ``finalize`` merges each rollout's chunks independently and assembles a
 ``Reduced.payload`` keyed the same way: ``"series": {name: ...}`` for the
 rollouts, ``"reference": ...`` as one distinguished entry (omitted on
-rollout-only plots like ``leakage_fraction``). The heatmap-shaped specs
-(``marginal_distance_summary``, ``sec_count_per_step_by_species``) and the
-router diagnostics are inherently one-matrix/one-checkpoint per rollout, so their
+rollout-only plots like ``leakage_fraction``). The two heatmap-shaped specs
+(``marginal_distance_summary``, ``n_sec_confusion``) and the router
+diagnostics are inherently one-matrix/one-checkpoint per rollout, so their
 ``"series"`` entries are whole per-rollout artifacts (a matrix, a gating
 dict) rather than a single number/array — ``render.py`` draws those as one
 panel per rollout instead of one line/bar per rollout.
@@ -60,6 +60,7 @@ from giant.analysis.reduce import (
     leakage_fraction,
     profile_finalize,
     profile_partial,
+    sec_count_by_event,
     species_share,
     sum_merge,
     transverse_expr,
@@ -71,15 +72,7 @@ from giant.analysis.router_gating import (
     compute_router_share_by_process,
     compute_router_specialization,
 )
-from giant.analysis.sources import (
-    RolloutSide,
-    RolloutSpec,
-    Side,
-    open_side,
-    physical_steps,
-    secondaries,
-    secondaries_by_step,
-)
+from giant.analysis.sources import RolloutSide, RolloutSpec, Side, open_side, physical_steps, secondaries
 from giant.analysis.type_embedding_distance import compute_type_embedding_l1_distance
 from giant.analysis.variables import RANGED_VARS, cos_scatter_expr
 
@@ -218,6 +211,31 @@ def _ks_statistic(r_counts, t_counts) -> float:
     return float(np.max(np.abs(r_cdf - t_cdf)))
 
 
+def _integer_confusion(
+    t: np.ndarray, r: np.ndarray, max_bins: int = 21, cap: int | None = None
+) -> tuple[list[str], np.ndarray]:
+    """Confusion matrix of two paired small-integer arrays (e.g. secondary counts).
+
+    Bins are consecutive integers ``0..cap``, with the last bin an overflow
+    ``"cap+"`` bucket, so an occasional pathological count doesn't blow up the
+    heatmap. Returns ``(labels, matrix)`` with ``matrix[i, j]`` counting pairs
+    with ``t == i`` and ``r == j`` (both clipped into ``[0, cap]``).
+
+    ``cap``, if given, is used as-is instead of being derived from ``t``/``r``
+    — lets a multi-rollout caller fix one shared cap (and so one shared label
+    set) across every rollout's matrix rather than each panel picking its own.
+    """
+    if cap is None:
+        cap = min(max(int(t.max()) if len(t) else 0, int(r.max()) if len(r) else 0, 1), max_bins - 1)
+    t_c = np.clip(t.astype(np.int64), 0, cap)
+    r_c = np.clip(r.astype(np.int64), 0, cap)
+    n = cap + 1
+    mat = np.zeros((n, n), dtype=np.int64)
+    np.add.at(mat, (t_c, r_c), 1)
+    labels = [str(i) for i in range(cap)] + [f"{cap}+"]
+    return labels, mat
+
+
 def _containment_depths(mat: np.ndarray, edges: np.ndarray, quantile: float) -> np.ndarray:
     """Per-event depth containing ``quantile`` of that event's deposited energy.
 
@@ -784,139 +802,6 @@ def _sec_count_per_species_finalize(parts: list[dict], ctx: Context) -> Reduced:
     )
 
 
-# Per-step secondary multiplicity. Fixed integer edges (bin i == exactly i
-# secondaries, the top bin an overflow bucket) keep both plots sum-mergeable
-# across chunks — no shared-range pass needed. The species heatmap gets a
-# shorter row axis because a single step rarely emits many of *one* species.
-_N_SEC_STEP_CAP = 20
-_N_SEC_SPECIES_CAP = 10
-_OTHER_KEY = "other"
-
-
-def _n_sec_edges(cap: int) -> np.ndarray:
-    return np.arange(-0.5, cap + 1.5)
-
-
-def _sec_step_key_lf(lf: pl.LazyFrame, side: Side) -> pl.LazyFrame:
-    """Secondaries with their emitting-step key.
-
-    The rollout side reads *all* rows, not just physical ones: a secondary
-    whose very first row is a synthetic termination row (born, then immediately
-    escaped or cut) was still produced by its parent step, and dropping it would
-    undercount that step's multiplicity.
-    """
-    return secondaries_by_step(lf, side)
-
-
-def _n_steps(lf: pl.LazyFrame) -> int:
-    """Number of (physical) step rows — the denominator the zero rows come from."""
-    return int(lf.select(pl.len()).collect(engine="streaming").item())
-
-
-def _sec_count_per_step_partial(b: Bundle) -> dict:
-    edges = _n_sec_edges(_N_SEC_STEP_CAP)
-
-    def _side(sec_lf: pl.LazyFrame, steps_lf: pl.LazyFrame) -> dict:
-        per_step = sec_lf.group_by("step_key").agg(pl.len().alias("n"))
-        return {
-            "h": _partial_hist(per_step, pl.col("n").clip(0, _N_SEC_STEP_CAP), edges),
-            "n_steps": _n_steps(steps_lf),
-        }
-
-    return {
-        "r": _per_rollout(b, lambda rs: _side(_sec_step_key_lf(rs.all, Side.rollout), rs.phys)),
-        "t": _side(_sec_step_key_lf(b.t_all, Side.reference), b.t_phys),
-    }
-
-
-def _zero_filled(part_hists: list[dict], n_steps: int, key, nbins: int) -> list[int]:
-    """Merged counts for one series, with bin 0 (= steps that emitted none) filled in.
-
-    The reduction only ever sees steps that produced at least one secondary, so
-    the empty ones are recovered by subtraction from the total step count.
-    """
-    counts = _finalize_counts(sum_merge(part_hists), key, nbins)
-    counts[0] = max(n_steps - int(sum(counts)), 0)
-    return [int(c) for c in counts]
-
-
-def _sec_count_per_step_finalize(parts: list[dict], ctx: Context) -> Reduced:
-    edges = _n_sec_edges(_N_SEC_STEP_CAP)
-    nb = len(edges) - 1
-    names = list(parts[0]["r"])
-    series = {
-        name: _zero_filled([p["r"][name]["h"] for p in parts], sum(p["r"][name]["n_steps"] for p in parts), 0, nb)
-        for name in names
-    }
-    return Reduced(
-        id="sec_count_per_step",
-        family="secondaries",
-        kind="overlay_hist",
-        title="Number of secondaries per step",
-        xlabel="secondaries per step",
-        payload={
-            "edges": edges.tolist(),
-            "series": series,
-            "reference": _zero_filled([p["t"]["h"] for p in parts], sum(p["t"]["n_steps"] for p in parts), 0, nb),
-            "log_y": True,
-        },
-    )
-
-
-def _species_key_expr(top_pdgs: list[int]) -> pl.Expr:
-    """``pdg`` bucketed into the shared top-K columns plus one ``other`` bin."""
-    return pl.when(pl.col("pdg").is_in(list(top_pdgs))).then(pl.col("pdg").cast(pl.Utf8)).otherwise(pl.lit(_OTHER_KEY))
-
-
-def _sec_count_per_step_by_species_partial(b: Bundle) -> dict:
-    edges = _n_sec_edges(_N_SEC_SPECIES_CAP)
-    group = _species_key_expr(b.ctx.top_pdgs)
-
-    def _side(sec_lf: pl.LazyFrame, steps_lf: pl.LazyFrame) -> dict:
-        per_step_species = sec_lf.group_by("step_key", "pdg").agg(pl.len().alias("n"))
-        return {
-            "h": _partial_hist(per_step_species, pl.col("n").clip(0, _N_SEC_SPECIES_CAP), edges, group=group),
-            "n_steps": _n_steps(steps_lf),
-        }
-
-    return {
-        "r": _per_rollout(b, lambda rs: _side(_sec_step_key_lf(rs.all, Side.rollout), rs.phys)),
-        "t": _side(_sec_step_key_lf(b.t_all, Side.reference), b.t_phys),
-    }
-
-
-def _sec_count_per_step_by_species_finalize(parts: list[dict], ctx: Context) -> Reduced:
-    edges = _n_sec_edges(_N_SEC_SPECIES_CAP)
-    nb = len(edges) - 1
-    names = list(parts[0]["r"])
-    keys = [str(p) for p in ctx.top_pdgs] + [_OTHER_KEY]
-
-    def _matrix(hists: list[dict], n_steps: int) -> list[list[int]]:
-        # columns = species, rows = multiplicity; every species gets its own
-        # zero row (steps that produced none of *that* species).
-        cols = [_zero_filled(hists, n_steps, k, nb) for k in keys]
-        return [[cols[j][i] for j in range(len(keys))] for i in range(nb)]
-
-    return Reduced(
-        id="sec_count_per_step_by_species",
-        family="secondaries",
-        kind="heatmap",
-        title="Per-step secondary multiplicity by species",
-        xlabel="species",
-        payload={
-            "series": {
-                n: _matrix([p["r"][n]["h"] for p in parts], sum(p["r"][n]["n_steps"] for p in parts)) for n in names
-            },
-            "reference": _matrix([p["t"]["h"] for p in parts], sum(p["t"]["n_steps"] for p in parts)),
-            "row_labels": [str(i) for i in range(_N_SEC_SPECIES_CAP)] + [f"{_N_SEC_SPECIES_CAP}+"],
-            "col_labels": [pdg_label(k) for k in ctx.top_pdgs] + [_OTHER_KEY],
-            "ylabel": "secondaries of this species per step",
-            "cbar_label": "step count",
-            "log_color": True,
-        },
-    )
-
-
 def _sec_energy_partial(b: Bundle) -> dict:
     edges = np.linspace(*b.ctx.sec_energy_range, b.ctx.n_sec_bins + 1)
     return {
@@ -973,6 +858,62 @@ def _sec_cos_angle_finalize(parts: list[dict], ctx: Context) -> Reduced:
     )
 
 
+def _n_sec_confusion_partial(b: Bundle) -> dict:
+    t_ids, t_n = sec_count_by_event(b.t_all, _t_sec(b))
+
+    def _r(rs: RolloutSide) -> dict:
+        ids, n = sec_count_by_event(rs.phys, _r_sec(rs))
+        return {"ids": ids.tolist(), "n": n.tolist()}
+
+    return {"r": _per_rollout(b, _r), "t": {"ids": t_ids.tolist(), "n": t_n.tolist()}}
+
+
+def _n_sec_confusion_finalize(parts: list[dict], ctx: Context) -> Reduced:
+    names = list(parts[0]["r"])
+    # event-disjoint chunking (see Bundle.open) means each event_id appears in
+    # exactly one part on each side, so a plain dict build is a safe merge.
+    t_ids = np.concatenate([np.asarray(p["t"]["ids"], dtype=np.int64) for p in parts])
+    t_n = np.concatenate([np.asarray(p["t"]["n"], dtype=np.int64) for p in parts])
+    t_map = dict(zip(t_ids.tolist(), t_n.tolist()))
+
+    pairs: dict[str, tuple[np.ndarray, np.ndarray]] = {}
+    max_val = 0
+    for name in names:
+        r_ids = np.concatenate([np.asarray(p["r"][name]["ids"], dtype=np.int64) for p in parts])
+        r_n = np.concatenate([np.asarray(p["r"][name]["n"], dtype=np.int64) for p in parts])
+        r_map = dict(zip(r_ids.tolist(), r_n.tolist()))
+        common = sorted(set(r_map) & set(t_map))
+        true_n = np.array([t_map[e] for e in common], dtype=np.int64)
+        pred_n = np.array([r_map[e] for e in common], dtype=np.int64)
+        pairs[name] = (true_n, pred_n)
+        if len(true_n):
+            max_val = max(max_val, int(true_n.max()), int(pred_n.max()))
+
+    cap = min(max(max_val, 1), 20)
+    matrices: dict[str, list[list[int]]] = {}
+    labels: list[str] = []
+    for name in names:
+        true_n, pred_n = pairs[name]
+        labels, mat = _integer_confusion(true_n, pred_n, cap=cap)
+        matrices[name] = mat.tolist()
+
+    return Reduced(
+        id="n_sec_confusion",
+        family="secondaries",
+        kind="heatmap",
+        title="Predicted vs true secondary count per event",
+        xlabel="predicted secondaries (rollout)",
+        payload={
+            "series": matrices,
+            "row_labels": labels,
+            "col_labels": labels,
+            "ylabel": "true secondaries (reference)",
+            "cbar_label": "event count",
+            "vmin": 0.0,
+        },
+    )
+
+
 # ---------------------------------------------------------------------------
 # router diagnostics (not chunked — already bounded/subsampled)
 # ---------------------------------------------------------------------------
@@ -1136,18 +1077,6 @@ def build_catalog() -> list[PlotSpec]:
             compute_partial=_sec_count_per_species_partial,
             finalize=_sec_count_per_species_finalize,
         ),
-        PlotSpec(
-            "sec_count_per_step",
-            "secondaries",
-            compute_partial=_sec_count_per_step_partial,
-            finalize=_sec_count_per_step_finalize,
-        ),
-        PlotSpec(
-            "sec_count_per_step_by_species",
-            "secondaries",
-            compute_partial=_sec_count_per_step_by_species_partial,
-            finalize=_sec_count_per_step_by_species_finalize,
-        ),
         PlotSpec(
             "sec_energy",
             "secondaries",
@@ -1160,6 +1089,12 @@ def build_catalog() -> list[PlotSpec]:
             compute_partial=_sec_cos_angle_partial,
             finalize=_sec_cos_angle_finalize,
         ),
+        PlotSpec(
+            "n_sec_confusion",
+            "secondaries",
+            compute_partial=_n_sec_confusion_partial,
+            finalize=_n_sec_confusion_finalize,
+        ),
         PlotSpec(
             "router_gating",
             "model",

giant/analysis/reduce.py

@@ -271,3 +271,20 @@ def leakage_fraction(lf: pl.LazyFrame) -> np.ndarray:
     escaped = per_event["escaped"].fill_null(0.0).to_numpy()
     total = deposited + escaped
     return np.where(total > 0, escaped / total, 0.0)
+
+
+def sec_count_by_event(lf_all: pl.LazyFrame, sec_lf: pl.LazyFrame) -> tuple[np.ndarray, np.ndarray]:
+    """Per-event secondary count, zero-filled for events that produced none.
+
+    Two bounded per-event ``group_by``s — the full event set (from ``lf_all``)
+    and the secondary counts (from ``sec_lf``, see ``sources.secondaries``) —
+    merged in Python via a dict. Both results are event-granularity (not
+    per-row), so this stays in the same bounded-memory budget as
+    ``event_scalars``; a plain ``group_by`` on ``sec_lf`` alone would silently
+    drop zero-secondary events instead of zero-filling them.
+    """
+    ev = lf_all.select("event_id").unique().collect(engine="streaming")["event_id"].to_numpy()
+    cnt_df = sec_lf.group_by("event_id").agg(pl.len().alias("n")).collect(engine="streaming")
+    cnt = dict(zip(cnt_df["event_id"].to_list(), cnt_df["n"].to_list()))
+    counts = np.array([cnt.get(int(e), 0) for e in ev], dtype=np.int64)
+    return ev, counts

giant/analysis/reduced.py

@@ -27,7 +27,7 @@ from pathlib import Path
 #   "router_specialization"  max gate weight vs energy (one scalar trend line
 #                    summarizing "router_gating"), per rollout with an enabled router
 #   "heatmap"        row x col matrix + colorbar, one panel per rollout (a
-#                    distance scorecard)
+#                    distance scorecard or a predicted-vs-true confusion matrix)
 #   "unavailable"    plot not applicable to this run (e.g. no MoE checkpoint)
 
 

giant/analysis/render.py

@@ -27,7 +27,6 @@ from pathlib import Path
 
 import numpy as np
 import plotstyle as ps
-from matplotlib.colors import LogNorm
 import yaml
 
 from giant.analysis.reduced import Reduced
@@ -377,16 +376,10 @@ def _render_router_specialization(r: Reduced, params: dict):
 
 
 def _render_heatmap(r: Reduced, params: dict):
-    series = dict(r.payload["series"])
+    series = r.payload["series"]
     row_labels = r.payload["row_labels"]
     col_labels = r.payload["col_labels"]
-    # A heatmap-shaped plot is one matrix per rollout, so the reference (when the
-    # comparison has one — the distance scorecard doesn't) becomes one more panel
-    # rather than another line.
-    if r.payload.get("reference") is not None:
-        series["reference"] = r.payload["reference"]
     names = list(series)
-    norm = LogNorm(vmin=1) if r.payload.get("log_color") else None
     fig, axes = ps.new_figure(
         "slide-16x9" if len(names) > 1 else "thesis-single",
         title=r.title,
@@ -404,9 +397,8 @@ def _render_heatmap(r: Reduced, params: dict):
             origin="upper",
             aspect="auto",
             cmap=r.payload.get("cmap", "viridis"),
-            norm=norm,
-            vmin=None if norm else r.payload.get("vmin"),
-            vmax=None if norm else r.payload.get("vmax"),
+            vmin=r.payload.get("vmin"),
+            vmax=r.payload.get("vmax"),
         )
         ax.set_xticks(range(len(col_labels)))
         ax.set_xticklabels(col_labels, rotation=45, ha="right")
@@ -543,7 +535,7 @@ def render_run(run_dir: str | Path, *, run_gallery: bool = False) -> list[Path]:
     (checkpoint, paths, cutoffs) from ``run_meta.json`` into every plot's
     gallery metadata and renders.
     """
-    from giant.analysis.condor import RunMeta, merge_all
+    from giant.analysis.run import RunMeta, merge_all
 
     run_dir = Path(run_dir)
     merge_all(run_dir)

giant/analysis/run.py

@@ -1,4 +1,6 @@
-"""HTCondor orchestration driven by one or more ``giant rollout`` YAML sidecars.
+"""Analysis run directories: prep, per-(plot, chunk) compute, and merge.
+
+Driven by one or more ``giant rollout`` YAML sidecars.
 
 A rollout writes a YAML sidecar (``giant/cli.py:_write_prediction_ref`` +
 rollout extras) that already names both files we need and carries the run's
@@ -22,13 +24,16 @@ everything out under it:
     <run_dir>/reduced/<id>.json   merged, per plot
     <run_dir>/plots/<family>/<id>.pdf   rendered locally
 
-Job model (one condor job per (plot, chunk), compute/merge/render split):
+Job model (one job per (plot, chunk), compute/merge/render split). Job
+submission itself is b2luigi's (``giant/workflow/tasks.py`` — ``AnalysisPrepTask``
+/ ``AnalysisComputeTask`` / ``AnalysisRenderTask``); this module only provides
+the three steps they call:
 
-1. ``prep`` runs once on the submit node — reads the YAML, resolves the shared
+1. ``prep`` runs once locally — reads the YAML, resolves the shared
    context from a subsample, writes ``shared.json`` + ``run_meta.json``
    (including the run's configured ``n_chunks``).
 2. one job per catalog id x chunk index runs ``giant analyze compute-one
-   --run-dir`` on a worker — a single streaming pass over that
+   --run-dir`` (or ``compute_one`` in-process) on a worker — a single streaming pass over that
    ``event_id``-disjoint chunk, writing ``reduced_partial/<id>__<chunk>.json``
    (polars/numpy only, no LaTeX). Specs marked ``chunkable=False``
    (``PlotSpec``, ``catalog.py``) always run as a single chunk.
@@ -38,15 +43,15 @@ Job model (one condor job per (plot, chunk), compute/merge/render split):
    ``reduced/<id>.json``, then renders those into the styled PDF + gallery tree
    (that step imports plotstyle/LaTeX).
 
-Files on ``/ceph`` or ``/work`` are reached via ``ProvidesETPResources``; no
-HTCondor file transfer of the multi-GB inputs.
+Files on ``/ceph`` or ``/work`` are reached directly (see
+``giant/workflow/htcondor.py``); no HTCondor file transfer of the multi-GB
+inputs.
 """
 
 from __future__ import annotations
 
 import json
 import shutil
-import sys
 from collections.abc import Sequence
 from dataclasses import dataclass, field
 from pathlib import Path
@@ -57,7 +62,6 @@ import yaml
 from giant.analysis.catalog import Bundle, catalog_ids, get_spec
 from giant.analysis.context import Context, build_context
 from giant.analysis.reduced import Partial
-from giant.analysis.runtime_estimate import estimate_runtime_s
 from giant.analysis.sources import RolloutSpec, Side, open_side
 
 # Keys copied verbatim from a rollout YAML into each plot's gallery metadata.
@@ -429,136 +433,3 @@ def merge_one(spec_id: str, run_dir: str | Path) -> Path:
 def merge_all(run_dir: str | Path) -> list[Path]:
     """Merge every catalog plot's chunk partials into ``reduced/<id>.json``."""
     return [merge_one(spec_id, run_dir) for spec_id in catalog_ids()]
-
-
-# ---------------------------------------------------------------------------
-# submit description
-# ---------------------------------------------------------------------------
-
-
-@dataclass
-class SubmitConfig:
-    run_dir: Path
-    accounting_group: str
-    repo_dir: Path
-    docker_image: str = "cverstege/alma9-gridjob"
-    request_memory_mb: int = 8192
-    request_cpus: int = 1
-    remote: bool = False  # +RemoteJob (grid I/O) vs ProvidesETPResources (local files)
-    n_chunks: int = 1  # per-plot data chunks; ignored for chunkable=False specs
-
-
-_WRAPPER = """#!/bin/bash
-set -euo pipefail
-cd {repo_dir}
-exec {giant_exe} analyze compute-one --id "$1" --chunk "$2" --run-dir {run_dir}
-"""
-
-
-def _submit_description(cfg: SubmitConfig, wrapper: Path, jobs_file: Path) -> str:
-    reqs_attrs = "+RemoteJob = True\n" if cfg.remote else "requirements = TARGET.ProvidesETPResources\n"
-    return (
-        "universe = docker\n"
-        f"docker_image = {cfg.docker_image}\n"
-        f"executable = {wrapper}\n"
-        "arguments = $(plotid) $(chunk)\n"
-        "should_transfer_files = YES\n"
-        "when_to_transfer_output = ON_EXIT\n"
-        f"request_memory = {cfg.request_memory_mb}\n"
-        f"request_cpus = {cfg.request_cpus}\n"
-        "+RequestWalltime = $(walltime)\n"
-        f"accounting_group = {cfg.accounting_group}\n"
-        f"{reqs_attrs}"
-        f"output = {cfg.run_dir}/logs/$(plotid)__$(chunk).out\n"
-        f"error = {cfg.run_dir}/logs/$(plotid)__$(chunk).err\n"
-        f"log = {cfg.run_dir}/logs/condor.log\n"
-        f"queue plotid,chunk,walltime from {jobs_file}\n"
-    )
-
-
-def _job_walltimes(run_dir: Path, ids: list[str], n_chunks: int) -> list[tuple[str, int, int]]:
-    """``(spec_id, chunk, walltime_s)`` for every job, sized from ``run_meta.json``.
-
-    Row counts come from ``prep``'s ``RunMeta.rows_per_chunk``/``total_rows``;
-    ``chunkable=False`` specs (router diagnostics) always use the dataset
-    total since they run as a single job regardless of ``n_chunks``.
-    """
-    meta = RunMeta.load(run_dir / "run_meta.json")
-    jobs: list[tuple[str, int, int]] = []
-    for spec_id in ids:
-        chunkable = get_spec(spec_id).chunkable
-        chunks = range(n_chunks) if chunkable else [0]
-        for chunk in chunks:
-            n_rows = meta.rows_per_chunk[chunk] if chunkable else meta.total_rows
-            jobs.append((spec_id, chunk, estimate_runtime_s(spec_id, n_rows)))
-    return jobs
-
-
-def _resolve_giant_executable(repo_dir: Path) -> Path:
-    """Path to the ``giant`` entry point to bake into the condor wrapper script.
-
-    Prefers the venv currently running this process (``sys.executable``'s
-    sibling ``giant``) so a submit from a non-default venv (e.g. ``--extra
-    cuda`` on a dev box) doesn't silently pick up a different one; falls back
-    to ``repo_dir/.venv/bin/giant`` for the case this is invoked from outside
-    any venv (e.g. a system Python).
-    """
-    active = Path(sys.executable).parent / "giant"
-    if active.exists():
-        return active
-    venv_giant = repo_dir / ".venv" / "bin" / "giant"
-    if not venv_giant.exists():
-        raise FileNotFoundError(
-            f"no `giant` executable found next to {sys.executable} or at "
-            f"{venv_giant} — condor jobs run it directly (no `uv` on the "
-            f"worker image), so run `uv sync --extra cpu` in {repo_dir} "
-            "before submitting."
-        )
-    return venv_giant
-
-
-def write_submit(cfg: SubmitConfig, ids: list[str] | None = None) -> Path:
-    """Write the wrapper script, (plot, chunk) job list, and HTCondor submit
-    description.
-
-    Each catalog id gets ``cfg.n_chunks`` jobs, except ``chunkable=False``
-    specs (the router diagnostics), which always get exactly one regardless of
-    ``cfg.n_chunks``. Every job's ``+RequestWalltime`` is estimated from its
-    chunk's row count (``runtime_estimate.estimate_runtime_s``, requires
-    ``run_meta.json`` from ``prep`` to already carry ``rows_per_chunk``).
-    Returns the submit description path (``<run_dir>/analyze.sub``). Does not
-    submit — call ``condor_submit`` on the returned file.
-
-    ``cfg.n_chunks`` and the run directory's own ``RunMeta.n_chunks`` (fixed by
-    ``prep``, and what ``RunMeta.rows_per_chunk`` was sized against) are two
-    independent values — checked equal up front so a mismatch is a clear error
-    here rather than an ``IndexError`` out of ``_job_walltimes``.
-    """
-    giant_exe = _resolve_giant_executable(cfg.repo_dir)
-
-    ids = ids or catalog_ids()
-    run_dir = cfg.run_dir
-    meta = RunMeta.load(run_dir / "run_meta.json")
-    if cfg.n_chunks != meta.n_chunks:
-        raise ValueError(
-            f"SubmitConfig.n_chunks={cfg.n_chunks} does not match the "
-            f"n_chunks this run directory was prepped with "
-            f"(RunMeta.n_chunks={meta.n_chunks} in {run_dir}/run_meta.json) — "
-            "re-run `prep` with the desired n_chunks, or fix cfg.n_chunks to "
-            "match it."
-        )
-    (run_dir / "logs").mkdir(parents=True, exist_ok=True)
-    (run_dir / "reduced").mkdir(parents=True, exist_ok=True)
-    (run_dir / "reduced_partial").mkdir(parents=True, exist_ok=True)
-
-    wrapper = run_dir / "run_compute.sh"
-    wrapper.write_text(_WRAPPER.format(repo_dir=cfg.repo_dir, giant_exe=giant_exe, run_dir=run_dir))
-    wrapper.chmod(0o755)
-
-    jobs = _job_walltimes(run_dir, ids, cfg.n_chunks)
-    jobs_file = run_dir / "jobs.txt"
-    jobs_file.write_text("\n".join(f"{i},{k},{w}" for i, k, w in jobs) + "\n")
-
-    sub = run_dir / "analyze.sub"
-    sub.write_text(_submit_description(cfg, wrapper, jobs_file))
-    return sub

giant/analysis/runtime_estimate.py

@@ -1,4 +1,4 @@
-"""Per-(plot, chunk) HTCondor walltime estimates for `giant analyze submit`.
+"""Per-(plot, chunk) HTCondor walltime estimates for the analysis compute jobs.
 
 Each catalog spec's compute cost is close to linear in the number of input
 rows a `compute-one` job streams over — every spec is one (or a couple of)

giant/analysis/sources.py

@@ -236,34 +236,3 @@ def secondaries(lf: pl.LazyFrame, side: Side) -> pl.LazyFrame:
             pl.col("sec_dz_list").alias("sdz"),
         )
     )
-
-
-def secondaries_by_step(lf: pl.LazyFrame, side: Side) -> pl.LazyFrame:
-    """One row per produced secondary, tagged with the step that produced it.
-
-    Canonical columns: ``step_key`` (an opaque struct identifying the emitting
-    step) and ``pdg``. ``secondaries`` deliberately drops that link; the
-    per-step multiplicity plots need it, so this is a separate view rather than
-    extra columns every other consumer would pay for.
-
-    - rollout: a secondary's birth row carries ``parent_id`` and a birth
-      position copied verbatim from the parent step's ``post_pos``, so
-      ``(event_id, parent_id, pre_pos)`` identifies the emitting step exactly —
-      no join against the (large) step frame is needed.
-    - reference: secondaries already live on their parent step's row, so the
-      row index *is* the step key. It is only ever used as a group key inside
-      one chunk's own aggregation, so indices repeating across chunks is
-      harmless.
-    """
-    if side is Side.rollout:
-        return lf.filter((pl.col("generation") > 0) & (pl.col("step_no") == 0)).select(
-            pl.struct("event_id", "parent_id", "pre_x", "pre_y", "pre_z").alias("step_key"),
-            "pdg",
-        )
-    return (
-        lf.select("sec_pdg_list")
-        .with_row_index("_row")
-        .explode("sec_pdg_list")
-        .drop_nulls("sec_pdg_list")
-        .select(pl.struct("_row").alias("step_key"), pl.col("sec_pdg_list").cast(pl.Int64).alias("pdg"))
-    )

giant/checkpoint_io.py

@@ -14,18 +14,11 @@ directly and imported from non-CLI code (`giant.analysis.router_gating`,
 lazily — see that module's docstring for why). Failures raise
 `CheckpointCompatibilityError` with the same wording the CLI has always
 shown; the CLI layer catches it and does the `typer.echo`/`Exit(1)`.
-
-`load_for_inference`'s `config_overrides` (gitea #87) lets a caller change a
-checkpoint's `model_config` at load time, restricted to
-`giant.config.INFERENCE_OVERRIDES` — the allowlist of keys that only affect
-sampling, never module construction/shapes or the preprocessing normalizers/
-vocab maps were fit under.
 """
 
 from __future__ import annotations
 
-import copy
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from pathlib import Path
 
 import torch
@@ -36,45 +29,13 @@ from giant.constants import K_MAX
 from giant.data.loader import TopNMap
 from giant.data.setup_cache import topnmap_from_json
 from giant.data.transforms import Normalizer
-from giant.model.network import _migrate_legacy_model_config, build_models
+from giant.model.network import build_models
 
 
 class CheckpointCompatibilityError(Exception):
     """Checkpoint is missing something `load_for_inference` needs."""
 
 
-def apply_config_overrides(model_cfg: dict, overrides: dict[str, object] | None) -> dict:
-    """Deep-merge dotted-path *overrides* into a checkpoint's `model_config`,
-    validated against `giant.config.INFERENCE_OVERRIDES` — the allowlist of
-    keys that only affect sampling, not module construction/shapes or the
-    preprocessing normalizers/vocab maps were fit under (gitea #87).
-
-    Migrates a v0.2 flat `model_config` to the nested v0.3 shape first: a
-    dotted path like "stage1_model.ddpm.n_steps" would otherwise silently
-    write into a dict that `build_models` still reads as flat (it decides
-    v0.2-vs-v0.3 by `"stage1_model" in model_config`), suppressing migration.
-
-    Raises `CheckpointCompatibilityError` — never a bare `ValueError` or a
-    downstream `load_state_dict` size mismatch — for an unknown/disallowed
-    path or a value that fails its allowlisted check.
-    """
-    if not overrides:
-        return model_cfg
-    cfg = model_cfg if "stage1_model" in model_cfg else _migrate_legacy_model_config(model_cfg)
-    cfg = copy.deepcopy(cfg)
-    for path, value in overrides.items():
-        spec = gconfig.INFERENCE_OVERRIDES.get(path)
-        if spec is None:
-            allowed = ", ".join(sorted(gconfig.INFERENCE_OVERRIDES))
-            raise CheckpointCompatibilityError(f"{path!r} is not an inference-safe override — allowed paths: {allowed}")
-        try:
-            spec.check(path, value)
-        except ValueError as exc:
-            raise CheckpointCompatibilityError(str(exc)) from exc
-        gconfig._set_path(cfg, path, value)
-    return cfg
-
-
 def conditioning_axes(model_cfg: dict, default: str = "embedding") -> tuple[str, str]:
     """(particle_conditioning, material_conditioning) for
     `giant.data.transforms.build_cond_features`/`build_features` — from
@@ -167,7 +128,6 @@ class InferenceContext:
     model_config: dict
     epoch: int | None
     best_val_loss: float | None
-    config_overrides: dict[str, object] = field(default_factory=dict)
 
 
 def load_for_inference(
@@ -176,7 +136,6 @@ def load_for_inference(
     command_name: str,
     weights: str = "raw",
     require_stage2: bool = True,
-    config_overrides: dict[str, object] | None = None,
 ) -> InferenceContext:
     """Load *checkpoint* and reconstruct everything `predict`/`rollout` need
     to run it forward, on *device*, in `eval()` mode.
@@ -189,13 +148,6 @@ def load_for_inference(
     both stages) or an acceptable `stage2 = None` result — kept as a real
     parameter since `stage{1,2}_model.active` is a real, if currently
     stage1+stage2-only-in-practice, config option.
-
-    *config_overrides* deep-merges dotted `model_config` paths (e.g.
-    `{"stage2_model.n_sec.sampling": "sample"}`) before anything is
-    derived from `model_config` or built — see `apply_config_overrides` for
-    the allowlist and validation. Every derived `InferenceContext` field
-    (`other_policy`, `stage{1,2}_ddpm_steps`, the built modules, ...)
-    reflects the overridden config.
     """
     ckpt = torch.load(checkpoint, map_location="cpu", weights_only=False)
     for key in ("model_config", "sec_decoder"):
@@ -207,7 +159,7 @@ def load_for_inference(
 
     gconfig.warn_if_checkpoint_config_mismatch(checkpoint)
 
-    model_cfg = apply_config_overrides(ckpt["model_config"], config_overrides)
+    model_cfg = ckpt["model_config"]
     particle_conditioning, material_conditioning = conditioning_axes(model_cfg)
     pdg_topn_map = load_pdg_topn_map(ckpt)
     mat_topn_map = load_mat_topn_map(ckpt)
@@ -279,5 +231,4 @@ def load_for_inference(
         model_config=model_cfg,
         epoch=ckpt.get("epoch"),
         best_val_loss=ckpt.get("best_val_loss"),
-        config_overrides=dict(config_overrides) if config_overrides else {},
     )

giant/cli.py

@@ -1,19 +1,22 @@
-from __future__ import annotations
-
 from collections import Counter
 from datetime import datetime, timezone
 from enum import Enum
 import math
 from pathlib import Path
 import re
-from typing import TYPE_CHECKING, Optional
+import sys
+from typing import Optional
 import uuid as uuid_mod
 
+import numpy as np
+import yaml
+import torch
 import typer
 from typing_extensions import Annotated
 
-if TYPE_CHECKING:
-    import numpy as np
+import pyarrow as pa
+import pyarrow.parquet as pq
+from tqdm import tqdm
 
 from giant import config as gconfig
 from giant.constants import (
@@ -23,11 +26,30 @@ from giant.constants import (
     PREDICT_SCHEMA_VERSION_KEY,
     ROLLOUT_COORD_VALUE,
 )
-
-# giant.materials only pulls in numpy (no torch/pandas), and MATERIAL_PROPERTIES
-# is needed at decoration time below (a Typer option default), so it can't be
-# deferred into a command body like the rest of this module's heavy imports.
+from giant.data.loader import (
+    event_id_offset,
+    find_parquet_files,
+    iter_file_chunks,
+    iter_cond_chunks,
+)
+from giant.data.transforms import (
+    build_features,
+    build_cond_features,
+    energy_simplex_decode,
+    inv_local_frame_rotation,
+    inv_log_transform,
+    reconstruct_post_pos,
+)
+from giant.checkpoint_io import CheckpointCompatibilityError, load_for_inference
+from giant.geometry import GeometryOracle
 from giant.materials import MATERIAL_PROPERTIES
+from giant.pipeline import run_train_job
+from giant.rollout import (
+    L1DistCollector,
+    decode_secondary_identity,
+    rollout as run_rollout,
+)
+from giant.sample import resolve_n_sec, sample_stage1, sample_stage2
 
 app = typer.Typer(no_args_is_help=True)
 
@@ -120,23 +142,6 @@ def _parse_router_axis_flags(specs: list[str]) -> dict[str, object]:
     return out
 
 
-def _parse_set_flags(specs: Optional[list[str]]) -> dict[str, object]:
-    """Parse repeated `--set dotted.path=value` flags into a dict, typing
-    each value with `_coerce_scalar` the same way a TOML file's native types
-    would arrive. Validation against the inference-safe allowlist happens
-    downstream in `giant.checkpoint_io.apply_config_overrides` — this only
-    parses syntax.
-    """
-    out: dict[str, object] = {}
-    for spec in specs or []:
-        path, sep, val = spec.partition("=")
-        if not sep:
-            typer.echo(f"error: --set {spec!r} must be 'dotted.path=value'", err=True)
-            raise typer.Exit(1)
-        out[path] = _coerce_scalar(val)
-    return out
-
-
 def _router_cli_overrides(
     router: bool | None,
     router_type: str | None,
@@ -187,10 +192,17 @@ def _write_prediction_ref(
     out: Path,
     dataset_path: Path,
     comment: str | None = None,
+    explicit_out: bool = False,
 ) -> Path:
-    """Write a YAML sidecar in the checkpoint directory and return its path."""
-    import yaml
+    """Write the YAML sidecar and return its path.
 
+    With an explicit ``--out`` the sidecar sits next to the output file as
+    ``out.with_suffix(".yaml")`` — a *deterministic* path, which is what lets
+    a workflow task (``giant/workflow/tasks.py``) declare it as a target.
+    Without one, the historic uuid-named file under the checkpoint directory
+    is kept, so ad-hoc runs and the ``/ceph`` predictions convention are
+    unaffected.
+    """
     ref = {
         "prediction_id": pred_uuid,
         "output": str(out),
@@ -200,7 +212,7 @@ def _write_prediction_ref(
     }
     if comment is not None:
         ref["comment"] = comment
-    ref_path = checkpoint.parent / f"{pred_uuid}.yaml"
+    ref_path = out.with_suffix(".yaml") if explicit_out else checkpoint.parent / f"{pred_uuid}.yaml"
     ref_path.write_text(yaml.dump(ref, default_flow_style=False, sort_keys=False))
     return ref_path
 
@@ -620,10 +632,6 @@ def train(
     ] = None,
 ) -> None:
     """Train the GIANT surrogate model."""
-    import torch
-
-    from giant.pipeline import run_train_job
-
     batch_size_auto = False
     batch_size_value: Optional[int] = None
     if batch_size is not None:
@@ -1022,36 +1030,8 @@ def predict(
             help="Free-text note recorded in the prediction's YAML sidecar",
         ),
     ] = None,
-    set_: Annotated[
-        Optional[list[str]],
-        typer.Option(
-            "--set",
-            help="Override a sampling-only model_config key on this checkpoint, "
-            "'dotted.path=value' (repeatable) — see giant.config.INFERENCE_OVERRIDES "
-            "for the allowlist, e.g. --set stage2_model.n_sec.sampling=sample",
-        ),
-    ] = None,
 ) -> None:
     """Run trained model on a parquet file and save predictions."""
-    import numpy as np
-    import pyarrow as pa
-    import pyarrow.parquet as pq
-    import torch
-    from tqdm import tqdm
-
-    from giant.checkpoint_io import CheckpointCompatibilityError, load_for_inference
-    from giant.data.loader import event_id_offset, find_parquet_files, iter_cond_chunks, iter_file_chunks
-    from giant.data.transforms import (
-        build_cond_features,
-        build_features,
-        energy_simplex_decode,
-        inv_local_frame_rotation,
-        inv_log_transform,
-        reconstruct_post_pos,
-    )
-    from giant.rollout import decode_secondary_identity
-    from giant.sample import resolve_n_sec, sample_stage1, sample_stage2
-
     batch_size_auto = False
     batch_size_value: Optional[int] = None
     if batch_size.strip().lower() == "auto":
@@ -1070,11 +1050,8 @@ def predict(
     typer.echo(f"device: {_device}")
 
     # --- Load checkpoint ---
-    config_overrides = _parse_set_flags(set_)
     try:
-        ctx = load_for_inference(
-            checkpoint, _device, "predict", weights=weights.value, config_overrides=config_overrides
-        )
+        ctx = load_for_inference(checkpoint, _device, "predict", weights=weights.value)
     except CheckpointCompatibilityError as exc:
         typer.echo(f"error: {exc}", err=True)
         raise typer.Exit(1)
@@ -1109,6 +1086,7 @@ def predict(
     bs = batch_size_value
 
     # --- Output path ---
+    explicit_out = out is not None
     out, dataset_path, pred_uuid = _resolve_prediction_output(data, out)
     out.parent.mkdir(parents=True, exist_ok=True)
     typer.echo(f"output: {out}")
@@ -1328,7 +1306,7 @@ def predict(
     if writer is not None:
         writer.close()
 
-    ref_path = _write_prediction_ref(checkpoint, pred_uuid, out, dataset_path, comment)
+    ref_path = _write_prediction_ref(checkpoint, pred_uuid, out, dataset_path, comment, explicit_out=explicit_out)
     typer.echo(f"reference: {ref_path}")
 
     if skipped:
@@ -1347,10 +1325,6 @@ def _seed_from_data(files: list[Path], n_events: int | None) -> dict[str, np.nda
     the codebase's convention for the primary (a secondary always carries less
     energy than its parent). See giant/analysis/reduce.py:entry_axis.
     """
-    import numpy as np
-
-    from giant.data.loader import event_id_offset, iter_cond_chunks
-
     best_E: dict[int, float] = {}
     best: dict[int, tuple] = {}
     for file_idx, path in enumerate(files):
@@ -1444,28 +1418,8 @@ def rollout(
         Optional[int],
         typer.Option("--seed", help="Torch/numpy seed for reproducibility"),
     ] = None,
-    set_: Annotated[
-        Optional[list[str]],
-        typer.Option(
-            "--set",
-            help="Override a sampling-only model_config key on this checkpoint, "
-            "'dotted.path=value' (repeatable) — see giant.config.INFERENCE_OVERRIDES "
-            "for the allowlist, e.g. --set stage2_model.n_sec.sampling=sample",
-        ),
-    ] = None,
 ) -> None:
     """Roll the surrogate forward into full showers (autoregressive)."""
-    import numpy as np
-    import pyarrow as pa
-    import pyarrow.parquet as pq
-    import torch
-    import yaml
-
-    from giant.checkpoint_io import CheckpointCompatibilityError, load_for_inference
-    from giant.data.loader import find_parquet_files
-    from giant.geometry import GeometryOracle
-    from giant.rollout import L1DistCollector, rollout as run_rollout
-
     if seed is not None:
         torch.manual_seed(seed)
         np.random.seed(seed)
@@ -1473,11 +1427,8 @@ def rollout(
     _device = torch.device(device) if device else gconfig.auto_device()
     typer.echo(f"device: {_device}")
 
-    config_overrides = _parse_set_flags(set_)
     try:
-        ctx = load_for_inference(
-            checkpoint, _device, "rollout", weights=weights.value, config_overrides=config_overrides
-        )
+        ctx = load_for_inference(checkpoint, _device, "rollout", weights=weights.value)
     except CheckpointCompatibilityError as exc:
         typer.echo(f"error: {exc}", err=True)
         raise typer.Exit(1)
@@ -1503,6 +1454,7 @@ def rollout(
     seeds = _seed_from_data(files, n_events)
     typer.echo(f"seeded {len(seeds['event_id']):,} shower(s)")
 
+    explicit_out = out is not None
     out, dataset_path, pred_uuid = _resolve_prediction_output(data, out)
     out.parent.mkdir(parents=True, exist_ok=True)
 
@@ -1563,7 +1515,7 @@ def rollout(
 
     l1_summary = l1_dist_collector.summary()
 
-    ref_path = _write_prediction_ref(checkpoint, pred_uuid, out, dataset_path)
+    ref_path = _write_prediction_ref(checkpoint, pred_uuid, out, dataset_path, explicit_out=explicit_out)
     ref = yaml.safe_load(ref_path.read_text())
     ref.update(
         {
@@ -1592,7 +1544,6 @@ def rollout(
             # model knob (router type/n_experts, noise_dim, vocab sizes, ...)
             # is available downstream without touching this command again.
             "model_config": dict(model_cfg),
-            "config_overrides": dict(ctx.config_overrides),
             "training_epoch": ctx.epoch,
             "best_val_loss": ctx.best_val_loss,
             # [train]/[meta] from the sibling config.toml (giant.config.save_config)
@@ -1608,6 +1559,49 @@ def rollout(
     typer.echo(f"reference: {ref_path}")
 
 
+workflow_app = typer.Typer(
+    no_args_is_help=True,
+    help="b2luigi pipeline orchestration: one spec file -> cache-warm, train, rollout, analysis.",
+)
+app.add_typer(workflow_app, name="workflow")
+
+
+@workflow_app.command("run")
+def workflow_run(
+    spec: Annotated[Path, typer.Argument(help="Workflow TOML (see configs/workflow_example.toml)")],
+    batch: Annotated[
+        bool,
+        typer.Option("--batch/--local", help="Submit batch-system tasks to HTCondor, or run everything locally"),
+    ] = False,
+    workers: Annotated[int, typer.Option("--workers", help="Concurrent luigi workers")] = 1,
+    mode: Annotated[
+        str,
+        typer.Option(
+            "--mode",
+            help="run | dry-run (print pending tasks) | show-output (print every target) | remove (delete outputs)",
+        ),
+    ] = "run",
+    scheduler_host: Annotated[Optional[str], typer.Option("--scheduler-host", help="luigid host")] = None,
+    scheduler_port: Annotated[Optional[int], typer.Option("--scheduler-port", help="luigid port")] = None,
+) -> None:
+    """Run a workflow spec end to end (the only sanctioned multi-step entry point).
+
+    A thin exec of `giant/workflow/run.py`, which b2luigi also re-executes on
+    every worker — so there is one documented entry point and one code path.
+    """
+    import subprocess
+
+    script = Path(__file__).resolve().parent / "workflow" / "run.py"
+    cmd = [sys.executable, str(script), "--spec", str(spec), "--workers", str(workers), "--mode", mode]
+    if batch:
+        cmd.append("--batch")
+    if scheduler_host:
+        cmd += ["--scheduler-host", scheduler_host]
+    if scheduler_port:
+        cmd += ["--scheduler-port", str(scheduler_port)]
+    raise typer.Exit(subprocess.run(cmd).returncode)
+
+
 analyze_app = typer.Typer(
     no_args_is_help=True,
     help="Rollout-vs-reference analysis: parallel compute on HTCondor + local render.",
@@ -1738,83 +1732,5 @@ def analyze_metrics(
     typer.echo(f"rendered {len(paths)} plots -> {paths[0].parent if paths else '(nothing to render)'}")
 
 
-@analyze_app.command("submit")
-def analyze_submit(
-    rollout_yamls: Annotated[
-        list[Path],
-        typer.Argument(
-            help="giant rollout YAML sidecar(s). Multiple compare N rollouts against one "
-            "shared reference — every YAML must name the same `dataset`."
-        ),
-    ],
-    accounting_group: Annotated[str, typer.Option("--accounting-group")],
-    label: Annotated[
-        Optional[list[str]],
-        typer.Option(
-            "--label",
-            help="Series name for a rollout YAML, positionally matched to it — give none, "
-            'or exactly one per YAML. Defaults to the YAML stem (or "rollout" for a '
-            "single YAML).",
-        ),
-    ] = None,
-    run_dir: Annotated[
-        Optional[Path],
-        typer.Option(
-            "--run-dir",
-            "-o",
-            help="Override the run directory (default: <cwd>/analysis_runs/analysis_<id>)",
-        ),
-    ] = None,
-    docker_image: Annotated[str, typer.Option("--docker-image")] = "cverstege/alma9-gridjob",
-    request_memory: Annotated[int, typer.Option("--request-memory", help="MB")] = 8192,
-    remote: Annotated[
-        bool,
-        typer.Option("--remote/--local", help="+RemoteJob vs ProvidesETPResources"),
-    ] = False,
-    chunks: Annotated[
-        int,
-        typer.Option(
-            "--chunks",
-            help="Split each plot's data into this many event_id chunks/jobs",
-        ),
-    ] = 1,
-    n_energy_bins: Annotated[int, typer.Option("--energy-bins")] = 4,
-    n_marginal_bins: Annotated[int, typer.Option("--bins")] = 50,
-    top_k_pdg: Annotated[int, typer.Option("--top-pdg")] = 6,
-    dry_run: Annotated[bool, typer.Option("--dry-run", help="Write files but don't condor_submit")] = False,
-) -> None:
-    """prep + write the HTCondor submit description (one job per plot x chunk), then submit."""
-    import subprocess
-
-    from giant.analysis import SubmitConfig, prep, write_submit
-
-    path = prep(
-        rollout_yamls,
-        run_dir,
-        n_chunks=chunks,
-        default_base=Path.cwd() / "analysis_runs",
-        labels=label,
-        n_energy_bins=n_energy_bins,
-        n_marginal_bins=n_marginal_bins,
-        top_k_pdg=top_k_pdg,
-    )
-    cfg = SubmitConfig(
-        run_dir=path,
-        accounting_group=accounting_group,
-        repo_dir=Path.cwd(),
-        docker_image=docker_image,
-        request_memory_mb=request_memory,
-        remote=remote,
-        n_chunks=chunks,
-    )
-    sub = write_submit(cfg)
-    typer.echo(f"run directory: {path}")
-    typer.echo(f"wrote submit description: {sub}")
-    if dry_run:
-        typer.echo("dry-run: not submitting")
-        return
-    subprocess.run(["condor_submit", str(sub)], check=True)
-
-
 if __name__ == "__main__":
     app()

giant/config.py

@@ -1,5 +1,3 @@
-from __future__ import annotations
-
 import copy
 import difflib
 import hashlib
@@ -12,12 +10,12 @@ from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from enum import Enum
 from pathlib import Path
-from typing import TYPE_CHECKING
+
+import numpy as np
+import torch
 
 from giant._migration import V02_FIXED_FACTS, V02_MODEL_KEY_TO_STAGES, reject_legacy_router_expert_sizing
-
-if TYPE_CHECKING:
-    import torch
+from giant.model.history import HISTORY_REGISTRY
 
 
 class Conditioning(str, Enum):
@@ -406,16 +404,6 @@ class Stage2RouterConfig(RouterConfig):
         return {"tie_to_stage1": self.tie_to_stage1, **super().to_dict()}
 
 
-# stage2_model.n_sec.sampling choices — single source of truth for both
-# validate_config's train-time check and INFERENCE_OVERRIDES below.
-STOP_SAMPLING_CHOICES = ("greedy", "sample")
-
-# stage2_model.particle_type.other_policy choices — see ParticleTypeConfig's
-# docstring for what each means; only documented there until now, since
-# nothing validated it at train time.
-OTHER_POLICY_CHOICES = ("sample", "modal", "drop")
-
-
 @dataclass(frozen=True)
 class NSecConfig:
     # "head": a classifier over {0..k_max} on the condition encoding alone
@@ -436,15 +424,11 @@ class NSecConfig:
     # n_sec head was trained against Stage 1's own ConditionEncoder output and so has
     # to stay attached there, not just be labeled as such).
     owner: str = "stage2"
-    # How resolve_n_sec/sample_secondaries_ar turn a count-bearing head's output into an
-    # actual n_sec decision. mode="head": "greedy" is argmax over the classifier logits
-    # (deterministic — the conditional mode, not a sample); "sample" is a categorical draw
-    # from softmax(logits) (a real sample from the learned count distribution). mode=
-    # "stop_token": "greedy" is sigmoid(stop_logit) >= 0.5 per slot (deterministic);
-    # "sample" is a Bernoulli draw at sigmoid(stop_logit) per slot. Renamed from
-    # "stop_sampling" (gitea #86), which is still accepted as a deprecated alias since it
-    # appears in existing checkpoints' model_config.
-    sampling: str = "greedy"
+    # mode="stop_token" only: how sample_secondaries_ar turns a slot's stop logit into a
+    # stop/continue decision. "greedy": sigmoid(logit) >= 0.5 (deterministic). "sample":
+    # a Bernoulli draw at sigmoid(logit) (a real sample from the learned length
+    # distribution, at the cost of an extra RNG draw per slot).
+    stop_sampling: str = "greedy"
 
     @classmethod
     def from_dict(cls, d: dict | None) -> "NSecConfig":
@@ -453,7 +437,7 @@ class NSecConfig:
             mode=d.get("mode", "head"),
             lambda_weight=d.get("lambda", 0.1),
             owner=d.get("owner", "stage2"),
-            sampling=d.get("sampling", d.get("stop_sampling", "greedy")),
+            stop_sampling=d.get("stop_sampling", "greedy"),
         )
 
     def to_dict(self) -> dict:
@@ -461,7 +445,7 @@ class NSecConfig:
             "mode": self.mode,
             "lambda": self.lambda_weight,
             "owner": self.owner,
-            "sampling": self.sampling,
+            "stop_sampling": self.stop_sampling,
         }
 
 
@@ -943,8 +927,6 @@ def git_hash() -> str:
 
 
 def auto_device() -> torch.device:
-    import torch
-
     if torch.cuda.is_available():
         return torch.device("cuda")
     if torch.backends.mps.is_available():
@@ -990,8 +972,6 @@ def estimate_batch_size(
     inference (e.g. `predict`), which uses a much lower per-sample memory
     calibration since there's no backward graph or optimizer state.
     """
-    import torch
-
     if device.type != "cuda":
         raise ValueError(f"--batch-size auto is only supported on cuda devices, got {device.type!r}")
     device_index = device.index if device.index is not None else torch.cuda.current_device()
@@ -1095,27 +1075,6 @@ def _set_path(d: dict, dotted: str, value) -> None:
     cur[parts[-1]] = value
 
 
-def _pop_path(d: dict, dotted: str) -> None:
-    """Remove a dotted path from a nested dict, if present. No-op if any
-    component along the path is missing."""
-    parts = dotted.split(".")
-    cur = d
-    for part in parts[:-1]:
-        if not isinstance(cur, dict) or part not in cur:
-            return
-        cur = cur[part]
-    if isinstance(cur, dict):
-        cur.pop(parts[-1], None)
-
-
-# Config keys renamed within v0.3 itself (not part of the v0.2->v0.3 migration
-# above) — normalized by migrate_config so a config.toml still using an older
-# v0.3 key name keeps passing validate_config_keys.
-_RENAMED_KEYS = {
-    "stage2_model.n_sec.stop_sampling": "stage2_model.n_sec.sampling",  # gitea #86
-}
-
-
 def _deep_merge(base: dict, override: dict) -> dict:
     """Recursively merge `override` onto a copy of `base`.
 
@@ -1134,72 +1093,6 @@ def _deep_merge(base: dict, override: dict) -> dict:
     return result
 
 
-@dataclass(frozen=True)
-class InferenceOverride:
-    """One dotted `model_config` path that `giant.checkpoint_io.load_for_inference`
-    is allowed to change on an already-trained checkpoint, without retraining.
-
-    A path only belongs here if it affects neither module construction/tensor
-    shapes nor the data preprocessing the normalizers/vocab maps were fit
-    under — see the module docstring on `giant.model.summary` for the class
-    of key this targets (`_fingerprint`'s "plain scalar attribute" leaves),
-    and `giant.checkpoint_io.apply_config_overrides` for where this is used.
-    """
-
-    why: str
-    choices: tuple[str, ...] | None = None
-    minimum: float | None = None
-    numeric: bool = False  # int/float leaf (vs. str, the default)
-
-    def check(self, path: str, value: object) -> None:
-        if self.choices is not None:
-            if value not in self.choices:
-                raise ValueError(f"{path} = {value!r} — must be one of {self.choices}")
-            return
-        if self.numeric:
-            if isinstance(value, bool) or not isinstance(value, (int, float)):
-                raise ValueError(f"{path} = {value!r} — must be a number")
-            if self.minimum is not None and value < self.minimum:
-                raise ValueError(f"{path} = {value!r} — must be >= {self.minimum}")
-
-
-# Inference-safe dotted `model_config` paths — the allowlist gitea #87 asked
-# for, so a typo or a shape-bearing key (e.g. "stage1_model.hidden_dim")
-# raises a clear CheckpointCompatibilityError instead of surfacing as an
-# opaque load_state_dict size mismatch later. Extend this table, not a
-# per-call bypass, when a new inference-only key needs the capability.
-INFERENCE_OVERRIDES: dict[str, InferenceOverride] = {
-    "stage2_model.n_sec.sampling": InferenceOverride(
-        why="giant.sample's n_sec head/stop-token sampling reads this at sample time only (gitea #86)",
-        choices=STOP_SAMPLING_CHOICES,
-    ),
-    "stage1_model.ddpm.n_steps": InferenceOverride(
-        why="giant.model.schedule.CosineSchedule's step count, resolved at sample time",
-        numeric=True,
-        minimum=1,
-    ),
-    "stage2_model.ddpm.n_steps": InferenceOverride(
-        why="giant.model.schedule.CosineSchedule's step count, resolved at sample time",
-        numeric=True,
-        minimum=1,
-    ),
-    "stage2_model.particle_type.other_policy": InferenceOverride(
-        why="giant.rollout resolves an 'other'-bucket secondary's PDG code with this at rollout time",
-        choices=OTHER_POLICY_CHOICES,
-    ),
-    "stage1_model.router.temperature": InferenceOverride(
-        why="giant.model.routers.EnergyRouter.temperature, a plain constructor attribute",
-        numeric=True,
-        minimum=1e-6,
-    ),
-    "stage2_model.router.temperature": InferenceOverride(
-        why="giant.model.routers.EnergyRouter.temperature, a plain constructor attribute",
-        numeric=True,
-        minimum=1e-6,
-    ),
-}
-
-
 @dataclass(frozen=True)
 class FlagSpec:
     """One CLI flag's mapping into the config-overrides tree.
@@ -1378,21 +1271,11 @@ def migrate_config(cfg: dict) -> dict:
     (which additionally carries n_sec_head ownership and needs
     `network.build_models`'s cooperation) is a separate migration surface,
     deferred to the network.py refactor.
-
-    Independently of the v0.2/v0.3 branch below, `_RENAMED_KEYS` normalizes
-    keys renamed within v0.3 itself (e.g. `stop_sampling` -> `sampling`,
-    gitea #86) so a config.toml written against an older v0.3 key name still
-    passes `validate_config_keys`.
     """
-    cfg = copy.deepcopy(cfg)
-    for old_path, new_path in _RENAMED_KEYS.items():
-        if _get_path(cfg, old_path) is not None and _get_path(cfg, new_path) is None:
-            _set_path(cfg, new_path, _get_path(cfg, old_path))
-        _pop_path(cfg, old_path)
-
     if _get_path(cfg, "meta.config_version") == CONFIG_VERSION:
-        return cfg
+        return copy.deepcopy(cfg)
 
+    cfg = copy.deepcopy(cfg)
     old_train = cfg.pop("train", {})
     old_model = cfg.pop("model", {})
     old_router = dict(old_model.pop("router", {}))
@@ -1629,9 +1512,9 @@ def validate_config(cfg: dict, *, resume: bool = False) -> None:
                 "conditioning to hang an EOS decision off"
             )
 
-    n_sec_sampling = _get_path(cfg, "stage2_model.n_sec.sampling")
-    if n_sec_sampling not in STOP_SAMPLING_CHOICES:
-        raise ValueError(f"stage2_model.n_sec.sampling = {n_sec_sampling!r} — must be 'greedy' or 'sample'")
+    stop_sampling = _get_path(cfg, "stage2_model.n_sec.stop_sampling")
+    if stop_sampling not in ("greedy", "sample"):
+        raise ValueError(f"stage2_model.n_sec.stop_sampling = {stop_sampling!r} — must be 'greedy' or 'sample'")
 
     precision = _get_path(cfg, "train.precision")
     if precision not in ("fp32", "bf16"):
@@ -1686,8 +1569,6 @@ def validate_config(cfg: dict, *, resume: bool = False) -> None:
                 "'energy_desc' (the only implemented ordering; see "
                 "AutoregressiveConfig.order's docstring)"
             )
-        from giant.model.history import HISTORY_REGISTRY
-
         history = _get_path(cfg, "stage2_model.autoregressive.history")
         if history not in HISTORY_REGISTRY:
             raise ValueError(
@@ -1888,10 +1769,20 @@ def resolve_default_out_dir(cfg: dict, base: Path = Path("checkpoints")) -> Path
     return out_dir
 
 
-def seed_everything(seed: int) -> None:
-    import numpy as np
-    import torch
+def epoch_seed(seed: int, epoch: int) -> int:
+    """Per-epoch derivative of the run seed.
 
+    Reseeding the global RNGs from this at the top of every epoch makes epoch
+    *k* draw the same noise whether it runs inside one long `giant train` or
+    as its own resumed job in a per-epoch workflow chain
+    (`giant/workflow/tasks.py:TrainEpochTask`) — without it, a fresh process
+    would restart the stream at epoch 1's state. Mirrors what
+    `StreamingStepsDataset.set_epoch` does for the batch order.
+    """
+    return (int(seed) * 1_000_003 + int(epoch)) % (2**32)
+
+
+def seed_everything(seed: int) -> None:
     random.seed(seed)
     np.random.seed(seed)
     torch.manual_seed(seed)
@@ -1945,8 +1836,6 @@ def build_run_meta(
     n_val_events: int,
     n_train_steps: int,
 ) -> dict:
-    import torch
-
     return {
         "config_version": CONFIG_VERSION,
         "git_hash": git_hash(),

giant/data/dataset.py

@@ -97,6 +97,7 @@ class StreamingStepsDataset(IterableDataset):
         mat_topn_map: dict[str, int] | None = None,
         sec_type_class_map: dict | None = None,
         k_max: int = K_MAX,
+        seed: int = 0,
     ) -> None:
         self.files = list(files)
         self._offsets = {path: event_id_offset(i) for i, path in enumerate(self.files)}
@@ -117,16 +118,35 @@ class StreamingStepsDataset(IterableDataset):
         self.mat_topn_map = mat_topn_map
         self.sec_type_class_map = sec_type_class_map
         self.k_max = k_max
+        self.seed = seed
+        self.epoch = 0
+        self._rng = np.random.default_rng()
+
+    def set_epoch(self, epoch: int) -> None:
+        """Select the shuffle stream for `epoch` (the DistributedSampler convention).
+
+        The training loop calls this at the top of every epoch. Shuffling is
+        seeded from `(seed, epoch, worker_id)` rather than the global numpy
+        state so epoch *k*'s batch order is the same whether it runs as epoch
+        *k* of one long `giant train`, or as its own resumed job in a
+        per-epoch workflow chain (`giant/workflow/tasks.py:TrainEpochTask`).
+        Workers are re-forked from this object each epoch (no
+        `persistent_workers`), so setting it here reaches them.
+        """
+        self.epoch = int(epoch)
 
     def __iter__(self):
         worker_info = torch.utils.data.get_worker_info()
         files = self.files
+        worker_id = worker_info.id if worker_info is not None else 0
         if worker_info is not None:
-            files = files[worker_info.id :: worker_info.num_workers]
+            files = files[worker_id :: worker_info.num_workers]
+
+        self._rng = np.random.default_rng([self.seed, self.epoch, worker_id])
 
         if self.shuffle:
             files = list(files)
-            np.random.default_rng().shuffle(files)
+            self._rng.shuffle(files)
 
         buf_cont: list[np.ndarray] = []
         buf_cat: list[np.ndarray] = []
@@ -222,7 +242,7 @@ class StreamingStepsDataset(IterableDataset):
         styp = np.concatenate(buf_type)
 
         if self.shuffle:
-            idx = np.random.permutation(len(cont))
+            idx = self._rng.permutation(len(cont))
             cont, cat, tgt = cont[idx], cat[idx], tgt[idx]
             nsec, sec, proc, styp = nsec[idx], sec[idx], proc[idx], styp[idx]
 

giant/model/builders.py

@@ -138,7 +138,7 @@ def build_models(model_config: dict) -> dict[str, nn.Module | None]:
                 n_sec_head_cfg=s2_spec.heads.n_sec.to_dict(),
                 type_head_cfg=s2_spec.heads.type.to_dict(),
                 build_stop_head=stop_token,
-                n_sec_sampling=s2_spec.n_sec.sampling,
+                stop_sampling=s2_spec.n_sec.stop_sampling,
                 stop_head_cfg=s2_spec.heads.n_sec.to_dict(),
             )
         else:
@@ -168,7 +168,6 @@ def build_models(model_config: dict) -> dict[str, nn.Module | None]:
                 cond_enc=shared_cond_enc,
                 n_sec_head_cfg=s2_spec.heads.n_sec.to_dict(),
                 type_head_cfg=s2_spec.heads.type.to_dict(),
-                n_sec_sampling=s2_spec.n_sec.sampling,
             )
 
     return result

giant/model/models.py

@@ -370,7 +370,6 @@ class Stage2OneShot(StageModel):
         cond_enc: ConditionEncoder | None = None,
         n_sec_head_cfg: dict | None = None,
         type_head_cfg: dict | None = None,
-        n_sec_sampling: str = "greedy",
     ) -> None:
         super().__init__(
             pdg_vocab,
@@ -384,7 +383,6 @@ class Stage2OneShot(StageModel):
             particle_type_cfg=particle_type_cfg,
             cond_enc=cond_enc,
         )
-        self.n_sec_sampling = n_sec_sampling
         self._build_context_fusion(x_dim, context_dim, cond_out_dim)
         target = self.particle_type_cfg.target
         type_head_out_dim = None if target == "physical" else k_max * self.type_dim
@@ -500,7 +498,7 @@ class Stage2Autoregressive(StageModel):
         n_sec_head_cfg: dict | None = None,
         type_head_cfg: dict | None = None,
         build_stop_head: bool = False,
-        n_sec_sampling: str = "greedy",
+        stop_sampling: str = "greedy",
         stop_head_cfg: dict | None = None,
     ) -> None:
         super().__init__(
@@ -516,7 +514,7 @@ class Stage2Autoregressive(StageModel):
             cond_enc=cond_enc,
         )
         self.history_kind = history
-        self.n_sec_sampling = n_sec_sampling
+        self.stop_sampling = stop_sampling
         self.context_adapter = ContextAdapter(x_dim, context_dim)
         self.base_fuse = nn.Sequential(
             nn.Linear(cond_out_dim + context_dim, cond_out_dim),

giant/model/summary.py

@@ -37,7 +37,7 @@ from dataclasses import dataclass, field
 
 import torch.nn as nn
 
-from giant.config import INFERENCE_OVERRIDES, _get_path, _set_path, leaf_paths
+from giant.config import _get_path, _set_path, leaf_paths
 from giant.model.builders import build_critics, build_models
 from giant.model.trunks import RoutedTrunk
 
@@ -111,7 +111,6 @@ class ModelSummary:
     pdg_vocab: int
     mat_vocab: int
     vocab_caveats: list[str] = field(default_factory=list)
-    overridable: list[str] = field(default_factory=list)
 
 
 def _build_model_config(cfg: dict, pdg_vocab: int, mat_vocab: int) -> dict:
@@ -141,7 +140,7 @@ def _fingerprint(modules: dict[str, nn.Module]) -> list:
     exist, every parameter's/buffer's shape+dtype (never values — those are
     randomly initialized and irrelevant to *structure*), and every plain
     scalar attribute any module stores on itself (e.g. `Stage2Autoregressive
-    .n_sec_sampling`, `EnergyRouter.temperature`) — this is what makes a
+    .stop_sampling`, `EnergyRouter.temperature`) — this is what makes a
     non-parametric key's effect on construction observable."""
     sig = []
     for stage_name, module in modules.items():
@@ -235,8 +234,6 @@ def summarize_model(cfg: dict, pdg_vocab: int, mat_vocab: int) -> ModelSummary:
         else:
             inert.append(path)
 
-    overridable = sorted(p for p in in_scope if p in INFERENCE_OVERRIDES)
-
     return ModelSummary(
         modules=modules,
         consumed=sorted(consumed),
@@ -245,7 +242,6 @@ def summarize_model(cfg: dict, pdg_vocab: int, mat_vocab: int) -> ModelSummary:
         pdg_vocab=pdg_vocab,
         mat_vocab=mat_vocab,
         vocab_caveats=_vocab_caveats(cfg),
-        overridable=overridable,
     )
 
 
@@ -313,12 +309,6 @@ def render_summary(summary: ModelSummary) -> str:
     else:
         lines.append("  (none)")
 
-    if summary.overridable:
-        lines.append("")
-        lines.append("inference-overridable without retraining (giant predict/rollout --set):")
-        for path in summary.overridable:
-            lines.append(f"  {path}  ({INFERENCE_OVERRIDES[path].why})")
-
     if summary.vocab_caveats:
         lines.append("")
         lines.append("vocab placeholder caveats:")

giant/pipeline.py

@@ -435,6 +435,7 @@ def run_train_job(
         mat_topn_map=cond_mat_topn,
         sec_type_class_map=sec_type_class_map,
         k_max=k_max,
+        seed=t["seed"],
     )
     val_ds = StreamingStepsDataset(
         files=files,

giant/sample.py

@@ -261,7 +261,7 @@ def sample_secondaries_ar(
     slot's own stop logit (`predict_stop`, evaluated on the same prefix
     conditioning as the token itself — see `predict_type`'s docstring for
     why this needs no extra state) decides whether generation should have
-    already stopped, per `sec_decoder.n_sec_sampling` ("greedy": threshold at
+    already stopped, per `sec_decoder.stop_sampling` ("greedy": threshold at
     0; "sample": a Bernoulli draw at `sigmoid(logit)`). A row's own
     `n_sec_pred` is the first slot index where this fires; once every row in
     the batch has fired, the loop breaks before spending a model call on the
@@ -345,7 +345,7 @@ def sample_secondaries_ar(
                 slot_idx,
                 hist=hist,
             ).squeeze(1)
-            if sec_decoder.n_sec_sampling == "sample":
+            if sec_decoder.stop_sampling == "sample":
                 stop_now = torch.rand(B, device=device) < torch.sigmoid(stop_logit)
             else:
                 stop_now = stop_logit >= 0.0
@@ -496,12 +496,7 @@ def resolve_n_sec(
 
     Raises if neither stage owns any n_sec mechanism at all — the only way
     that happens is `stage2_model.n_sec.mode = "truth"`, which is not a valid
-    rollout-/predict-capable checkpoint.
-
-    `n_sec.mode = "head"` resolves the classifier logits per
-    `sec_decoder.n_sec_sampling`: "greedy" (default) takes the conditional
-    mode via argmax; "sample" draws a real sample from the learned count
-    distribution via `torch.multinomial` on the softmax — see gitea #86."""
+    rollout-/predict-capable checkpoint."""
     if n_sec_pred is not None:
         return n_sec_pred
     if getattr(sec_decoder, "stop_head", None) is not None:
@@ -513,6 +508,4 @@ def resolve_n_sec(
             "'truth' is standalone-evaluation-only"
         )
     logits = sec_decoder.predict_n_sec(cond_cont, cond_cat, stage1_out)
-    if sec_decoder.n_sec_sampling == "sample":
-        return torch.multinomial(logits.softmax(dim=-1), 1).squeeze(-1)
     return logits.argmax(dim=-1)

giant/tools/dwarf.py

@@ -5,8 +5,6 @@ simulation-fanout tools into one Typer app so there's a single command name
 (and `--help`) to remember instead of five differently-hyphenated ones.
 """
 
-from __future__ import annotations
-
 import os
 from enum import Enum
 from pathlib import Path
@@ -16,15 +14,20 @@ import typer
 from typing_extensions import Annotated
 
 from giant.config import Conditioning
-
-# DATA_DEFAULT/SCAN_DIR_DEFAULT are Typer option defaults (evaluated at
-# decoration time below), so that one name has to stay eager — the module
-# itself is stdlib-only, so it costs nothing. Every other giant.tools.*
-# import here is deferred into the one command body that uses it, since
-# several (steps_to_parquet: uproot/awkward/polars; warm_setup_cache:
-# giant.pipeline -> torch; geometry_oracle: pandas) are expensive and
-# `dwarf --help`/tab-completion shouldn't pay for all of them upfront.
-from giant.tools.hparam_scan import DATA_DEFAULT, SCAN_DIR_DEFAULT
+from giant.tools.bump_dataset_version import (
+    run_bump_gen,
+    run_bump_schema,
+    run_create_manifest,
+    run_status,
+    run_update_manifest,
+)
+from giant.tools.create_root_files import run_make_root
+from giant.tools.geometry_oracle import run_build_geometry_oracle
+from giant.tools.hparam_scan import DATA_DEFAULT, SCAN_DIR_DEFAULT, run_hparam_scan
+from giant.tools.migrate_geant_steps import run_migration
+from giant.tools.steps_to_parquet import convert_steps_to_parquet
+from giant.tools.steps_to_parquet_parallel import run_parallel_job
+from giant.tools.warm_setup_cache import run_warm_setup_cache
 
 app = typer.Typer(no_args_is_help=True)
 
@@ -118,9 +121,6 @@ def convert(
     ] = None,
 ) -> None:
     """Convert ROOT Steps tree(s) to Parquet."""
-    from giant.tools.steps_to_parquet import convert_steps_to_parquet
-    from giant.tools.steps_to_parquet_parallel import run_parallel_job
-
     if jobs < 1:
         typer.echo("error: --jobs must be >= 1", err=True)
         raise typer.Exit(1)
@@ -183,8 +183,6 @@ def migrate(
     ] = False,
 ) -> None:
     """One-time migration into the versioned raw/processed/pools/derived layout."""
-    from giant.tools.migrate_geant_steps import run_migration
-
     run_migration(str(root), execute=execute, copy=copy)
 
 
@@ -209,8 +207,6 @@ def bump_gen(
     root: Annotated[Path, typer.Option("--root", help="Dataset root")] = _DATASET_ROOT_DEFAULT,
 ) -> None:
     """Cut a new raw generation."""
-    from giant.tools.bump_dataset_version import run_bump_gen
-
     run_bump_gen(
         kind=kind,
         reason=reason,
@@ -244,8 +240,6 @@ def bump_schema(
     root: Annotated[Path, typer.Option("--root", help="Dataset root")] = _DATASET_ROOT_DEFAULT,
 ) -> None:
     """Cut a new schema within a gen."""
-    from giant.tools.bump_dataset_version import run_bump_schema
-
     run_bump_schema(
         kind=kind,
         gen=gen,
@@ -263,8 +257,6 @@ def status(
     root: Annotated[Path, typer.Option("--root", help="Dataset root")] = _DATASET_ROOT_DEFAULT,
 ) -> None:
     """List existing gens/schemas per kind."""
-    from giant.tools.bump_dataset_version import run_status
-
     run_status(str(root))
 
 
@@ -289,8 +281,6 @@ def update_manifest(
     ] = False,
 ) -> None:
     """Repoint manifest(s) to a new gen and/or schema, verifying all target files exist."""
-    from giant.tools.bump_dataset_version import run_update_manifest
-
     run_update_manifest([str(m) for m in manifests], schema=schema, execute=execute, gen=gen)
 
 
@@ -321,8 +311,6 @@ def create_manifest(
     ] = False,
 ) -> None:
     """Create a new manifest from a list of parquet files."""
-    from giant.tools.bump_dataset_version import run_create_manifest
-
     run_create_manifest(
         [str(f) for f in files],
         execute=execute,
@@ -370,8 +358,6 @@ def make_root(
     ] = False,
 ) -> None:
     """Generate new ROOT shards via a minicalosim executable."""
-    from giant.tools.create_root_files import run_make_root
-
     _warn_if_exceeds_shared_quota(jobs, "--jobs")
     run_make_root(
         executable=executable,
@@ -437,8 +423,6 @@ def build_geometry_oracle(
     ] = 2000,
 ) -> None:
     """Fit a position -> (material, layer_id) oracle for `giant rollout`."""
-    from giant.tools.geometry_oracle import run_build_geometry_oracle
-
     run_build_geometry_oracle(
         data=data,
         out=out,
@@ -531,8 +515,6 @@ def warm_cache(
     such entry across every run) skips straight to training. See
     giant/data/setup_cache.py.
     """
-    from giant.tools.warm_setup_cache import run_warm_setup_cache
-
     flag_overrides = {
         "--val-fraction": val_fraction,
         "--seed": seed,
@@ -575,8 +557,6 @@ def hparam_scan(
     dry_run: Annotated[bool, typer.Option("--dry-run")] = False,
 ) -> None:
     """Grid-scan dropout x n_blocks x hidden_dim via sequential `giant train` runs."""
-    from giant.tools.hparam_scan import run_hparam_scan
-
     run_hparam_scan(data=data, scan_dir=scan_dir, seed=seed, dry_run=dry_run)
 
 

giant/tools/profile_analysis_costs.py

@@ -26,7 +26,7 @@ import numpy as np
 import polars as pl
 
 from giant.analysis.catalog import catalog_ids, get_spec
-from giant.analysis.condor import compute_reduced
+from giant.analysis.run import compute_reduced
 from giant.analysis.context import build_context
 from giant.analysis.sources import RolloutSpec
 

giant/training/loop.py

@@ -18,6 +18,7 @@ import torch
 from torch.utils.data import DataLoader
 from tqdm import tqdm
 
+from giant import config
 from giant.data.loader import TopNMap
 from giant.data.setup_cache import topnmap_to_json
 from giant.training.checkpoint import build_checkpoint, init_stages_from_checkpoints, load_checkpoint
@@ -184,6 +185,17 @@ def train(
             if device.type == "cuda":
                 torch.cuda.reset_peak_memory_stats(device)
             collector.start_epoch(epoch)
+            # Epoch-aware RNG: same noise (and, below, same batch order) for
+            # epoch k whether the run is one process or a chain of per-epoch
+            # jobs. See giant.config.epoch_seed.
+            config.seed_everything(config.epoch_seed(t["seed"], epoch))
+            # Epoch-aware shuffle stream (see StreamingStepsDataset.set_epoch):
+            # keeps epoch k's batch order identical whether it runs here or as
+            # its own resumed per-epoch job in a b2luigi workflow.
+            # (tests hand `train` a plain list of batches, which has neither)
+            set_epoch = getattr(getattr(train_loader, "dataset", None), "set_epoch", None)
+            if callable(set_epoch):
+                set_epoch(epoch)
             for trainer in trainers.values():
                 trainer.train_mode()
 

giant/training/plots.py

@@ -115,7 +115,7 @@ def derive_metrics_dir(
 
     Precedence: an explicit `out_dir` always wins. Otherwise
     `default_base / f"metrics_{run_dir.name}"` (the CLI passes the repo's
-    gitignored `analysis_runs/`, matching `giant.analysis.condor.derive_run_dir`'s
+    gitignored `analysis_runs/`, matching `giant.analysis.run.derive_run_dir`'s
     convention) — training-progress plots live alongside rollout-vs-reference
     analysis runs, not inside the training run directory itself.
     """

giant/workflow/__init__.py

@@ -0,0 +1,40 @@
+"""b2luigi orchestration of the full GIANT pipeline.
+
+One workflow TOML (``spec.py``) parameterises an entire experiment — dataset,
+geometry oracle, N trainings, N rollouts, N analyses — and ``giant workflow
+run <spec.toml>`` turns it into a b2luigi task graph (``tasks.py``) whose
+targets are files on ``/ceph``: nothing is recomputed that already exists,
+every step waits for its inputs, and HTCondor submission/polling is b2luigi's
+job rather than a hand-rolled submit-file generator.
+
+This is the only sanctioned way to run a multi-step pipeline; ``giant`` and
+``dwarf`` stay single-step primitives that these tasks invoke.
+
+``tasks``/``run`` import b2luigi, so they are *not* imported here — a plain
+``import giant.workflow`` (or ``giant.workflow.spec``) works without the
+``workflow`` extra installed.
+"""
+
+from giant.workflow.spec import (
+    AnalysisSpec,
+    CondorSpec,
+    DatasetSpec,
+    GeometrySpec,
+    RolloutSpec,
+    TrainSpec,
+    WorkflowSpec,
+    load_spec,
+    spec_hash,
+)
+
+__all__ = [
+    "AnalysisSpec",
+    "CondorSpec",
+    "DatasetSpec",
+    "GeometrySpec",
+    "RolloutSpec",
+    "TrainSpec",
+    "WorkflowSpec",
+    "load_spec",
+    "spec_hash",
+]

giant/workflow/htcondor.py

@@ -0,0 +1,82 @@
+"""HTCondor job descriptions for the workflow tasks.
+
+b2luigi writes every key of a task's ``htcondor_settings`` dict straight into
+that job's submit description, so these helpers are just the ETP-specific
+resource/requirement conventions in one place:
+
+* **CPU jobs** (setup cache, geometry oracle, analysis compute) keep what
+  the deleted ``giant analyze submit`` used: ``+RemoteJob`` for grid I/O, or
+  ``TARGET.ProvidesETPResources`` when the files are local to the cluster.
+* **GPU jobs** (training epochs, rollout) are remote-only, so they always
+  carry ``+RemoteJob`` and reach ``/ceph`` through
+  ``TARGET.ProvidesEtpCeph`` — the requirement strings are ported from the
+  ``condor-gpu-train-rollout`` branch's ``giant/condor.py`` rather than
+  rewritten, since they encode what the ETP HTCondor wiki documents for
+  TOpAS/NEMO2 GPU workers.
+"""
+
+from __future__ import annotations
+
+from giant.workflow.spec import CondorSpec
+
+__all__ = ["cpu_settings", "gpu_settings", "gpu_requirements"]
+
+
+def cpu_settings(
+    condor: CondorSpec,
+    *,
+    request_memory_mb: int | None = None,
+    request_cpus: int | None = None,
+    walltime_s: int | None = None,
+) -> dict:
+    settings: dict = {
+        "universe": "docker",
+        "docker_image": condor.docker_image_cpu,
+        "request_memory": request_memory_mb if request_memory_mb is not None else condor.request_memory_mb,
+        "request_cpus": request_cpus if request_cpus is not None else condor.request_cpus,
+        "accounting_group": condor.accounting_group,
+        "should_transfer_files": "YES",
+        "when_to_transfer_output": "ON_EXIT",
+    }
+    if condor.remote:
+        settings["+RemoteJob"] = "True"
+    else:
+        settings["requirements"] = "TARGET.ProvidesETPResources"
+    if walltime_s is not None:
+        settings["+RequestWalltime"] = int(walltime_s)
+    return settings
+
+
+def gpu_requirements(gpu_type: str | None = None, gpu_memory_mb: int | None = None) -> str:
+    """``TARGET.ProvidesEtpCeph`` (remote /ceph access) ANDed with any GPU pin."""
+    clauses = ["TARGET.ProvidesEtpCeph =?= True"]
+    if gpu_type is not None:
+        clauses.append(f'TARGET.GPUs_DeviceName =?= "{gpu_type}"')
+    if gpu_memory_mb is not None:
+        clauses.append(f"TARGET.GPUs_GlobalMemoryMb >= {gpu_memory_mb}")
+    return " && ".join(clauses)
+
+
+def gpu_settings(
+    condor: CondorSpec,
+    *,
+    request_gpus: int = 1,
+    gpu_type: str | None = None,
+    gpu_memory_mb: int | None = None,
+    request_memory_mb: int = 16384,
+    request_cpus: int = 4,
+    walltime_s: int = 86400,
+) -> dict:
+    return {
+        "universe": "docker",
+        "docker_image": condor.docker_image_gpu,
+        "request_memory": request_memory_mb,
+        "request_cpus": request_cpus,
+        "RequestGPUs": request_gpus,
+        "+RequestWalltime": int(walltime_s),
+        "accounting_group": condor.accounting_group,
+        "should_transfer_files": "YES",
+        "when_to_transfer_output": "ON_EXIT",
+        "+RemoteJob": "True",
+        "requirements": f"({gpu_requirements(gpu_type, gpu_memory_mb)})",
+    }

giant/workflow/run.py

@@ -0,0 +1,103 @@
+#!/usr/bin/env python
+"""Entry point b2luigi re-executes on every worker.
+
+Locally this is what ``giant workflow run <spec.toml>`` execs; on a batch
+worker it is what the generated wrapper script runs (after ``cd repo_dir`` and
+sourcing ``env_script``), with ``--spec`` forwarded via the
+``task_cmd_additional_args`` setting so the worker resolves exactly the same
+spec — and therefore the same task graph and output paths — as the submitter.
+
+b2luigi needs a real script path for that re-execution, which is why this is a
+script rather than a ``python -m`` module.
+"""
+
+from __future__ import annotations
+
+import argparse
+import sys
+from pathlib import Path
+
+# Allow `python giant/workflow/run.py` from a checkout that isn't installed.
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+
+import b2luigi  # noqa: E402
+
+from giant.workflow.spec import WorkflowSpec, load_spec  # noqa: E402
+from giant.workflow.tasks import WorkflowTask, set_spec  # noqa: E402
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(description="Run a GIANT workflow spec with b2luigi.")
+    parser.add_argument("--spec", required=True, help="Workflow TOML (see configs/workflow_example.toml)")
+    parser.add_argument("--workers", type=int, default=1, help="Concurrent luigi workers")
+    parser.add_argument(
+        "--batch",
+        action="store_true",
+        help="Submit batch-system tasks to HTCondor (otherwise everything runs locally)",
+    )
+    parser.add_argument(
+        "--mode",
+        choices=("run", "dry-run", "show-output", "remove"),
+        default="run",
+        help="run (default), dry-run (print pending tasks), show-output (print every target), remove (delete outputs)",
+    )
+    parser.add_argument("--scheduler-host", default=None, help="luigid host (default: local scheduler)")
+    parser.add_argument("--scheduler-port", type=int, default=None, help="luigid port")
+    return parser
+
+
+def configure(spec: WorkflowSpec, spec_path: Path, batch: bool) -> None:
+    """Wire b2luigi's settings from the spec.
+
+    ``/ceph`` is shared between submit host and workers, so there is
+    deliberately no ``transfer_files``: ``result_dir``/``log_dir`` must live
+    somewhere both sides can see.
+    """
+    set_spec(spec)
+
+    b2luigi.set_setting("result_dir", spec.result_dir)
+    b2luigi.set_setting("log_dir", spec.log_dir)
+    b2luigi.set_setting("task_file_dir", str(Path(spec.result_dir) / "task_files"))
+    b2luigi.set_setting("use_parameter_name_in_output", True)
+    b2luigi.set_setting("batch_system", "htcondor" if batch else "local")
+    b2luigi.set_setting("working_dir", spec.condor.repo_dir)
+    b2luigi.set_setting("job_name", spec.name)
+    if spec.condor.env_script:
+        b2luigi.set_setting("env_script", spec.condor.env_script)
+    # The worker command is `<executable> [<basename of this file>] --batch-runner
+    # --task-id ...`, run after `cd working_dir`. Only the *basename* would be
+    # used, so the filename is dropped and the repo-relative script path is
+    # made part of the executable instead.
+    b2luigi.set_setting("add_filename_to_cmd", False)
+    b2luigi.set_setting("executable", [".venv/bin/python", "giant/workflow/run.py"])
+    b2luigi.set_setting("task_cmd_additional_args", ["--spec", str(spec_path)])
+
+
+def main(argv: list[str] | None = None) -> None:
+    args, _ = build_parser().parse_known_args(argv)
+    spec_path = Path(args.spec).resolve()
+    spec = load_spec(spec_path)
+    configure(spec, spec_path, batch=args.batch)
+
+    kwargs: dict = {}
+    if args.scheduler_host:
+        kwargs["scheduler_host"] = args.scheduler_host
+    if args.scheduler_port:
+        kwargs["scheduler_port"] = args.scheduler_port
+
+    b2luigi.process(
+        WorkflowTask(workflow_name=spec.name),
+        workers=args.workers,
+        batch=args.batch,
+        dry_run=args.mode == "dry-run",
+        show_output=args.mode == "show-output",
+        remove=args.mode == "remove",
+        auto_confirm=args.mode == "remove",
+        # run.py owns --spec/--mode/...; b2luigi must not choke on them.
+        ignore_additional_command_line_args=True,
+        **kwargs,
+    )
+
+
+if __name__ == "__main__":
+    main()

giant/workflow/spec.py

@@ -0,0 +1,383 @@
+"""Workflow TOML -> frozen dataclasses, validation, and per-task spec hashes.
+
+One spec file is the only place a pipeline is parameterised (see
+``configs/workflow_example.toml``):
+
+    [workflow]  name / result_dir / log_dir
+    [condor]    accounting group, repo dir, env script, docker images
+    [dataset]   steps (training) + reference (rollout seeds & analysis truth)
+    [geometry]  geometry-oracle build options
+    [[train]]   one per training run      (name, config, epochs, overrides, ...)
+    [[rollout]] one per rollout           (name, train = <a [[train]].name>, ...)
+    [[analysis]] one per comparison       (name, rollouts = [<[[rollout]].name>, ...])
+
+Every task carries its ``name`` plus a short ``spec_hash`` — 8 hex of the
+canonical JSON of its own resolved sub-spec **including its transitive
+parents**. That is what makes an edited spec produce a fresh result directory
+instead of silently reusing outputs computed under different settings: change
+the dataset and every hash downstream of it changes too.
+
+Unknown keys are rejected (with the valid ones listed), in the same spirit as
+``giant.config.validate_config_keys`` — a typo in a workflow spec would
+otherwise be a silently ignored setting on a multi-day pipeline.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import tomllib
+from dataclasses import MISSING, dataclass, field, fields, is_dataclass
+from pathlib import Path
+from typing import Any
+
+__all__ = [
+    "AnalysisSpec",
+    "CondorSpec",
+    "DatasetSpec",
+    "GeometrySpec",
+    "RolloutSpec",
+    "TrainSpec",
+    "WorkflowSpec",
+    "load_spec",
+    "spec_hash",
+]
+
+
+class WorkflowSpecError(ValueError):
+    """Raised for any malformed workflow spec (unknown key, bad reference, ...)."""
+
+
+# ---------------------------------------------------------------------------
+# sub-specs
+# ---------------------------------------------------------------------------
+
+
+@dataclass(frozen=True)
+class CondorSpec:
+    """Where and how jobs run — the batch-system half of the spec.
+
+    ``repo_dir`` doubles as b2luigi's ``working_dir`` (jobs ``cd`` there before
+    running ``giant/workflow/run.py``), and ``env_script`` is sourced first,
+    since submit and worker machines don't share an environment.
+    """
+
+    accounting_group: str
+    repo_dir: str
+    env_script: str = ""
+    docker_image_cpu: str = "cverstege/alma9-gridjob"
+    docker_image_gpu: str = "mschnepf/slc7-condocker"
+    remote: bool = True
+    request_cpus: int = 1
+    request_memory_mb: int = 8192
+
+
+@dataclass(frozen=True)
+class DatasetSpec:
+    """The two datasets every pipeline needs.
+
+    ``steps`` is what training reads; ``reference`` is the held-out file
+    rollouts are seeded from and the analysis compares against (the "one
+    ground truth" premise of ``giant.analysis``).
+    """
+
+    steps: str
+    reference: str
+
+
+@dataclass(frozen=True)
+class GeometrySpec:
+    """``dwarf build-geometry-oracle`` options (see giant/tools/geometry_oracle.py)."""
+
+    method: str = "slab"
+    k: int = 1
+    subsample: int = 500_000
+    escape_factor: float = 5.0
+    seed: int = 0
+    depth_axis: int = 2
+    n_bins: int = 2000
+
+
+@dataclass(frozen=True)
+class TrainSpec:
+    """One training run, fanned out into ``ceil(epochs / epochs_per_job)`` jobs.
+
+    ``overrides`` are ``[train]``/model config keys merged on top of ``config``
+    exactly as ``giant train``'s flags are (``giant.config.merge_cli_overrides``),
+    so anything expressible on the CLI is expressible here.
+    """
+
+    name: str
+    config: str | None = None
+    epochs: int = 1
+    epochs_per_job: int = 1
+    overrides: dict[str, Any] = field(default_factory=dict)
+    request_gpus: int = 1
+    gpu_type: str | None = None
+    gpu_memory_mb: int | None = None
+    request_memory_mb: int = 16384
+    request_cpus: int = 4
+    walltime_s: int = 86400
+    num_workers: int = 4
+    shuffle_buffer: int = 65536
+    device: str | None = None
+
+
+@dataclass(frozen=True)
+class RolloutSpec:
+    """One ``giant rollout`` run against the checkpoint of ``train``."""
+
+    name: str
+    train: str
+    n_events: int | None = None
+    energy_cutoff: float = 0.1
+    max_steps: int = 1000
+    steps: int = 10
+    batch_size: int = 4096
+    max_tracks_per_event: int | None = None
+    escape_threshold: float | None = None
+    weights: str = "raw"
+    seed: int | None = None
+    request_gpus: int = 1
+    gpu_type: str | None = None
+    gpu_memory_mb: int | None = None
+    request_memory_mb: int = 16384
+    request_cpus: int = 2
+    walltime_s: int = 86400
+    device: str | None = None
+
+
+@dataclass(frozen=True)
+class AnalysisSpec:
+    """One rollout-vs-reference comparison (N rollout series, one reference)."""
+
+    name: str
+    rollouts: tuple[str, ...]
+    chunks: int = 1
+    energy_bins: int = 4
+    bins: int = 50
+    top_pdg: int = 6
+    gallery: bool = False
+    request_memory_mb: int = 8192
+    request_cpus: int = 1
+
+
+@dataclass(frozen=True)
+class WorkflowSpec:
+    """A whole pipeline: the parsed spec file plus name-keyed lookups."""
+
+    name: str
+    result_dir: str
+    log_dir: str
+    condor: CondorSpec
+    dataset: DatasetSpec
+    geometry: GeometrySpec
+    trains: tuple[TrainSpec, ...]
+    rollouts: tuple[RolloutSpec, ...]
+    analyses: tuple[AnalysisSpec, ...]
+    path: str = ""
+
+    # -- lookups ----------------------------------------------------------
+    def train(self, name: str) -> TrainSpec:
+        return _lookup(self.trains, name, "train")
+
+    def rollout(self, name: str) -> RolloutSpec:
+        return _lookup(self.rollouts, name, "rollout")
+
+    def analysis(self, name: str) -> AnalysisSpec:
+        return _lookup(self.analyses, name, "analysis")
+
+    # -- hashes -----------------------------------------------------------
+    # Each one folds in everything upstream of it, so a change anywhere in a
+    # task's ancestry moves its result directory (and only the affected
+    # subtree's).
+    def dataset_hash(self) -> str:
+        return spec_hash(self.dataset)
+
+    def warm_cache_hash(self, train_name: str) -> str:
+        # The setup cache depends on the dataset and on what this training's
+        # config asks of it (val split, conditioning, router) — not on how
+        # many epochs it runs for, so epochs/resources are deliberately left
+        # out and two trainings sharing a config share one warm-cache job.
+        t = self.train(train_name)
+        return spec_hash(self.dataset, t.config, t.overrides)
+
+    def geometry_hash(self) -> str:
+        return spec_hash(self.dataset, self.geometry)
+
+    def train_hash(self, name: str) -> str:
+        return spec_hash(self.dataset, self.train(name))
+
+    def rollout_hash(self, name: str) -> str:
+        ro = self.rollout(name)
+        return spec_hash(self.dataset, self.geometry, self.train(ro.train), ro)
+
+    def analysis_hash(self, name: str) -> str:
+        an = self.analysis(name)
+        parents = [self.rollout(r) for r in an.rollouts]
+        train_parents = [self.train(r.train) for r in parents]
+        return spec_hash(self.dataset, self.geometry, train_parents, parents, an)
+
+
+def _lookup(items, name: str, kind: str):
+    for item in items:
+        if item.name == name:
+            return item
+    known = ", ".join(sorted(i.name for i in items)) or "(none defined)"
+    raise WorkflowSpecError(f"no [[{kind}]] named {name!r} in this workflow — defined: {known}")
+
+
+# ---------------------------------------------------------------------------
+# hashing
+# ---------------------------------------------------------------------------
+
+
+def spec_hash(*parts: Any, length: int = 8) -> str:
+    """Short stable hash of one or more (sub-)specs.
+
+    Canonical JSON (sorted keys, dataclasses expanded) so the value depends
+    only on the resolved settings — not on key order in the TOML, nor on
+    which defaults were written out explicitly.
+    """
+    payload = json.dumps([_canonical(p) for p in parts], sort_keys=True, separators=(",", ":"))
+    return hashlib.sha256(payload.encode()).hexdigest()[:length]
+
+
+def _canonical(value: Any) -> Any:
+    if is_dataclass(value) and not isinstance(value, type):
+        return {f.name: _canonical(getattr(value, f.name)) for f in fields(value)}
+    if isinstance(value, dict):
+        return {str(k): _canonical(v) for k, v in value.items()}
+    if isinstance(value, (list, tuple)):
+        return [_canonical(v) for v in value]
+    if isinstance(value, Path):
+        return str(value)
+    return value
+
+
+# ---------------------------------------------------------------------------
+# parsing
+# ---------------------------------------------------------------------------
+
+
+def _build(cls, data: dict, where: str):
+    """Instantiate a frozen sub-spec, rejecting unknown/missing keys loudly."""
+    valid = {f.name for f in fields(cls)}
+    unknown = sorted(set(data) - valid)
+    if unknown:
+        raise WorkflowSpecError(f"{where}: unknown key(s) {unknown} — valid keys: {sorted(valid)}")
+    required = {f.name for f in fields(cls) if f.default is MISSING and f.default_factory is MISSING}
+    missing = sorted(required - set(data))
+    if missing:
+        raise WorkflowSpecError(f"{where}: missing required key(s) {missing}")
+    return cls(**data)
+
+
+def load_spec(path: str | Path) -> WorkflowSpec:
+    """Parse and validate a workflow TOML file."""
+    path = Path(path)
+    try:
+        raw = tomllib.loads(path.read_text())
+    except tomllib.TOMLDecodeError as exc:
+        raise WorkflowSpecError(f"{path}: not valid TOML — {exc}") from exc
+    return parse_spec(raw, path=path)
+
+
+def parse_spec(raw: dict, path: str | Path = "") -> WorkflowSpec:
+    """Validate an already-parsed workflow spec mapping."""
+    top_valid = {"workflow", "condor", "dataset", "geometry", "train", "rollout", "analysis"}
+    unknown = sorted(set(raw) - top_valid)
+    if unknown:
+        raise WorkflowSpecError(
+            f"{path or '<spec>'}: unknown top-level table(s) {unknown} — valid: {sorted(top_valid)}"
+        )
+
+    for required in ("workflow", "condor", "dataset"):
+        if required not in raw:
+            raise WorkflowSpecError(f"{path or '<spec>'}: missing required [{required}] table")
+
+    wf = dict(raw["workflow"])
+    wf_valid = {"name", "result_dir", "log_dir"}
+    wf_unknown = sorted(set(wf) - wf_valid)
+    if wf_unknown:
+        raise WorkflowSpecError(f"[workflow]: unknown key(s) {wf_unknown} — valid keys: {sorted(wf_valid)}")
+    if "name" not in wf or "result_dir" not in wf:
+        raise WorkflowSpecError("[workflow]: 'name' and 'result_dir' are required")
+    result_dir = str(Path(wf["result_dir"]).expanduser())
+    log_dir = str(Path(wf.get("log_dir", Path(result_dir) / "logs")).expanduser())
+
+    condor = _build(CondorSpec, dict(raw["condor"]), "[condor]")
+    dataset = _build(DatasetSpec, dict(raw["dataset"]), "[dataset]")
+    geometry = _build(GeometrySpec, dict(raw.get("geometry", {})), "[geometry]")
+
+    trains = tuple(_build(TrainSpec, dict(t), f"[[train]] #{i}") for i, t in enumerate(raw.get("train", [])))
+    rollouts = tuple(_build(RolloutSpec, dict(r), f"[[rollout]] #{i}") for i, r in enumerate(raw.get("rollout", [])))
+    analyses = tuple(
+        _build(AnalysisSpec, {**a, "rollouts": tuple(a.get("rollouts", ()))}, f"[[analysis]] #{i}")
+        for i, a in enumerate(raw.get("analysis", []))
+    )
+
+    _check_unique(trains, "train")
+    _check_unique(rollouts, "rollout")
+    _check_unique(analyses, "analysis")
+
+    train_names = {t.name for t in trains}
+    for ro in rollouts:
+        if ro.train not in train_names:
+            raise WorkflowSpecError(
+                f"[[rollout]] {ro.name!r}: train={ro.train!r} names no [[train]] — defined: {sorted(train_names)}"
+            )
+    rollout_names = {r.name for r in rollouts}
+    for an in analyses:
+        if not an.rollouts:
+            raise WorkflowSpecError(f"[[analysis]] {an.name!r}: 'rollouts' must name at least one [[rollout]]")
+        for r in an.rollouts:
+            if r not in rollout_names:
+                raise WorkflowSpecError(
+                    f"[[analysis]] {an.name!r}: rollout {r!r} is not defined — "
+                    f"defined: {sorted(rollout_names) or '(none)'}"
+                )
+        if len(set(an.rollouts)) != len(an.rollouts):
+            raise WorkflowSpecError(f"[[analysis]] {an.name!r}: repeated rollout name(s) in 'rollouts'")
+        if an.chunks < 1:
+            raise WorkflowSpecError(f"[[analysis]] {an.name!r}: chunks must be >= 1, got {an.chunks}")
+
+    for t in trains:
+        if t.epochs < 1:
+            raise WorkflowSpecError(f"[[train]] {t.name!r}: epochs must be >= 1, got {t.epochs}")
+        if t.epochs_per_job < 1:
+            raise WorkflowSpecError(f"[[train]] {t.name!r}: epochs_per_job must be >= 1, got {t.epochs_per_job}")
+
+    return WorkflowSpec(
+        name=wf["name"],
+        result_dir=result_dir,
+        log_dir=log_dir,
+        condor=condor,
+        dataset=dataset,
+        geometry=geometry,
+        trains=trains,
+        rollouts=rollouts,
+        analyses=analyses,
+        path=str(path),
+    )
+
+
+def _check_unique(items, kind: str) -> None:
+    names = [i.name for i in items]
+    dupes = sorted({n for n in names if names.count(n) > 1})
+    if dupes:
+        raise WorkflowSpecError(f"[[{kind}]] names must be unique — repeated: {dupes}")
+
+
+def epoch_milestones(train: TrainSpec) -> list[int]:
+    """Cumulative epoch counts, one per chained ``TrainEpochTask``.
+
+    ``epochs_per_job`` trades queue waits against job length: with
+    ``epochs=10, epochs_per_job=3`` this is ``[3, 6, 9, 10]``, i.e. job *k*
+    resumes job *k-1*'s ``last.pt`` and trains up to its own milestone.
+    """
+    step = train.epochs_per_job
+    milestones = list(range(step, train.epochs + 1, step))
+    if not milestones or milestones[-1] != train.epochs:
+        milestones.append(train.epochs)
+    return milestones

giant/workflow/tasks.py

@@ -0,0 +1,647 @@
+"""The b2luigi task graph: cache-warm -> train -> rollout -> analysis.
+
+    DatasetTask (external)  ─┬─> WarmCacheTask(train) ─> TrainEpochTask(train, 1..N) ─> TrainTask(train) ─┐
+                             └─> GeometryOracleTask ──┐                                                   │
+                                                      └──> RolloutTask(rollout) <────────────────────────┘
+                                                                       │
+    AnalysisPrepTask(analysis) ─> AnalysisComputeTask(analysis, plot, chunk) ─> AnalysisRenderTask(analysis)
+                                                                                        ^
+    WorkflowTask (wrapper) ─────────────────────────────────────────────────────────────┘
+
+Every task's output directory is ``<result_dir>/<kind>/name=<name>/spec_hash=
+<hash>/…`` — the hash covers the task's resolved sub-spec *and its transitive
+parents* (``giant/workflow/spec.py``), so editing the spec produces a fresh
+directory for exactly the affected subtree instead of silently reusing stale
+outputs.
+
+Task bodies never reimplement anything: they call the same entry points the
+CLIs do (``run_warm_setup_cache``, ``run_build_geometry_oracle``,
+``run_train_job``, ``giant.analysis.prep``/``compute_one``/``merge_all``,
+``render_run``), or shell out to ``giant rollout``, which has no library-level
+entry point of its own.
+
+Training is fanned out into **one short GPU job per epoch** (or per
+``epochs_per_job`` epochs): job *k* runs ``run_train_job`` with ``epochs = k``
+and ``resume = <job k-1>/last.pt``, which the training loop already handles
+(``giant/training/loop.py`` sets ``start_epoch = ckpt["epoch"] + 1`` and
+returns early when the checkpoint already covers ``epochs``). A 200-epoch run
+then becomes 200 schedulable jobs that survive preemption and give luigi a
+real progress signal, at the cost of one (cache-warmed) setup scan and one
+queue wait per job.
+"""
+
+from __future__ import annotations
+
+import json
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+
+import b2luigi
+
+from giant.workflow.htcondor import cpu_settings, gpu_settings
+from giant.workflow.spec import WorkflowSpec, epoch_milestones
+
+__all__ = [
+    "AnalysisComputeTask",
+    "AnalysisPrepTask",
+    "AnalysisRenderTask",
+    "DatasetTask",
+    "GeometryOracleTask",
+    "RolloutTask",
+    "TrainEpochTask",
+    "TrainTask",
+    "WarmCacheTask",
+    "WorkflowTask",
+    "analysis_dir",
+    "analysis_jobs",
+    "get_spec",
+    "set_spec",
+]
+
+
+# ---------------------------------------------------------------------------
+# the active spec
+# ---------------------------------------------------------------------------
+# luigi parameters must be simple scalars, so tasks carry only `name` +
+# `spec_hash` and read the rest out of the one spec this process was started
+# with. Batch workers re-execute `run.py --spec <same file>` (see
+# `task_cmd_additional_args` there), so they resolve the identical spec.
+
+_SPEC: WorkflowSpec | None = None
+
+
+def set_spec(spec: WorkflowSpec) -> None:
+    global _SPEC
+    _SPEC = spec
+
+
+def get_spec() -> WorkflowSpec:
+    if _SPEC is None:
+        raise RuntimeError("no workflow spec loaded — call giant.workflow.tasks.set_spec() first")
+    return _SPEC
+
+
+def _result_dir(*parts: str) -> Path:
+    return Path(get_spec().result_dir).joinpath(*parts)
+
+
+def _task_dir(kind: str, name: str, spec_hash: str) -> Path:
+    """``<result_dir>/<kind>/name=<name>/spec_hash=<hash>``."""
+    return _result_dir(kind, f"name={name}", f"spec_hash={spec_hash}")
+
+
+def analysis_dir(spec: WorkflowSpec, name: str) -> Path:
+    """The analysis run directory — what ``prep`` lays out and every later step reads."""
+    return Path(spec.result_dir) / "analysis" / f"name={name}" / f"spec_hash={spec.analysis_hash(name)}"
+
+
+def analysis_jobs(spec: WorkflowSpec, name: str) -> list[tuple[str, int]]:
+    """Every ``(plot_id, chunk)`` compute job of one analysis.
+
+    ``chunkable=False`` specs (the checkpoint-bound diagnostics, already
+    bounded/subsampled) always run as a single chunk — the same rule the
+    deleted ``_job_walltimes`` applied.
+    """
+    from giant.analysis.catalog import catalog_ids, get_spec as get_plot_spec
+
+    chunks = spec.analysis(name).chunks
+    jobs: list[tuple[str, int]] = []
+    for plot_id in catalog_ids():
+        n = chunks if get_plot_spec(plot_id).chunkable else 1
+        jobs.extend((plot_id, chunk) for chunk in range(n))
+    return jobs
+
+
+def _giant_cmd() -> list[str]:
+    """How to invoke the ``giant`` CLI from inside a task (worker or locally)."""
+    return [sys.executable, "-m", "giant.cli"]
+
+
+# ---------------------------------------------------------------------------
+# inputs
+# ---------------------------------------------------------------------------
+
+
+class DatasetTask(b2luigi.ExternalTask):
+    """A steps parquet file or directory that must already exist.
+
+    Nothing produces it, so a missing path is a hard, immediate error rather
+    than a job that fails hours later — the usual cause being ``/ceph`` not
+    mounted on the machine the workflow was started from.
+    """
+
+    path = b2luigi.Parameter()
+
+    def output(self):
+        return b2luigi.LocalTarget(str(self.path))
+
+    def complete(self):
+        if not Path(str(self.path)).exists():
+            raise FileNotFoundError(
+                f"dataset {self.path!r} does not exist — is /ceph mounted on this machine? "
+                "(see CLAUDE.md's Compute environment section)"
+            )
+        return True
+
+
+# ---------------------------------------------------------------------------
+# setup stage
+# ---------------------------------------------------------------------------
+
+
+class WarmCacheTask(b2luigi.Task):
+    """Precompute one training's setup-stage sidecar (vocab maps, event split,
+    normalizer stats) so every per-epoch job is a cache hit instead of a
+    full rescan.
+
+    The real product (``<data>.giant_train_cache.json``) lives next to the
+    dataset, not under ``result_dir``, so the target here is a small stamp
+    recording that sidecar's path/mtime/size.
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+
+    @property
+    def htcondor_settings(self):
+        spec = get_spec()
+        return cpu_settings(spec.condor, request_memory_mb=32768, request_cpus=4, walltime_s=21600)
+
+    def requires(self):
+        yield DatasetTask(path=get_spec().dataset.steps)
+
+    def output(self):
+        return b2luigi.LocalTarget(str(_task_dir("warm_cache", str(self.name), str(self.spec_hash)) / "stamp.json"))
+
+    def run(self):
+        from giant.data.setup_cache import sidecar_path
+        from giant.tools.warm_setup_cache import run_warm_setup_cache
+
+        spec = get_spec()
+        train = spec.train(str(self.name))
+        run_warm_setup_cache(
+            data=spec.dataset.steps,
+            config_path=Path(train.config) if train.config else None,
+        )
+        sidecar = Path(sidecar_path(spec.dataset.steps))
+        stamp = {
+            "sidecar": str(sidecar),
+            "mtime": sidecar.stat().st_mtime if sidecar.exists() else None,
+            "size": sidecar.stat().st_size if sidecar.exists() else None,
+        }
+        out = Path(self.output().path)
+        out.parent.mkdir(parents=True, exist_ok=True)
+        out.write_text(json.dumps(stamp, indent=2))
+
+
+class GeometryOracleTask(b2luigi.Task):
+    """Build the position -> (material, layer_id) oracle every rollout needs."""
+
+    spec_hash = b2luigi.Parameter()
+
+    @property
+    def htcondor_settings(self):
+        spec = get_spec()
+        return cpu_settings(spec.condor, request_memory_mb=32768, request_cpus=4, walltime_s=21600)
+
+    def requires(self):
+        yield DatasetTask(path=get_spec().dataset.steps)
+
+    def output(self):
+        return b2luigi.LocalTarget(
+            str(_result_dir("geometry", f"spec_hash={self.spec_hash}") / "oracle.pkl"),
+        )
+
+    def run(self):
+        from giant.tools.geometry_oracle import run_build_geometry_oracle
+
+        spec = get_spec()
+        g = spec.geometry
+        out = Path(self.output().path)
+        out.parent.mkdir(parents=True, exist_ok=True)
+        run_build_geometry_oracle(
+            data=Path(spec.dataset.steps),
+            out=out,
+            method=g.method,
+            k=g.k,
+            subsample=g.subsample,
+            escape_factor=g.escape_factor,
+            seed=g.seed,
+            depth_axis=g.depth_axis,
+            n_bins=g.n_bins,
+        )
+
+
+# ---------------------------------------------------------------------------
+# training
+# ---------------------------------------------------------------------------
+
+
+def _train_cfg(spec: WorkflowSpec, name: str, epochs: int) -> dict:
+    """The merged config one training job runs, resolved exactly as `giant train` does."""
+    from giant import config as gconfig
+
+    train = spec.train(name)
+    flags = {**train.overrides, "epochs": epochs}
+    overrides = gconfig.overrides_from_flags(flags)
+    cfg = gconfig.merge_cli_overrides(
+        gconfig.DEFAULT_CONFIG,
+        Path(train.config) if train.config else None,
+        overrides,
+    )
+    gconfig.validate_config(cfg, resume=True)
+    return cfg
+
+
+class TrainEpochTask(b2luigi.Task):
+    """Epochs up to ``milestone`` of one training, resuming the previous job.
+
+    Target is ``last.pt``. ``best.pt`` is written by the loop *only when that
+    epoch improved*, and ``best_val_loss`` travels inside the checkpoint, so
+    the global best comparison stays correct across jobs: "``best.pt`` exists
+    in milestone dir *k*" means exactly "one of that job's epochs was the best
+    so far".
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+    milestone = b2luigi.IntParameter()
+
+    @property
+    def htcondor_settings(self):
+        spec = get_spec()
+        train = spec.train(str(self.name))
+        return gpu_settings(
+            spec.condor,
+            request_gpus=train.request_gpus,
+            gpu_type=train.gpu_type,
+            gpu_memory_mb=train.gpu_memory_mb,
+            request_memory_mb=train.request_memory_mb,
+            request_cpus=train.request_cpus,
+            walltime_s=train.walltime_s,
+        )
+
+    @property
+    def _dir(self) -> Path:
+        return _task_dir("train_epoch", str(self.name), str(self.spec_hash)) / f"epochs={int(self.milestone)}"
+
+    def _previous_milestone(self) -> int | None:
+        spec = get_spec()
+        milestones = epoch_milestones(spec.train(str(self.name)))
+        index = milestones.index(int(self.milestone))
+        return milestones[index - 1] if index > 0 else None
+
+    def requires(self):
+        previous = self._previous_milestone()
+        if previous is None:
+            yield WarmCacheTask(name=self.name, spec_hash=get_spec().warm_cache_hash(str(self.name)))
+        else:
+            yield TrainEpochTask(name=self.name, spec_hash=self.spec_hash, milestone=previous)
+
+    def output(self):
+        return b2luigi.LocalTarget(str(self._dir / "last.pt"))
+
+    def run(self):
+        import torch
+
+        from giant import config as gconfig
+        from giant.pipeline import run_train_job
+
+        spec = get_spec()
+        train = spec.train(str(self.name))
+        cfg = _train_cfg(spec, str(self.name), int(self.milestone))
+
+        previous = self._previous_milestone()
+        resume = None
+        if previous is not None:
+            resume = _task_dir("train_epoch", str(self.name), str(self.spec_hash)) / f"epochs={previous}" / "last.pt"
+
+        device = torch.device(train.device) if train.device else gconfig.auto_device()
+        out_dir = self._dir
+        out_dir.mkdir(parents=True, exist_ok=True)
+        run_train_job(
+            data=Path(spec.dataset.steps),
+            cfg=cfg,
+            out_dir=out_dir,
+            device=device,
+            shuffle_buffer=train.shuffle_buffer,
+            num_workers=train.num_workers,
+            resume=resume,
+            cache_setup=True,
+        )
+
+
+class TrainTask(b2luigi.Task):
+    """Publish one training's canonical outputs, hiding the epoch fan-out.
+
+    Everything downstream (``RolloutTask``, humans, ``giant analyze metrics``)
+    points here and never has to know which milestone directory happened to
+    hold the best checkpoint.
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+    batch_system = "local"
+
+    @property
+    def _milestones(self) -> list[int]:
+        return epoch_milestones(get_spec().train(str(self.name)))
+
+    def requires(self):
+        yield TrainEpochTask(name=self.name, spec_hash=self.spec_hash, milestone=self._milestones[-1])
+
+    @property
+    def _dir(self) -> Path:
+        return _task_dir("train", str(self.name), str(self.spec_hash))
+
+    def output(self):
+        d = self._dir
+        return {
+            "best.pt": b2luigi.LocalTarget(str(d / "best.pt")),
+            "last.pt": b2luigi.LocalTarget(str(d / "last.pt")),
+            "metrics.csv": b2luigi.LocalTarget(str(d / "metrics.csv")),
+        }
+
+    def run(self):
+        epoch_base = _task_dir("train_epoch", str(self.name), str(self.spec_hash))
+        milestone_dirs = [epoch_base / f"epochs={m}" for m in self._milestones]
+
+        best_dirs = [d for d in milestone_dirs if (d / "best.pt").exists()]
+        if not best_dirs:
+            raise FileNotFoundError(
+                f"no best.pt in any milestone directory under {epoch_base} — "
+                "did every epoch job run with a validation split?"
+            )
+        out = self._dir
+        out.mkdir(parents=True, exist_ok=True)
+        shutil.copy2(best_dirs[-1] / "best.pt", out / "best.pt")
+        shutil.copy2(milestone_dirs[-1] / "last.pt", out / "last.pt")
+        for extra in ("config.toml", "run_meta.json"):
+            src = milestone_dirs[-1] / extra
+            if src.exists():
+                shutil.copy2(src, out / extra)
+
+        # One metrics.csv for the whole run: the first job's header, then
+        # every job's rows in epoch order, so `giant analyze metrics` sees a
+        # single continuous training curve.
+        lines: list[str] = []
+        header: str | None = None
+        for d in milestone_dirs:
+            csv = d / "metrics.csv"
+            if not csv.exists():
+                continue
+            rows = csv.read_text().splitlines()
+            if not rows:
+                continue
+            if header is None:
+                header = rows[0]
+            lines.extend(rows[1:])
+        (out / "metrics.csv").write_text("\n".join([header or ""] + lines) + "\n")
+
+
+# ---------------------------------------------------------------------------
+# rollout
+# ---------------------------------------------------------------------------
+
+
+class RolloutTask(b2luigi.Task):
+    """Roll one trained checkpoint forward into full showers.
+
+    ``giant rollout`` has no library-level entry point, so this shells out to
+    the CLI — with an explicit ``--out``, which puts the YAML sidecar at the
+    deterministic ``rollout.yaml`` next to the parquet (see
+    ``giant/cli.py:_write_prediction_ref``).
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+
+    @property
+    def htcondor_settings(self):
+        spec = get_spec()
+        ro = spec.rollout(str(self.name))
+        return gpu_settings(
+            spec.condor,
+            request_gpus=ro.request_gpus,
+            gpu_type=ro.gpu_type,
+            gpu_memory_mb=ro.gpu_memory_mb,
+            request_memory_mb=ro.request_memory_mb,
+            request_cpus=ro.request_cpus,
+            walltime_s=ro.walltime_s,
+        )
+
+    @property
+    def _dir(self) -> Path:
+        return _task_dir("rollout", str(self.name), str(self.spec_hash))
+
+    def requires(self):
+        spec = get_spec()
+        ro = spec.rollout(str(self.name))
+        yield TrainTask(name=ro.train, spec_hash=spec.train_hash(ro.train))
+        yield GeometryOracleTask(spec_hash=spec.geometry_hash())
+        yield DatasetTask(path=spec.dataset.reference)
+
+    def output(self):
+        d = self._dir
+        return {
+            "rollout.parquet": b2luigi.LocalTarget(str(d / "rollout.parquet")),
+            "rollout.yaml": b2luigi.LocalTarget(str(d / "rollout.yaml")),
+        }
+
+    def run(self):
+        spec = get_spec()
+        ro = spec.rollout(str(self.name))
+        out = self._dir / "rollout.parquet"
+        out.parent.mkdir(parents=True, exist_ok=True)
+
+        checkpoint = _task_dir("train", ro.train, spec.train_hash(ro.train)) / "best.pt"
+        oracle = _result_dir("geometry", f"spec_hash={spec.geometry_hash()}") / "oracle.pkl"
+
+        cmd = [
+            *_giant_cmd(),
+            "rollout",
+            spec.dataset.reference,
+            "--checkpoint",
+            str(checkpoint),
+            "--geometry",
+            str(oracle),
+            "--out",
+            str(out),
+            "--energy-cutoff",
+            str(ro.energy_cutoff),
+            "--max-steps",
+            str(ro.max_steps),
+            "--steps",
+            str(ro.steps),
+            "--batch-size",
+            str(ro.batch_size),
+            "--weights",
+            ro.weights,
+        ]
+        for flag, value in (
+            ("--n-events", ro.n_events),
+            ("--max-tracks-per-event", ro.max_tracks_per_event),
+            ("--escape-threshold", ro.escape_threshold),
+            ("--seed", ro.seed),
+            ("--device", ro.device),
+        ):
+            if value is not None:
+                cmd += [flag, str(value)]
+        subprocess.run(cmd, check=True)
+
+
+# ---------------------------------------------------------------------------
+# analysis
+# ---------------------------------------------------------------------------
+
+
+class AnalysisPrepTask(b2luigi.Task):
+    """Resolve the shared bin edges/group sets once, for every compute job.
+
+    Cheap and streaming, so it runs locally: everything after it needs
+    ``shared.json``/``run_meta.json`` to already exist.
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+    batch_system = "local"
+
+    def requires(self):
+        spec = get_spec()
+        for rollout_name in spec.analysis(str(self.name)).rollouts:
+            yield RolloutTask(name=rollout_name, spec_hash=spec.rollout_hash(rollout_name))
+
+    @property
+    def _dir(self) -> Path:
+        return analysis_dir(get_spec(), str(self.name))
+
+    def output(self):
+        d = self._dir
+        return {
+            "shared.json": b2luigi.LocalTarget(str(d / "shared.json")),
+            "run_meta.json": b2luigi.LocalTarget(str(d / "run_meta.json")),
+        }
+
+    def run(self):
+        from giant.analysis import prep
+
+        spec = get_spec()
+        an = spec.analysis(str(self.name))
+        yamls = [_task_dir("rollout", r, spec.rollout_hash(r)) / "rollout.yaml" for r in an.rollouts]
+        prep(
+            yamls,
+            run_dir=self._dir,
+            n_chunks=an.chunks,
+            labels=list(an.rollouts),
+            n_energy_bins=an.energy_bins,
+            n_marginal_bins=an.bins,
+            top_k_pdg=an.top_pdg,
+        )
+
+
+class AnalysisComputeTask(b2luigi.Task):
+    """One (plot, chunk) streaming reduction — the replaced ``jobs.txt`` row.
+
+    The output path is the on-disk contract ``compute-one``/``merge_one``
+    already share (``reduced_partial/<id>__<chunk>.json``), declared
+    explicitly rather than through b2luigi's own output naming so that
+    contract is untouched.
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+    plot_id = b2luigi.Parameter()
+    chunk = b2luigi.IntParameter()
+
+    @property
+    def htcondor_settings(self):
+        # A property, so it is evaluated at submit time — i.e. after prep has
+        # written run_meta.json, whose row counts size the walltime request.
+        from giant.analysis import RunMeta
+        from giant.analysis.runtime_estimate import estimate_runtime_s
+
+        spec = get_spec()
+        an = spec.analysis(str(self.name))
+        walltime = None
+        meta_path = analysis_dir(spec, str(self.name)) / "run_meta.json"
+        if meta_path.exists():
+            from giant.analysis.catalog import get_spec as get_plot_spec
+
+            meta = RunMeta.load(meta_path)
+            chunkable = get_plot_spec(str(self.plot_id)).chunkable
+            n_rows = meta.rows_per_chunk[int(self.chunk)] if chunkable and meta.rows_per_chunk else meta.total_rows
+            walltime = estimate_runtime_s(str(self.plot_id), n_rows)
+        return cpu_settings(
+            spec.condor,
+            request_memory_mb=an.request_memory_mb,
+            request_cpus=an.request_cpus,
+            walltime_s=walltime,
+        )
+
+    def requires(self):
+        yield AnalysisPrepTask(name=self.name, spec_hash=self.spec_hash)
+
+    def output(self):
+        run_dir = analysis_dir(get_spec(), str(self.name))
+        return b2luigi.LocalTarget(str(run_dir / "reduced_partial" / f"{self.plot_id}__{int(self.chunk)}.json"))
+
+    def run(self):
+        from giant.analysis import compute_one
+
+        compute_one(str(self.plot_id), analysis_dir(get_spec(), str(self.name)), chunk_index=int(self.chunk))
+
+
+class AnalysisRenderTask(b2luigi.Task):
+    """Merge every plot's chunk partials, then render the PDFs + gallery.
+
+    Always local — this is the only step that imports plotstyle/LaTeX, which
+    the compute worker images don't have.
+    """
+
+    name = b2luigi.Parameter()
+    spec_hash = b2luigi.Parameter()
+    batch_system = "local"
+
+    def requires(self):
+        spec = get_spec()
+        for plot_id, chunk in analysis_jobs(spec, str(self.name)):
+            yield AnalysisComputeTask(
+                name=self.name,
+                spec_hash=self.spec_hash,
+                plot_id=plot_id,
+                chunk=chunk,
+            )
+
+    def output(self):
+        run_dir = analysis_dir(get_spec(), str(self.name))
+        return b2luigi.LocalTarget(str(run_dir / "plots" / "metadata.yaml"))
+
+    def run(self):
+        # render_run joins every plot's chunk partials (merge_all) before
+        # rendering, so this one call is the whole merge+render step.
+        from giant.analysis.render import render_run
+
+        spec = get_spec()
+        render_run(analysis_dir(spec, str(self.name)), run_gallery=spec.analysis(str(self.name)).gallery)
+
+
+class WorkflowTask(b2luigi.WrapperTask):
+    """The whole pipeline: every analysis in the spec, rendered."""
+
+    workflow_name = b2luigi.Parameter()
+
+    def requires(self):
+        spec = get_spec()
+        if not spec.analyses:
+            # A spec with no [[analysis]] still has work to do — fall back to
+            # the deepest tasks it does define.
+            for ro in spec.rollouts:
+                yield RolloutTask(name=ro.name, spec_hash=spec.rollout_hash(ro.name))
+            if not spec.rollouts:
+                for tr in spec.trains:
+                    yield TrainTask(name=tr.name, spec_hash=spec.train_hash(tr.name))
+            return
+        for an in spec.analyses:
+            yield AnalysisRenderTask(name=an.name, spec_hash=spec.analysis_hash(an.name))

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "giant"
-version = "0.3.15"
+version = "0.3.10"
 description = "Geant4 step-function surrogate via conditional flow matching"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -28,7 +28,7 @@ dev = [
     "ty>=0.0.50,<0.1",
     "bump-my-version>=1.2,<2",
     "git-cliff>=2,<3",
-    "giant[convert,analysis,geometry,wandb]",
+    "giant[convert,analysis,geometry,wandb,workflow]",
 ]
 geometry = [
     "scikit-learn>=1.4,<2",
@@ -49,6 +49,11 @@ analysis = [
     # `giant analyze render` step imports it; compute workers never do.
     "plotstyle>=1.0.0",
 ]
+# b2luigi pulls luigi + tenacity; the only sanctioned way to chain a
+# multi-step pipeline (see giant/workflow/).
+workflow = [
+    "b2luigi>=1.0,<2",
+]
 
 [project.scripts]
 giant = "giant.cli:app"
@@ -103,3 +108,12 @@ explicit = true
 name = "larsbogner"
 url = "https://git.larsbogner.de/api/packages/lars/pypi/simple/"
 explicit = true
+
+# luigi builds task constructors from class-level Parameter descriptors, so a
+# static checker sees no keyword parameters at all on `Task(name=..., ...)`.
+# The workflow code is written against that API; nothing else in the repo is.
+[[tool.ty.overrides]]
+include = ["giant/workflow/**", "tests/test_workflow_tasks.py"]
+
+[tool.ty.overrides.rules]
+unknown-argument = "ignore"

tests/test_analysis_reduce.py

@@ -13,7 +13,6 @@ from giant.analysis.sources import (
     open_side,
     physical_steps,
     secondaries,
-    secondaries_by_step,
 )
 from giant.data.loader import EVENT_ID_FILE_STRIDE
 
@@ -160,17 +159,18 @@ def test_secondaries_rollout_vs_reference_align():
     assert t["pdg"].to_list() == [22, 22]
 
 
-def test_secondaries_by_step_keys_each_secondary_to_its_emitting_step():
-    r = secondaries_by_step(_rollout_frame(), Side.rollout).collect()
-    assert r["pdg"].to_list() == [22]
-    # the rollout key is (event_id, parent_id, birth position) — the parent
-    # step's post_pos, copied verbatim onto the child's birth row.
-    assert r["step_key"][0] == {"event_id": 1, "parent_id": 0, "pre_x": 0.0, "pre_y": 0.0, "pre_z": 1.0}
+def test_sec_count_by_event_zero_fills_events_with_no_secondaries():
+    r_phys = physical_steps(_rollout_frame(), Side.rollout)
+    r_sec = secondaries(_rollout_frame(), Side.rollout)
+    ev, n = R.sec_count_by_event(r_phys, r_sec)
+    # event 1 has one secondary track; event 2 has none and must still appear (as 0),
+    # not silently drop out of a plain group_by on the secondaries frame alone.
+    assert dict(zip(ev.tolist(), n.tolist())) == {1: 1, 2: 0}
 
-    t = secondaries_by_step(_reference_frame(), Side.reference).collect()
-    assert t["pdg"].to_list() == [22, 22]
-    # one row per emitting step; the empty-list step drops out entirely
-    assert [k["_row"] for k in t["step_key"]] == [0, 2]
+    t_all = _reference_frame()
+    t_sec = secondaries(t_all, Side.reference)
+    ev, n = R.sec_count_by_event(t_all, t_sec)
+    assert dict(zip(ev.tolist(), n.tolist())) == {1: 1, 2: 1}
 
 
 def test_leakage_fraction():

tests/test_analysis_run.py

@@ -1,8 +1,7 @@
-"""Tests for the rollout-YAML(s) → run-directory flow, compute, and submit."""
+"""Tests for the rollout-YAML(s) → run-directory flow, compute, and merge."""
 
 from __future__ import annotations
 
-import sys
 from pathlib import Path
 
 import pyarrow.parquet as pq
@@ -11,8 +10,6 @@ import yaml
 
 from giant.analysis import (
     RunMeta,
-    SubmitConfig,
-    catalog_ids,
     compute_one,
     compute_reduced,
     derive_run_dir,
@@ -20,10 +17,8 @@ from giant.analysis import (
     load_rollout_yamls,
     merge_one,
     prep,
-    write_submit,
 )
-from giant.analysis.catalog import get_spec
-from giant.analysis.condor import Context
+from giant.analysis.run import Context
 from giant.analysis.reduced import Partial, Reduced
 from giant.constants import PREDICT_COORD_METADATA_KEY, ROLLOUT_COORD_VALUE
 from tests.test_analysis_reduce import _reference_frame, _rollout_frame
@@ -86,14 +81,6 @@ def _write_two_inputs(tmp_path: Path) -> tuple[Path, Path]:
     return paths[0], paths[1]
 
 
-def _fake_venv(repo_dir: Path) -> None:
-    """Stand in for a `uv sync`'d venv: write_submit checks `.venv/bin/giant` exists."""
-    giant = repo_dir / ".venv" / "bin" / "giant"
-    giant.parent.mkdir(parents=True, exist_ok=True)
-    giant.write_text("#!/bin/bash\n")
-    giant.chmod(0o755)
-
-
 def _prep(rollout_yamls, run_dir: str | Path | None = None, chunks: int = 1, labels=None) -> Path:
     """``prep`` with small test-sized context bins/sampling."""
     return prep(
@@ -313,73 +300,6 @@ def test_compute_reduced_rejects_out_of_range_chunk(tmp_path: Path):
         compute_one("marginal_edep", run_dir, chunk_index=1)
 
 
-def test_write_submit_description(tmp_path: Path):
-    run_dir = _prep([_write_inputs(tmp_path)])
-    _fake_venv(tmp_path)
-    cfg = SubmitConfig(run_dir=run_dir, accounting_group="cms", repo_dir=tmp_path)
-    txt = write_submit(cfg).read_text()
-    assert "universe = docker" in txt
-    assert "docker_image = cverstege/alma9-gridjob" in txt
-    assert "requirements = TARGET.ProvidesETPResources" in txt
-    assert "accounting_group = cms" in txt
-    assert "+RequestWalltime = $(walltime)" in txt
-    assert "queue plotid,chunk,walltime from" in txt
-    jobs = [line.split(",") for line in (run_dir / "jobs.txt").read_text().split()]
-    assert [i for i, _, _ in jobs] == catalog_ids()
-    assert all(k == "0" for _, k, _ in jobs)  # n_chunks=1 default
-    assert all(int(w) > 0 for _, _, w in jobs)
-    wrapper = run_dir / "run_compute.sh"
-    assert wrapper.exists() and (wrapper.stat().st_mode & 0o111)
-    body = wrapper.read_text()
-    assert "giant analyze compute-one --id" in body
-    assert "--chunk" in body and "--run-dir" in body
-
-
-def test_write_submit_requires_synced_venv(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
-    run_dir = _prep([_write_inputs(tmp_path)])
-    cfg = SubmitConfig(run_dir=run_dir, accounting_group="cms", repo_dir=tmp_path)
-    # No `giant` next to the (fake) active interpreter, so this falls through
-    # to repo_dir/.venv/bin/giant, which _write_inputs/_prep also didn't create.
-    monkeypatch.setattr(sys, "executable", str(tmp_path / "not-a-venv" / "bin" / "python"))
-    with pytest.raises(FileNotFoundError, match="uv sync"):
-        write_submit(cfg)
-
-
-def test_write_submit_remote_flag(tmp_path: Path):
-    run_dir = _prep([_write_inputs(tmp_path)])
-    _fake_venv(tmp_path)
-    cfg = SubmitConfig(run_dir=run_dir, accounting_group="cms", repo_dir=tmp_path, remote=True)
-    txt = write_submit(cfg).read_text()
-    assert "+RemoteJob = True" in txt
-    assert "ProvidesETPResources" not in txt
-
-
-def test_write_submit_chunks_respect_chunkable(tmp_path: Path):
-    assert get_spec("router_gating").chunkable is False
-    run_dir = _prep([_write_inputs(tmp_path)], chunks=4)
-    _fake_venv(tmp_path)
-    cfg = SubmitConfig(run_dir=run_dir, accounting_group="cms", repo_dir=tmp_path, n_chunks=4)
-    write_submit(cfg)
-    jobs = [line.split(",") for line in (run_dir / "jobs.txt").read_text().split()]
-    counts: dict[str, int] = {}
-    for spec_id, _, _ in jobs:
-        counts[spec_id] = counts.get(spec_id, 0) + 1
-    assert counts["marginal_edep"] == 4
-    assert counts["router_gating"] == 1  # chunkable=False, ignores n_chunks
-
-
-def test_write_submit_rejects_n_chunks_mismatch_with_run_meta(tmp_path: Path):
-    """cfg.n_chunks must match the n_chunks the run_dir was actually prepped
-    with — RunMeta.rows_per_chunk is sized to the prepped value, so a
-    mismatch would otherwise surface as a confusing IndexError deep inside
-    _job_walltimes instead of a clear error here."""
-    run_dir = _prep([_write_inputs(tmp_path)], chunks=2)
-    _fake_venv(tmp_path)
-    cfg = SubmitConfig(run_dir=run_dir, accounting_group="cms", repo_dir=tmp_path, n_chunks=4)
-    with pytest.raises(ValueError, match="n_chunks"):
-        write_submit(cfg)
-
-
 def test_estimate_runtime_s_scales_with_rows_and_margin():
     from giant.analysis import RUNTIME_SAFETY_MARGIN, estimate_runtime_s
     from giant.analysis.runtime_estimate import _FIXED_OVERHEAD_S
@@ -389,18 +309,3 @@ def test_estimate_runtime_s_scales_with_rows_and_margin():
     large = estimate_runtime_s("marginal_edep", 100_000_000)
     assert small >= (1 + RUNTIME_SAFETY_MARGIN) * _FIXED_OVERHEAD_S
     assert large > small  # bigger chunk -> longer estimate
-
-
-def test_write_submit_walltime_grows_with_chunk_rows(tmp_path: Path):
-    """A chunked run's later job walltimes track that chunk's row count."""
-    from giant.analysis.runtime_estimate import estimate_runtime_s
-
-    run_dir = _prep([_write_inputs(tmp_path)], chunks=2)
-    meta = RunMeta.load(run_dir / "run_meta.json")
-    _fake_venv(tmp_path)
-    cfg = SubmitConfig(run_dir=run_dir, accounting_group="cms", repo_dir=tmp_path, n_chunks=2)
-    write_submit(cfg)
-    jobs = {(i, int(k)): int(w) for i, k, w in (line.split(",") for line in (run_dir / "jobs.txt").read_text().split())}
-    for chunk in range(2):
-        expected = estimate_runtime_s("marginal_edep", meta.rows_per_chunk[chunk])
-        assert jobs[("marginal_edep", chunk)] == expected

tests/test_catalog.py

@@ -10,10 +10,10 @@ from giant.analysis.catalog import (
     Bundle,
     PlotSpec,
     _containment_depths,
+    _integer_confusion,
     _ks_statistic,
 )
 from giant.analysis.context import Context, build_context
-from giant.analysis.grouping import pdg_label
 from giant.analysis.sources import RolloutSpec
 from tests.test_analysis_reduce import _reference_frame, _rollout_frame
 
@@ -160,9 +160,8 @@ def _validate_payload(r, names: list[str]) -> None:
 # data-dependent edges (event_total_edep), concat-then-mean/std (shower_
 # longitudinal), concat-then-max-edge (leakage_fraction), pdg-keyed sum with a
 # ratio (species_edep_share), a chunkable=False passthrough (router_gating),
-# sum-mergeable-with-a-zero-fill-denominator (sec_count_per_step{,_by_species}),
-# nested sum-merge into a scorecard (marginal_distance_summary), and
-# concat-then-per-event-derived-quantity
+# nested sum-merge into a scorecard (marginal_distance_summary), concat-then-
+# event-id-join (n_sec_confusion), and concat-then-per-event-derived-quantity
 # (shower_containment_depth_90, reusing the profile matrix's own merge shape).
 _CHUNK_EQUIVALENCE_IDS = [
     "marginal_edep",
@@ -171,10 +170,9 @@ _CHUNK_EQUIVALENCE_IDS = [
     "shower_longitudinal",
     "leakage_fraction",
     "sec_count_per_species",
-    "sec_count_per_step",
-    "sec_count_per_step_by_species",
     "router_gating",
     "marginal_distance_summary",
+    "n_sec_confusion",
     "shower_containment_depth_90",
 ]
 
@@ -219,7 +217,7 @@ def test_chunked_matches_unchunked(two_ctx: Context, spec_id: str):
 
 
 # ---------------------------------------------------------------------------
-# new (gitea #76) reductions: KS distance and containment depth
+# new (gitea #76) reductions: KS distance, confusion matrix, containment depth
 # ---------------------------------------------------------------------------
 
 
@@ -230,6 +228,28 @@ def test_ks_statistic():
     assert _ks_statistic([10, 0], [0, 0]) == 1.0  # one side empty, other isn't -> maximal mismatch
 
 
+def test_integer_confusion_matches_event_pairing():
+    # true (reference) n_sec = [1, 1]; predicted (rollout) n_sec = [1, 0]
+    labels, mat = _integer_confusion(np.array([1, 1]), np.array([1, 0]))
+    assert labels == ["0", "1+"]
+    assert mat.tolist() == [[0, 0], [1, 1]]  # row=true, col=pred
+
+
+def test_integer_confusion_caps_pathological_outliers():
+    labels, mat = _integer_confusion(np.array([0, 500]), np.array([0, 0]), max_bins=5)
+    assert labels[-1] == "4+"
+    assert mat.shape == (5, 5)
+    assert mat.sum() == 2
+
+
+def test_integer_confusion_explicit_cap_overrides_local_range():
+    # Even though this pair's own max is 1, an explicit shared cap forces a
+    # wider (and so cross-rollout-consistent) label set.
+    labels, mat = _integer_confusion(np.array([1, 1]), np.array([0, 1]), cap=3)
+    assert labels == ["0", "1", "2", "3+"]
+    assert mat.shape == (4, 4)
+
+
 def test_containment_depths_simple_ramp():
     # one event, edep concentrated in the first bin -> 90%/95% containment
     # depth is the first bin's right edge; a zero-energy event is dropped.
@@ -239,25 +259,17 @@ def test_containment_depths_simple_ramp():
     assert depths.tolist() == [1.0]
 
 
-def test_sec_count_per_step_counts_empty_steps(bundle):
-    spec = get_spec("sec_count_per_step")
+def test_n_sec_confusion_spec(bundle):
+    spec = get_spec("n_sec_confusion")
     r = spec.finalize([spec.compute_partial(bundle)], bundle.ctx)
-    # reference: 3 steps, two of which emit exactly one secondary
-    assert r.payload["reference"][:2] == [1, 2]
-    # rollout: 4 physical steps, one of which emits a single secondary
-    assert r.payload["series"]["rollout"][:2] == [3, 1]
-    assert sum(r.payload["reference"]) == 3
+    assert r.payload["row_labels"] == r.payload["col_labels"] == ["0", "1+"]
+    assert r.payload["series"]["rollout"] == [[0, 0], [1, 1]]
 
 
-def test_sec_count_per_step_by_species_zero_row_is_per_species(bundle):
-    spec = get_spec("sec_count_per_step_by_species")
-    r = spec.finalize([spec.compute_partial(bundle)], bundle.ctx)
-    cols = r.payload["col_labels"]
-    ref = r.payload["reference"]
-    g = cols.index(pdg_label(22))
-    # two reference steps emit one photon each; the third emits none
-    assert [row[g] for row in ref][:2] == [1, 2]
-    # every other species column is "no such secondary" on all 3 steps
-    for j, _ in enumerate(cols):
-        if j != g:
-            assert ref[0][j] == 3 and sum(row[j] for row in ref[1:]) == 0
+def test_n_sec_confusion_shares_one_cap_across_rollouts(two_bundle):
+    spec = get_spec("n_sec_confusion")
+    r = spec.finalize([spec.compute_partial(two_bundle)], two_bundle.ctx)
+    assert list(r.payload["series"]) == ["flow", "wgan"]
+    # both rollouts share the same fixture data here, so their matrices (and
+    # the shared label set) must be identical.
+    assert r.payload["series"]["flow"] == r.payload["series"]["wgan"]

tests/test_checkpoint_io.py

@@ -14,7 +14,6 @@ from giant import config as gconfig
 from giant.checkpoint_io import (
     CheckpointCompatibilityError,
     InferenceContext,
-    apply_config_overrides,
     conditioning_axes,
     load_for_inference,
     stage_cfg,
@@ -279,122 +278,3 @@ def test_stage_cfg_new_shape_returns_subdict():
 def test_stage_cfg_v02_flat_shape_returns_empty_dict():
     model_cfg = {"hidden_dim": 32, "n_blocks": 4}
     assert stage_cfg(model_cfg, "stage2") == {}
-
-
-# ---------------------------------------------------------------------------
-# config_overrides (gitea #87)
-# ---------------------------------------------------------------------------
-
-
-def _router_model_cfg() -> dict:
-    cfg = _model_cfg()
-    cfg["stage1_model"]["router"] = {"enabled": True, "type": "energy", "n_experts": 2}
-    return cfg
-
-
-def test_config_override_n_sec_sampling_changes_stage2_attribute(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    ctx = load_for_inference(
-        checkpoint,
-        torch.device("cpu"),
-        "predict",
-        config_overrides={"stage2_model.n_sec.sampling": "sample"},
-    )
-    assert ctx.stage2 is not None
-    assert ctx.stage2.n_sec_sampling == "sample"
-    assert ctx.config_overrides == {"stage2_model.n_sec.sampling": "sample"}
-
-
-def test_config_override_ddpm_n_steps_changes_context_fields(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    ctx = load_for_inference(
-        checkpoint,
-        torch.device("cpu"),
-        "predict",
-        config_overrides={"stage1_model.ddpm.n_steps": 42, "stage2_model.ddpm.n_steps": 7},
-    )
-    assert ctx.stage1_ddpm_steps == 42
-    assert ctx.stage2_ddpm_steps == 7
-
-
-def test_config_override_other_policy_changes_context_field(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    ctx = load_for_inference(
-        checkpoint,
-        torch.device("cpu"),
-        "predict",
-        config_overrides={"stage2_model.particle_type.other_policy": "modal"},
-    )
-    assert ctx.other_policy == "modal"
-
-
-def test_config_override_router_temperature_changes_router_attribute(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path, model_cfg=_router_model_cfg())
-    ctx = load_for_inference(
-        checkpoint,
-        torch.device("cpu"),
-        "predict",
-        config_overrides={"stage1_model.router.temperature": 1.5},
-    )
-    assert ctx.stage1 is not None
-    assert ctx.stage1.trunk.router.temperature == pytest.approx(1.5)
-
-
-def test_config_override_no_overrides_defaults_to_empty_dict(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    ctx = load_for_inference(checkpoint, torch.device("cpu"), "predict")
-    assert ctx.config_overrides == {}
-
-
-def test_config_override_unknown_path_raises(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    with pytest.raises(CheckpointCompatibilityError, match="not an inference-safe override"):
-        load_for_inference(
-            checkpoint,
-            torch.device("cpu"),
-            "predict",
-            config_overrides={"stage2_model.n_sec.typo": "sample"},
-        )
-
-
-def test_config_override_shape_bearing_key_raises_up_front(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    with pytest.raises(CheckpointCompatibilityError, match="not an inference-safe override"):
-        load_for_inference(
-            checkpoint,
-            torch.device("cpu"),
-            "predict",
-            config_overrides={"stage1_model.hidden_dim": 999},
-        )
-
-
-def test_config_override_bad_value_raises(tmp_path):
-    checkpoint = _write_checkpoint(tmp_path)
-    with pytest.raises(CheckpointCompatibilityError, match="must be one of"):
-        load_for_inference(
-            checkpoint,
-            torch.device("cpu"),
-            "predict",
-            config_overrides={"stage2_model.n_sec.sampling": "maybe"},
-        )
-
-
-def test_apply_config_overrides_no_overrides_returns_same_object():
-    cfg = _model_cfg()
-    assert apply_config_overrides(cfg, None) is cfg
-    assert apply_config_overrides(cfg, {}) is cfg
-
-
-def test_apply_config_overrides_migrates_legacy_flat_model_config_first():
-    legacy_cfg = {
-        "pdg_vocab": len(PDG_MAP),
-        "mat_vocab": len(MAT_MAP),
-        "hidden_dim": 32,
-        "n_blocks": 4,
-        "emb_dim": 8,
-        "dropout": 0.1,
-        "k_max": 5,
-    }
-    merged = apply_config_overrides(legacy_cfg, {"stage1_model.ddpm.n_steps": 10})
-    assert merged["stage1_model"]["ddpm"]["n_steps"] == 10
-    assert merged["stage1_model"]["hidden_dim"] == 32

tests/test_cli_predict.py

@@ -172,43 +172,3 @@ def test_predict_exits_1_on_checkpoint_missing_model_config(tmp_path):
 
     assert result.exit_code == 1
     assert "checkpoint has no model_config" in result.output
-
-
-# ---------------------------------------------------------------------------
-# --set (gitea #87)
-# ---------------------------------------------------------------------------
-
-
-def test_predict_set_flag_without_equals_exits_1(tmp_path):
-    checkpoint = tmp_path / "missing.pt"
-
-    result = runner.invoke(
-        app,
-        ["predict", "dummy.parquet", "--checkpoint", str(checkpoint), "--set", "sampling"],
-    )
-
-    assert result.exit_code == 1
-    assert "must be 'dotted.path=value'" in result.output
-
-
-def test_predict_set_flag_disallowed_path_surfaces_compat_error(tmp_path):
-    checkpoint = tmp_path / "ckpt.pt"
-    torch.save(
-        {"model_config": {"stage1_model": {}, "stage2_model": {}}, "sec_decoder": {}, "normalizer": {"sec_phys": {}}},
-        checkpoint,
-    )
-
-    result = runner.invoke(
-        app,
-        [
-            "predict",
-            "dummy.parquet",
-            "--checkpoint",
-            str(checkpoint),
-            "--set",
-            "stage1_model.hidden_dim=999",
-        ],
-    )
-
-    assert result.exit_code == 1
-    assert "not an inference-safe override" in result.output

tests/test_cli_rollout.py

@@ -31,28 +31,3 @@ def test_rollout_exits_1_on_checkpoint_missing_model_config(tmp_path):
 
     assert result.exit_code == 1
     assert "checkpoint has no model_config" in result.output
-
-
-def test_rollout_set_flag_disallowed_path_surfaces_compat_error(tmp_path):
-    checkpoint = tmp_path / "ckpt.pt"
-    torch.save(
-        {"model_config": {"stage1_model": {}, "stage2_model": {}}, "sec_decoder": {}, "normalizer": {"sec_phys": {}}},
-        checkpoint,
-    )
-
-    result = runner.invoke(
-        app,
-        [
-            "rollout",
-            "dummy.parquet",
-            "--checkpoint",
-            str(checkpoint),
-            "--geometry",
-            "dummy_geometry.pkl",
-            "--set",
-            "stage2_model.n_sec.typo=sample",
-        ],
-    )
-
-    assert result.exit_code == 1
-    assert "not an inference-safe override" in result.output

tests/test_cli_train_overrides.py

@@ -20,7 +20,7 @@ def _invoke_and_capture_cfg(monkeypatch, tmp_path: Path, args: list[str]) -> dic
     def _fake_run_train_job(*, data, cfg, out_dir, **kwargs):
         captured["cfg"] = cfg
 
-    monkeypatch.setattr("giant.pipeline.run_train_job", _fake_run_train_job)
+    monkeypatch.setattr(cli, "run_train_job", _fake_run_train_job)
 
     result = runner.invoke(
         cli.app,
@@ -125,7 +125,7 @@ def test_stage2_init_from_and_freeze_flags_land_in_cfg_and_dont_touch_stage1(mon
 
 
 def test_batch_size_invalid_string_errors(monkeypatch, tmp_path):
-    monkeypatch.setattr("giant.pipeline.run_train_job", lambda *a, **kw: None)
+    monkeypatch.setattr(cli, "run_train_job", lambda *a, **kw: None)
     result = runner.invoke(
         cli.app,
         ["train", "dummy.parquet", "--out", str(tmp_path / "run"), "--batch-size", "not-a-number"],
@@ -140,7 +140,7 @@ def test_out_dir_resolution_prefers_explicit_out_over_resume(monkeypatch, tmp_pa
     def _fake_run_train_job(*, data, cfg, out_dir, **kwargs):
         captured["out_dir"] = out_dir
 
-    monkeypatch.setattr("giant.pipeline.run_train_job", _fake_run_train_job)
+    monkeypatch.setattr(cli, "run_train_job", _fake_run_train_job)
 
     resume_dir = tmp_path / "resumed_run"
     resume_dir.mkdir()
@@ -161,7 +161,7 @@ def test_out_dir_resolution_falls_back_to_resume_parent(monkeypatch, tmp_path):
     def _fake_run_train_job(*, data, cfg, out_dir, **kwargs):
         captured["out_dir"] = out_dir
 
-    monkeypatch.setattr("giant.pipeline.run_train_job", _fake_run_train_job)
+    monkeypatch.setattr(cli, "run_train_job", _fake_run_train_job)
 
     resume_dir = tmp_path / "resumed_run"
     resume_dir.mkdir()
@@ -178,7 +178,7 @@ def test_out_dir_resolution_defaults_when_neither_out_nor_resume_given(monkeypat
     def _fake_run_train_job(*, data, cfg, out_dir, **kwargs):
         captured["out_dir"] = out_dir
 
-    monkeypatch.setattr("giant.pipeline.run_train_job", _fake_run_train_job)
+    monkeypatch.setattr(cli, "run_train_job", _fake_run_train_job)
     monkeypatch.chdir(tmp_path)
 
     result = runner.invoke(cli.app, ["train", "dummy.parquet"])
@@ -192,7 +192,7 @@ def test_batch_size_auto_estimates_and_echoes(monkeypatch, tmp_path):
     def _fake_run_train_job(*, data, cfg, out_dir, num_workers, **kwargs):
         captured["batch_size"] = cfg["train"]["batch_size"]
 
-    monkeypatch.setattr("giant.pipeline.run_train_job", _fake_run_train_job)
+    monkeypatch.setattr(cli, "run_train_job", _fake_run_train_job)
     monkeypatch.setattr(cli.gconfig, "estimate_batch_size", lambda hidden_dim, n_blocks, device: 123)
 
     result = runner.invoke(

tests/test_config.py

@@ -171,40 +171,17 @@ def test_n_sec_config_owner_defaults_to_stage2():
 def test_n_sec_config_owner_round_trips():
     n_sec = gconfig.NSecConfig.from_dict({"mode": "head", "owner": "stage1"})
     assert n_sec.owner == "stage1"
-    assert n_sec.to_dict() == {"mode": "head", "lambda": 0.1, "owner": "stage1", "sampling": "greedy"}
+    assert n_sec.to_dict() == {"mode": "head", "lambda": 0.1, "owner": "stage1", "stop_sampling": "greedy"}
 
 
-def test_n_sec_config_sampling_defaults_to_greedy():
-    assert gconfig.NSecConfig().sampling == "greedy"
+def test_n_sec_config_stop_sampling_defaults_to_greedy():
+    assert gconfig.NSecConfig().stop_sampling == "greedy"
 
 
-def test_n_sec_config_sampling_round_trips():
-    n_sec = gconfig.NSecConfig.from_dict({"mode": "stop_token", "sampling": "sample"})
-    assert n_sec.sampling == "sample"
-    assert n_sec.to_dict()["sampling"] == "sample"
-
-
-def test_n_sec_config_stop_sampling_alias_still_honored():
-    """gitea #86: stop_sampling was renamed to sampling; old checkpoints'
-    model_config still carries the old key and must keep working."""
+def test_n_sec_config_stop_sampling_round_trips():
     n_sec = gconfig.NSecConfig.from_dict({"mode": "stop_token", "stop_sampling": "sample"})
-    assert n_sec.sampling == "sample"
-    assert "stop_sampling" not in n_sec.to_dict()
-
-
-def test_n_sec_config_sampling_key_wins_over_stop_sampling_alias():
-    n_sec = gconfig.NSecConfig.from_dict({"sampling": "sample", "stop_sampling": "greedy"})
-    assert n_sec.sampling == "sample"
-
-
-def test_migrate_config_renames_stop_sampling_key():
-    cfg = {
-        "meta": {"config_version": gconfig.CONFIG_VERSION},
-        "stage2_model": {"n_sec": {"stop_sampling": "sample"}},
-    }
-    migrated = gconfig.migrate_config(cfg)
-    assert gconfig._get_path(migrated, "stage2_model.n_sec.sampling") == "sample"
-    assert gconfig._get_path(migrated, "stage2_model.n_sec.stop_sampling") is None
+    assert n_sec.stop_sampling == "sample"
+    assert n_sec.to_dict()["stop_sampling"] == "sample"
 
 
 # ---------------------------------------------------------------------------
@@ -874,13 +851,13 @@ def test_validate_config_stop_token_rejected_for_stage1_owner():
         assert "stop_token" in str(e) and "owner" in str(e)
 
 
-def test_validate_config_bad_n_sec_sampling_rejected():
-    cfg = _cfg_with(**{"stage2_model.n_sec.sampling": "bogus"})
+def test_validate_config_bad_stop_sampling_rejected():
+    cfg = _cfg_with(**{"stage2_model.n_sec.stop_sampling": "bogus"})
     try:
         gconfig.validate_config(cfg)
         assert False, "expected ValueError"
     except ValueError as e:
-        assert "sampling" in str(e)
+        assert "stop_sampling" in str(e)
 
 
 def test_validate_config_default_precision_is_fp32():

tests/test_model_summary.py

@@ -12,8 +12,6 @@ from giant.cli import app
 from giant.materials import MATERIAL_PROPERTIES
 from giant.model.summary import _NOT_BUILD_TIME, _built_modules, _vocab_caveats, summarize_model
 
-INFERENCE_OVERRIDES = gconfig.INFERENCE_OVERRIDES
-
 runner = CliRunner()
 
 _PDG_VOCAB = 300
@@ -55,16 +53,6 @@ def test_not_build_time_allow_list_has_no_stale_entries():
     assert not stale, f"_NOT_BUILD_TIME entries no longer in DEFAULT_CONFIG: {sorted(stale)}"
 
 
-def test_inference_overrides_allow_list_has_no_stale_entries():
-    in_scope = set(gconfig.leaf_paths(gconfig.DEFAULT_CONFIG))
-    stale = set(INFERENCE_OVERRIDES) - in_scope
-    assert not stale, f"INFERENCE_OVERRIDES entries no longer in DEFAULT_CONFIG: {sorted(stale)}"
-
-
-def test_default_config_overridable_lists_every_allowlisted_path(default_summary):
-    assert set(default_summary.overridable) == set(INFERENCE_OVERRIDES)
-
-
 def test_router_disabled_by_default_so_its_fields_are_inert(default_summary):
     assert "stage1_model.router.n_experts" in default_summary.inert
     assert "stage1_model.router.temperature" in default_summary.inert
@@ -147,5 +135,3 @@ def test_cli_default_smoke():
     assert "parameters" in result.output
     assert "trunk" in result.output
     assert "inert under this config" in result.output
-    assert "inference-overridable without retraining" in result.output
-    assert "stage2_model.n_sec.sampling" in result.output

tests/test_render.py

@@ -206,21 +206,21 @@ def test_render_all_run_gallery_invokes_subprocess(tmp_path: Path, monkeypatch):
     assert kwargs == {"check": True}
 
 
-def test_render_run_glues_condor_run_meta_into_render_all(tmp_path: Path, monkeypatch):
-    from giant.analysis import condor as condor_mod
+def test_render_run_glues_run_meta_into_render_all(tmp_path: Path, monkeypatch):
+    from giant.analysis import run as run_mod
 
     run_dir = tmp_path / "run"
     (run_dir / "reduced").mkdir(parents=True)
 
     merge_calls = []
-    monkeypatch.setattr(condor_mod, "merge_all", lambda rd: merge_calls.append(Path(rd)))
-    meta = condor_mod.RunMeta(
+    monkeypatch.setattr(run_mod, "merge_all", lambda rd: merge_calls.append(Path(rd)))
+    meta = run_mod.RunMeta(
         rollouts=[{"name": "rollout", "path": "rollout.parquet", "plot_meta": {"checkpoint": "ckpt/best.pt"}}],
         reference="reference.parquet",
         run_dir=str(run_dir),
         title="my-run",
     )
-    monkeypatch.setattr(condor_mod.RunMeta, "load", classmethod(lambda cls, p: meta))
+    monkeypatch.setattr(run_mod.RunMeta, "load", classmethod(lambda cls, p: meta))
 
     Reduced("s", "species", "single_hist", "Single", "x", {"edges": [0, 1], "series": {"rollout": [1]}}).save(
         run_dir / "reduced" / "s.json"
@@ -304,15 +304,13 @@ def test_render_one_of_each_kind(tmp_path: Path):
             "hm1",
             "secondaries",
             "heatmap",
-            "Heatmap (single rollout)",
+            "Confusion (single rollout)",
             "predicted",
             {
                 "series": {"flow": [[1, 0], [0, 1]]},
-                "reference": [[2, 0], [0, 1]],
                 "row_labels": ["0", "1+"],
                 "col_labels": ["0", "1+"],
                 "cbar_label": "count",
-                "log_color": True,
             },
         ),
     ]

tests/test_sample.py

@@ -14,7 +14,6 @@ from giant.model.network import (
     stage2_trunk_sec_dim,
 )
 from giant.sample import (
-    resolve_n_sec,
     sample_flow,
     sample_secondaries,
     sample_secondaries_ar,
@@ -73,7 +72,6 @@ def _stage2_ar(
     mat: int = 2,
     k_max: int = 5,
     history: str = "markov",
-    n_sec_sampling: str = "greedy",
 ) -> Stage2Autoregressive:
     particle_cfg, material_cfg = _particle_material_cfg(_conditioning_for(target), emb_dim)
     return Stage2Autoregressive(
@@ -91,7 +89,6 @@ def _stage2_ar(
         history=history,
         attn_n_heads=2,
         attn_n_layers=1,
-        n_sec_sampling=n_sec_sampling,
     ).eval()
 
 
@@ -102,7 +99,7 @@ def _expected_type_dim(target: str, emb_dim: int) -> int:
 def _stage2_ar_stop_token(
     target: str,
     generator: str,
-    n_sec_sampling: str = "greedy",
+    stop_sampling: str = "greedy",
     emb_dim: int = 6,
     pdg: int = 3,
     mat: int = 2,
@@ -123,7 +120,7 @@ def _stage2_ar_stop_token(
         particle_type_cfg=ParticleTypeConfig(target=target),
         build_n_sec_head=False,
         build_stop_head=True,
-        n_sec_sampling=n_sec_sampling,
+        stop_sampling=stop_sampling,
     ).eval()
 
 
@@ -270,14 +267,14 @@ def test_sample_secondaries_ar_first_slot_has_no_history():
 # ── Stage2Autoregressive: n_sec.mode = "stop_token" ─────────────────────────
 
 
-@pytest.mark.parametrize("n_sec_sampling", ["greedy", "sample"])
-def test_sample_secondaries_ar_stop_token_forced_stop_gives_zero_secondaries(n_sec_sampling):
+@pytest.mark.parametrize("stop_sampling", ["greedy", "sample"])
+def test_sample_secondaries_ar_stop_token_forced_stop_gives_zero_secondaries(stop_sampling):
     """A stop_head pinned to a large positive logit fires at slot 0 for
     every row under both policies (greedy: sigmoid(logit) >= 0.5; sample:
     a Bernoulli draw at sigmoid(logit) ~= 1) — the loop should break before
     generating any token."""
     B, k_max = 4, 5
-    decoder = _stage2_ar_stop_token("physical", "flow", n_sec_sampling=n_sec_sampling, k_max=k_max)
+    decoder = _stage2_ar_stop_token("physical", "flow", stop_sampling=stop_sampling, k_max=k_max)
     _force_stop_head_logit(decoder, 50.0)
     cond_cont, cond_cat = _cond(B)
     stage1_out = torch.randn(B, X_DIM)
@@ -286,13 +283,13 @@ def test_sample_secondaries_ar_stop_token_forced_stop_gives_zero_secondaries(n_s
     assert not sec_valid.any()
 
 
-@pytest.mark.parametrize("n_sec_sampling", ["greedy", "sample"])
-def test_sample_secondaries_ar_stop_token_forced_never_stop_runs_to_k_max(n_sec_sampling):
+@pytest.mark.parametrize("stop_sampling", ["greedy", "sample"])
+def test_sample_secondaries_ar_stop_token_forced_never_stop_runs_to_k_max(stop_sampling):
     """A stop_head pinned to a large negative logit never fires under either
     policy, so every row is capped at k_max (the safety cap, not a modeling
     ceiling)."""
     B, k_max = 4, 5
-    decoder = _stage2_ar_stop_token("physical", "flow", n_sec_sampling=n_sec_sampling, k_max=k_max)
+    decoder = _stage2_ar_stop_token("physical", "flow", stop_sampling=stop_sampling, k_max=k_max)
     _force_stop_head_logit(decoder, -50.0)
     cond_cont, cond_cat = _cond(B)
     stage1_out = torch.randn(B, X_DIM)
@@ -339,60 +336,3 @@ def test_sample_secondaries_ar_none_n_sec_pred_without_stop_head_raises():
     stage1_out = torch.randn(3, X_DIM)
     with pytest.raises(AssertionError):
         sample_secondaries_ar(decoder, cond_cont, cond_cat, stage1_out, None, steps=2)
-
-
-# ── resolve_n_sec: n_sec.mode = "head" sampling policy (gitea #86) ──────────
-
-
-def _force_n_sec_head_bias(decoder: Stage2Autoregressive, bias: torch.Tensor) -> None:
-    """Zeroes n_sec_head's weights and pins its bias, so predict_n_sec
-    returns `bias` (broadcast over the batch) as logits regardless of
-    conditioning — mirrors `_force_stop_head_logit`."""
-    assert decoder.n_sec_head is not None
-    last_linear = decoder.n_sec_head[-1]
-    with torch.no_grad():
-        last_linear.weight.zero_()
-        last_linear.bias.copy_(bias)
-
-
-@pytest.mark.parametrize("n_sec_sampling", ["greedy", "sample"])
-def test_resolve_n_sec_head_mode_sharply_peaked_logits_pick_dominant_class(n_sec_sampling):
-    """A logit vector overwhelmingly favoring one class gives the same
-    answer under both policies — greedy because it's the argmax, sample
-    because softmax puts ~all mass on it."""
-    B, k_max = 8, 5
-    decoder = _stage2_ar("physical", "flow", k_max=k_max, n_sec_sampling=n_sec_sampling)
-    bias = torch.full((k_max + 1,), -50.0)
-    bias[2] = 50.0
-    _force_n_sec_head_bias(decoder, bias)
-    cond_cont, cond_cat = _cond(B)
-    stage1_out = torch.randn(B, X_DIM)
-    n_sec = resolve_n_sec(decoder, decoder, cond_cont, cond_cat, stage1_out, None)
-    assert n_sec is not None
-    assert torch.equal(n_sec, torch.full((B,), 2, dtype=torch.long))
-
-
-def test_resolve_n_sec_head_mode_greedy_is_deterministic_under_flat_logits():
-    B, k_max = 32, 5
-    decoder = _stage2_ar("physical", "flow", k_max=k_max, n_sec_sampling="greedy")
-    _force_n_sec_head_bias(decoder, torch.zeros(k_max + 1))
-    cond_cont, cond_cat = _cond(B)
-    stage1_out = torch.randn(B, X_DIM)
-    n_sec = resolve_n_sec(decoder, decoder, cond_cont, cond_cat, stage1_out, None)
-    assert n_sec is not None
-    assert n_sec.unique().numel() == 1
-
-
-def test_resolve_n_sec_head_mode_sample_varies_under_flat_logits():
-    """Under a flat logit vector, a categorical draw across a large batch
-    should hit more than one class — the whole point of gitea #86: greedy
-    always collapses to one, sample should not."""
-    torch.manual_seed(0)
-    B, k_max = 256, 5
-    decoder = _stage2_ar("physical", "flow", k_max=k_max, n_sec_sampling="sample")
-    _force_n_sec_head_bias(decoder, torch.zeros(k_max + 1))
-    cond_cont, cond_cat = _cond(B)
-    stage1_out = torch.randn(B, X_DIM)
-    n_sec = resolve_n_sec(decoder, decoder, cond_cont, cond_cat, stage1_out, None)
-    assert n_sec is not None
-    assert n_sec.unique().numel() > 1

tests/test_workflow_spec.py

@@ -0,0 +1,148 @@
+"""Workflow spec parsing, validation, and spec hashes (gitea #83)."""
+
+import pytest
+
+from giant.workflow.spec import (
+    WorkflowSpecError,
+    epoch_milestones,
+    load_spec,
+    parse_spec,
+    spec_hash,
+)
+
+MINIMAL = {
+    "workflow": {"name": "wf", "result_dir": "/tmp/wf"},
+    "condor": {"accounting_group": "cms", "repo_dir": "/work/lbogner/giant"},
+    "dataset": {"steps": "/data/train", "reference": "/data/holdout"},
+    "train": [{"name": "a", "epochs": 3}],
+    "rollout": [{"name": "a", "train": "a"}],
+    "analysis": [{"name": "cmp", "rollouts": ["a"], "chunks": 4}],
+}
+
+
+def _spec(**patch):
+    raw = {k: (v.copy() if isinstance(v, dict) else list(v)) for k, v in MINIMAL.items()}
+    raw.update(patch)
+    return parse_spec(raw)
+
+
+def test_parses_minimal_spec():
+    spec = _spec()
+    assert spec.name == "wf"
+    assert spec.log_dir == "/tmp/wf/logs"  # derived from result_dir
+    assert spec.train("a").epochs == 3
+    assert spec.rollout("a").train == "a"
+    assert spec.analysis("cmp").rollouts == ("a",)
+    # defaults come from the dataclasses, not the file
+    assert spec.geometry.method == "slab"
+    assert spec.condor.docker_image_gpu == "mschnepf/slc7-condocker"
+
+
+def test_example_config_is_valid():
+    spec = load_spec("configs/workflow_example.toml")
+    assert {t.name for t in spec.trains} == {"baseline", "router-balanced"}
+    assert spec.analysis("baseline-vs-router").rollouts == ("baseline", "router-balanced")
+
+
+@pytest.mark.parametrize(
+    "patch, message",
+    [
+        ({"train": [{"name": "a"}, {"name": "a"}]}, "unique"),
+        ({"rollout": [{"name": "r", "train": "nope"}]}, "names no"),
+        ({"analysis": [{"name": "c", "rollouts": ["nope"]}]}, "not defined"),
+        ({"analysis": [{"name": "c", "rollouts": []}]}, "at least one"),
+        ({"analysis": [{"name": "c", "rollouts": ["a"], "chunks": 0}]}, "chunks must be"),
+        ({"train": [{"name": "a", "epochs": 0}]}, "epochs must be"),
+        ({"train": [{"name": "a", "epchs": 3}]}, "unknown key"),
+        ({"geometry": {"methd": "slab"}}, "unknown key"),
+    ],
+)
+def test_validation_errors(patch, message):
+    with pytest.raises(WorkflowSpecError, match=message):
+        _spec(**patch)
+
+
+def test_unknown_top_level_table_rejected():
+    with pytest.raises(WorkflowSpecError, match="unknown top-level"):
+        _spec(nonsense={})
+
+
+def test_missing_required_table_rejected():
+    raw = {k: v for k, v in MINIMAL.items() if k != "dataset"}
+    with pytest.raises(WorkflowSpecError, match=r"missing required \[dataset\]"):
+        parse_spec(raw)
+
+
+def test_unknown_lookup_names_are_explicit():
+    spec = _spec()
+    with pytest.raises(WorkflowSpecError, match="no \\[\\[train\\]\\] named 'zzz'"):
+        spec.train("zzz")
+
+
+def test_hash_is_stable_and_order_independent():
+    a = _spec()
+    b = parse_spec(
+        {
+            "dataset": MINIMAL["dataset"],
+            "condor": MINIMAL["condor"],
+            "workflow": MINIMAL["workflow"],
+            "train": MINIMAL["train"],
+            "rollout": MINIMAL["rollout"],
+            "analysis": MINIMAL["analysis"],
+        }
+    )
+    assert a.train_hash("a") == b.train_hash("a")
+    assert a.analysis_hash("cmp") == b.analysis_hash("cmp")
+    assert len(a.train_hash("a")) == 8
+
+
+def test_hash_changes_with_own_settings():
+    base = _spec()
+    changed = _spec(train=[{"name": "a", "epochs": 4}])
+    assert base.train_hash("a") != changed.train_hash("a")
+
+
+def test_hash_propagates_from_parents():
+    """A dataset change must move every downstream task's directory."""
+    base = _spec()
+    changed = _spec(dataset={"steps": "/data/other", "reference": "/data/holdout"})
+    assert base.train_hash("a") != changed.train_hash("a")
+    assert base.rollout_hash("a") != changed.rollout_hash("a")
+    assert base.analysis_hash("cmp") != changed.analysis_hash("cmp")
+
+    # ... and so must a change to a training the analysis transitively uses.
+    retrained = _spec(train=[{"name": "a", "epochs": 9}])
+    assert retrained.analysis_hash("cmp") != base.analysis_hash("cmp")
+    # while an unrelated knob on the analysis leaves the training alone
+    rebinned = _spec(analysis=[{"name": "cmp", "rollouts": ["a"], "chunks": 4, "bins": 99}])
+    assert rebinned.train_hash("a") == base.train_hash("a")
+    assert rebinned.analysis_hash("cmp") != base.analysis_hash("cmp")
+
+
+def test_warm_cache_hash_ignores_epochs():
+    """Epoch count doesn't change the setup cache, so it must not re-warm it."""
+    base = _spec()
+    longer = _spec(train=[{"name": "a", "epochs": 50}])
+    assert base.warm_cache_hash("a") == longer.warm_cache_hash("a")
+    other_cfg = _spec(train=[{"name": "a", "epochs": 3, "config": "configs/router.toml"}])
+    assert base.warm_cache_hash("a") != other_cfg.warm_cache_hash("a")
+
+
+def test_spec_hash_expands_dataclasses():
+    spec = _spec()
+    assert spec_hash(spec.dataset) == spec_hash(spec.dataset)
+    assert spec_hash(spec.dataset) != spec_hash(spec.geometry)
+
+
+@pytest.mark.parametrize(
+    "epochs, per_job, expected",
+    [
+        (3, 1, [1, 2, 3]),
+        (10, 3, [3, 6, 9, 10]),
+        (9, 3, [3, 6, 9]),
+        (1, 5, [1]),
+    ],
+)
+def test_epoch_milestones(epochs, per_job, expected):
+    spec = _spec(train=[{"name": "a", "epochs": epochs, "epochs_per_job": per_job}])
+    assert epoch_milestones(spec.train("a")) == expected

tests/test_workflow_tasks.py

@@ -0,0 +1,172 @@
+"""Workflow task graph: dependencies, output paths, condor settings (gitea #83)."""
+
+import pytest
+
+from giant.analysis.catalog import catalog_ids, get_spec as get_plot_spec
+from giant.workflow import tasks
+from giant.workflow.spec import parse_spec
+
+CONDOR = {
+    "accounting_group": "cms",
+    "repo_dir": "/work/lbogner/giant",
+    "env_script": "/work/lbogner/giant/condor_env.sh",
+}
+
+RAW = {
+    "workflow": {"name": "wf", "result_dir": "/results/wf"},
+    "condor": CONDOR,
+    "dataset": {"steps": "/data/train", "reference": "/data/holdout"},
+    "train": [
+        {"name": "base", "epochs": 3, "gpu_memory_mb": 20000},
+        {"name": "router", "epochs": 2},
+    ],
+    "rollout": [
+        {"name": "base", "train": "base"},
+        {"name": "router", "train": "router"},
+    ],
+    "analysis": [{"name": "cmp", "rollouts": ["base", "router"], "chunks": 4}],
+}
+
+
+@pytest.fixture
+def spec():
+    s = parse_spec(RAW)
+    tasks.set_spec(s)
+    return s
+
+
+def _requires(task):
+    return list(task.requires() or [])
+
+
+def test_epoch_chain_is_linear_and_rooted_at_warm_cache(spec):
+    h = spec.train_hash("base")
+    third = tasks.TrainEpochTask(name="base", spec_hash=h, milestone=3)
+    second = _requires(third)
+    assert [type(t) for t in second] == [tasks.TrainEpochTask]
+    assert second[0].milestone == 2
+    first = _requires(second[0])[0]
+    assert first.milestone == 1
+    root = _requires(first)
+    assert [type(t) for t in root] == [tasks.WarmCacheTask]
+    # the warm cache is keyed by its own hash, not the training's
+    assert root[0].spec_hash == spec.warm_cache_hash("base")
+
+
+def test_epoch_task_outputs_last_pt_per_milestone(spec):
+    h = spec.train_hash("base")
+    path = tasks.TrainEpochTask(name="base", spec_hash=h, milestone=2).output().path
+    assert path == f"/results/wf/train_epoch/name=base/spec_hash={h}/epochs=2/last.pt"
+
+
+def test_train_task_requires_final_epoch_and_publishes_canonical_outputs(spec):
+    h = spec.train_hash("base")
+    train = tasks.TrainTask(name="base", spec_hash=h)
+    (dep,) = _requires(train)
+    assert isinstance(dep, tasks.TrainEpochTask) and dep.milestone == 3
+    out = train.output()
+    assert set(out) == {"best.pt", "last.pt", "metrics.csv"}
+    assert out["best.pt"].path == f"/results/wf/train/name=base/spec_hash={h}/best.pt"
+    # local: it only copies files around, no reason to queue a job for it
+    assert train.batch_system == "local"
+
+
+def test_rollout_requires_training_geometry_and_reference(spec):
+    ro = tasks.RolloutTask(name="base", spec_hash=spec.rollout_hash("base"))
+    deps = _requires(ro)
+    assert [type(d) for d in deps] == [tasks.TrainTask, tasks.GeometryOracleTask, tasks.DatasetTask]
+    assert deps[0].name == "base"
+    assert deps[2].path == "/data/holdout"
+    out = ro.output()
+    assert out["rollout.yaml"].path.endswith("rollout.yaml")
+    # the sidecar sits next to the parquet — the deterministic path
+    # `giant rollout --out` now produces
+    assert out["rollout.yaml"].path[: -len(".yaml")] == out["rollout.parquet"].path[: -len(".parquet")]
+
+
+def test_analysis_prep_requires_every_named_rollout(spec):
+    prep = tasks.AnalysisPrepTask(name="cmp", spec_hash=spec.analysis_hash("cmp"))
+    deps = _requires(prep)
+    assert [d.name for d in deps] == ["base", "router"]
+    assert all(isinstance(d, tasks.RolloutTask) for d in deps)
+    assert prep.batch_system == "local"
+
+
+def test_compute_job_enumeration_collapses_non_chunkable_specs(spec):
+    jobs = tasks.analysis_jobs(spec, "cmp")
+    non_chunkable = [i for i in catalog_ids() if not get_plot_spec(i).chunkable]
+    expected = (len(catalog_ids()) - len(non_chunkable)) * 4 + len(non_chunkable)
+    assert len(jobs) == expected
+    assert non_chunkable, "expected some chunkable=False specs in the catalog"
+    for spec_id in non_chunkable:
+        assert [c for i, c in jobs if i == spec_id] == [0]
+
+
+def test_compute_output_matches_the_on_disk_contract(spec):
+    h = spec.analysis_hash("cmp")
+    task = tasks.AnalysisComputeTask(name="cmp", spec_hash=h, plot_id="event_mean_length", chunk=2)
+    assert task.output().path == (
+        f"/results/wf/analysis/name=cmp/spec_hash={h}/reduced_partial/event_mean_length__2.json"
+    )
+    (dep,) = _requires(task)
+    assert isinstance(dep, tasks.AnalysisPrepTask)
+
+
+def test_render_requires_every_compute_job_and_runs_locally(spec):
+    render = tasks.AnalysisRenderTask(name="cmp", spec_hash=spec.analysis_hash("cmp"))
+    deps = _requires(render)
+    assert len(deps) == len(tasks.analysis_jobs(spec, "cmp"))
+    assert render.batch_system == "local"  # the only step importing plotstyle/LaTeX
+    assert render.output().path.endswith("/plots/metadata.yaml")
+
+
+def test_workflow_task_wraps_every_analysis(spec):
+    deps = _requires(tasks.WorkflowTask(workflow_name="wf"))
+    assert [(type(d), d.name) for d in deps] == [(tasks.AnalysisRenderTask, "cmp")]
+
+
+def test_workflow_without_analysis_falls_back_to_rollouts():
+    raw = {k: v for k, v in RAW.items() if k != "analysis"}
+    tasks.set_spec(parse_spec(raw))
+    deps = _requires(tasks.WorkflowTask(workflow_name="wf"))
+    assert [type(d) for d in deps] == [tasks.RolloutTask, tasks.RolloutTask]
+
+
+def test_gpu_settings_carry_remote_ceph_and_pins(spec):
+    settings = tasks.TrainEpochTask(name="base", spec_hash=spec.train_hash("base"), milestone=1).htcondor_settings
+    assert settings["+RemoteJob"] == "True"
+    assert settings["RequestGPUs"] == 1
+    assert "TARGET.ProvidesEtpCeph =?= True" in settings["requirements"]
+    assert "TARGET.GPUs_GlobalMemoryMb >= 20000" in settings["requirements"]
+    assert settings["accounting_group"] == "cms"
+    assert settings["docker_image"] == "mschnepf/slc7-condocker"
+
+
+def test_cpu_settings_used_for_analysis_compute(spec):
+    task = tasks.AnalysisComputeTask(
+        name="cmp", spec_hash=spec.analysis_hash("cmp"), plot_id="event_mean_length", chunk=0
+    )
+    settings = task.htcondor_settings
+    assert settings["docker_image"] == "cverstege/alma9-gridjob"
+    assert "RequestGPUs" not in settings
+    # no run_meta.json yet (prep hasn't run), so no walltime is claimed
+    assert "+RequestWalltime" not in settings
+
+
+def test_cpu_settings_local_files_use_provides_etp_resources():
+    raw = {**RAW, "condor": {**CONDOR, "remote": False}}
+    spec = parse_spec(raw)
+    tasks.set_spec(spec)
+    settings = tasks.GeometryOracleTask(spec_hash=spec.geometry_hash()).htcondor_settings
+    assert settings["requirements"] == "TARGET.ProvidesETPResources"
+    assert "+RemoteJob" not in settings
+
+
+def test_missing_dataset_fails_immediately(spec):
+    with pytest.raises(FileNotFoundError, match="/ceph"):
+        tasks.DatasetTask(path="/data/train").complete()
+
+
+def test_dataset_that_exists_is_complete(tmp_path, spec):
+    (tmp_path / "steps.parquet").write_text("")
+    assert tasks.DatasetTask(path=str(tmp_path / "steps.parquet")).complete()

uv.lock

@@ -147,6 +147,26 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/44/a1/70ebfffd6c6edc6034a547838ee46287c65ed89f710592ddc39c76b4a5a8/awkward_cpp-53-cp314-cp314t-win_arm64.whl", hash = "sha256:1be0c1d87d9f4fdf94b767a061df849f1bb21579d302b2996fb101527fc80a97", size = 551257, upload-time = "2026-06-08T12:31:56.319Z" },
 ]
 
+[[package]]
+name = "b2luigi"
+version = "1.2.9"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cachetools" },
+    { name = "colorama" },
+    { name = "gitpython" },
+    { name = "jinja2" },
+    { name = "luigi" },
+    { name = "parse" },
+    { name = "setuptools" },
+    { name = "tenacity" },
+    { name = "webdavclient3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/e5/5d/0c3e0602b6cf80a2cfebbe54c330f3623b4de227c2b0f8cbff437dc8d62c/b2luigi-1.2.9.tar.gz", hash = "sha256:3f6734b06970cd5bf6bb62c45c095e9c02b6e283a45b76a29ffabdeaa8fce0c0", size = 786252, upload-time = "2026-04-17T16:38:23.316Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c3/82/1a8d3c20bc235e665300c42783eb8bf9f2aac49c7eae644273191e5e01e8/b2luigi-1.2.9-py3-none-any.whl", hash = "sha256:9ca28b4203f5946394b609da432e53b3c9e35540c840bdc6497cce6ab4c82e2e", size = 115103, upload-time = "2026-04-17T16:38:21.044Z" },
+]
+
 [[package]]
 name = "bracex"
 version = "3.0.1"
@@ -176,6 +196,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/b3/0b/5885530f79d4400368b9d4dcb9b39274c0d52e633f7871e7fc6feceea1e3/bump_my_version-1.5.1-py3-none-any.whl", hash = "sha256:df3e2989d0d7fe704718feb24a5880f089b6b6369e427a4445b89c3adebfcff1", size = 65090, upload-time = "2026-08-06T14:26:37.083Z" },
 ]
 
+[[package]]
+name = "cachetools"
+version = "7.1.7"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/70/d2/47e8bc06fe2a06d3f5bdf20f1126ab66c4e99dc48d940e7ba873f7ac7131/cachetools-7.1.7.tar.gz", hash = "sha256:a3e2a00b14d8f8a6b70c1dae7b4685e7ad3bc965c5b42124a2d6ce895da6cf50", size = 40680, upload-time = "2026-08-01T21:20:40.434Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e4/d8/767faeda872075724b95dd675466a645f1b92aadcdcf2d1429dcfd76c176/cachetools-7.1.7-py3-none-any.whl", hash = "sha256:ef98ef375ad188819ef2f9b3645e3987f4b8c5b7550e436ad998c2de78296df0", size = 16830, upload-time = "2026-08-01T21:20:38.977Z" },
+]
+
 [[package]]
 name = "certifi"
 version = "2026.7.22"
@@ -605,6 +634,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/05/7f/798705f5296a58ca505d600456748d1be48078eac8a7050d8a98bc9edb89/decorator-5.3.1-py3-none-any.whl", hash = "sha256:f47fe6fdbd2edd623ecfe36875d37aba411624e2670dd395dddae1358689bb3c", size = 10365, upload-time = "2026-05-18T06:03:26.517Z" },
 ]
 
+[[package]]
+name = "docutils"
+version = "0.23"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/39/a4/5180d9afc57e8fca05601dd652bdff19604c218814037fe90ffc7625a50a/docutils-0.23.tar.gz", hash = "sha256:746f5060322511280a1e50eb76846ed6bf2342984b2ac04dc42caa1a8d78799e", size = 2303823, upload-time = "2026-05-27T17:41:06.934Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/32/91/30151a39f7570f448ed84529390628a651d7f27c87d73c9b887f8189695e/docutils-0.23-py3-none-any.whl", hash = "sha256:25d013af9bf23bc1c7b2b093dff4208166c53a94786c9e447808335ef1185fea", size = 634701, upload-time = "2026-05-27T17:40:58.442Z" },
+]
+
 [[package]]
 name = "executing"
 version = "2.2.1"
@@ -675,7 +713,7 @@ wheels = [
 
 [[package]]
 name = "giant"
-version = "0.3.15"
+version = "0.3.10"
 source = { editable = "." }
 dependencies = [
     { name = "numpy" },
@@ -728,10 +766,14 @@ geometry = [
 wandb = [
     { name = "wandb" },
 ]
+workflow = [
+    { name = "b2luigi" },
+]
 
 [package.metadata]
 requires-dist = [
     { name = "awkward", marker = "extra == 'convert'", specifier = ">=2.6,<3" },
+    { name = "b2luigi", marker = "extra == 'workflow'", specifier = ">=1.0,<2" },
     { name = "bump-my-version", marker = "extra == 'dev'", specifier = ">=1.2,<2" },
     { name = "giant", extras = ["convert", "analysis", "geometry", "wandb"], marker = "extra == 'dev'" },
     { name = "git-cliff", marker = "extra == 'dev'", specifier = ">=2,<3" },
@@ -757,7 +799,7 @@ requires-dist = [
     { name = "uproot", marker = "extra == 'convert'", specifier = ">=5.3,<6" },
     { name = "wandb", marker = "extra == 'wandb'", specifier = ">=0.16,<1" },
 ]
-provides-extras = ["cpu", "cuda", "dev", "geometry", "wandb", "convert", "analysis"]
+provides-extras = ["cpu", "cuda", "dev", "geometry", "wandb", "convert", "analysis", "workflow"]
 
 [[package]]
 name = "git-cliff"
@@ -779,6 +821,30 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/99/b2/99fac50978b9a90bfec0f1b89354a667ec83f4990301f6c708abce05484e/git_cliff-2.13.1-py3-none-win_amd64.whl", hash = "sha256:856d831a0bede9c258229dbd4d4c2b1c0810d8fce3d3882729669e8dc09c72bf", size = 7714969, upload-time = "2026-04-26T10:33:40.163Z" },
 ]
 
+[[package]]
+name = "gitdb"
+version = "4.0.12"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "smmap" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/72/94/63b0fc47eb32792c7ba1fe1b694daec9a63620db1e313033d18140c2320a/gitdb-4.0.12.tar.gz", hash = "sha256:5ef71f855d191a3326fcfbc0d5da835f26b13fbcba60c32c21091c349ffdb571", size = 394684, upload-time = "2025-01-02T07:20:46.413Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a0/61/5c78b91c3143ed5c14207f463aecfc8f9dbb5092fb2869baf37c273b2705/gitdb-4.0.12-py3-none-any.whl", hash = "sha256:67073e15955400952c6565cc3e707c554a4eea2e428946f7a4c162fab9bd9bcf", size = 62794, upload-time = "2025-01-02T07:20:43.624Z" },
+]
+
+[[package]]
+name = "gitpython"
+version = "3.1.60"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "gitdb" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/84/14/e6b1a48d831755a53c2029351fcef82e70db4a08f338daefe29d8d0cf31c/gitpython-3.1.60.tar.gz", hash = "sha256:e936431879fa85581b4311fa63492ea52251909e2d655b6529c704c904ddcc24", size = 230793, upload-time = "2026-08-25T18:33:46.102Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/71/63/ba28697918b7c190af9f3f21940d03e8814e25dd4ddd39d6929f3a553995/gitpython-3.1.60-py3-none-any.whl", hash = "sha256:39548bffb8fa0f3a548133348868bb4838e79d73283052207dc97781a569b6b4", size = 221893, upload-time = "2026-08-25T18:33:44.75Z" },
+]
+
 [[package]]
 name = "h11"
 version = "0.16.0"
@@ -1073,6 +1139,141 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/b5/91/53255615acd2a1eaca307ede3c90eb550bae9c94581f8c00081b6b1c8f44/kiwisolver-1.5.0-graalpy312-graalpy250_312_native-win_amd64.whl", hash = "sha256:1f1489f769582498610e015a8ef2d36f28f505ab3096d0e16b4858a9ec214f57", size = 75987, upload-time = "2026-03-09T13:15:39.65Z" },
 ]
 
+[[package]]
+name = "lockfile"
+version = "0.12.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/17/47/72cb04a58a35ec495f96984dddb48232b551aafb95bde614605b754fe6f7/lockfile-0.12.2.tar.gz", hash = "sha256:6aed02de03cba24efabcd600b30540140634fc06cfa603822d508d5361e9f799", size = 20874, upload-time = "2015-11-25T18:29:58.279Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c8/22/9460e311f340cb62d26a38c419b1381b8593b0bb6b5d1f056938b086d362/lockfile-0.12.2-py2.py3-none-any.whl", hash = "sha256:6c3cb24f344923d30b2785d5ad75182c8ea7ac1b6171b08657258ec7429d50fa", size = 13564, upload-time = "2015-11-25T18:29:51.462Z" },
+]
+
+[[package]]
+name = "luigi"
+version = "3.7.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "python-daemon" },
+    { name = "python-dateutil" },
+    { name = "tenacity" },
+    { name = "tornado" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/b0/df/5235521bd30402e19e469b79c9e91757046014a3d9bc2c91fd341b0edcb6/luigi-3.7.3.tar.gz", hash = "sha256:82e3a5d482e9738f4dcae5acb631ee7ffa748896cb1a1d5d4922c2cac61903fe", size = 1295810, upload-time = "2026-02-12T01:49:34.147Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/63/42/a311f89099c5714dc9c69368d4197a99d453d5f3764713eb5c09b8a8cf18/luigi-3.7.3-py3-none-any.whl", hash = "sha256:f6aaac2417e28a833d30e6b1cd03fe53b2254024dd861fa104ab342cc4f3b843", size = 1186711, upload-time = "2026-02-12T01:49:32.78Z" },
+]
+
+[[package]]
+name = "lxml"
+version = "6.1.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/ad/a9/970b8fa0ecc4fbf1dfaed0d89bbc1fc1421b25ec26a2038c91e872dc6c8e/lxml-6.1.2.tar.gz", hash = "sha256:1055241852f2b02068af4a625a5d32c087db193c12251928af2562ecd2239f18", size = 4210626, upload-time = "2026-08-19T04:58:15.341Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ee/a4/55eb54507073089ab27743c5da2113c84f0d0b1715b33175fdd943c9652d/lxml-6.1.2-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:7d506bdba580ecb1a6ad2e2b5c49445e66d3e1f95894885739094393a1aad237", size = 8602111, upload-time = "2026-08-19T04:58:28.017Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/bf/6332f45d78da385bb01d5cac3fe4acda19f025d1307cbc7ad538355fecbb/lxml-6.1.2-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:12acd337d2821cb8b9247dfe4b7aa2f2769a3df5ae8511b7e550df42b8f4d3c3", size = 4638376, upload-time = "2026-08-19T04:58:41.181Z" },
+    { url = "https://files.pythonhosted.org/packages/68/e0/21fba0fe74d417fbe976903ae6bc77e92cdce01aae7b636abd87756f4588/lxml-6.1.2-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5078ff51e6316c0f75ea8127c2cd24374747fb351f62fb93d1761f8ae5a04a40", size = 4939689, upload-time = "2026-08-19T04:58:48.526Z" },
+    { url = "https://files.pythonhosted.org/packages/de/e5/ce3e885264fdd0bdcb6b49c1ea1842f94281b39e4ff956099e8d57532c60/lxml-6.1.2-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9477e14217c212e6023c994a71a1a349db19b0e10fd5bf189666b281ae63b1fd", size = 5105185, upload-time = "2026-08-19T04:59:15.533Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/b6/990a8446c488c70fa25681e150de94b7bf2eaaf387e374d195ab3c8faafb/lxml-6.1.2-cp312-cp312-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:261d98065326676d7253882db0198d0aa06748d7ee0443367acf10b148273f99", size = 5011863, upload-time = "2026-08-19T04:59:50.58Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/6a/f70f41363dae27e3bfd6224b128f5ba150874bd32ca4938552930ffa33b0/lxml-6.1.2-cp312-cp312-manylinux_2_26_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:0666943ee1576fa890a6dc6316ef42e8241b5dd56f67bc5475acb2ac298c6ca9", size = 5638234, upload-time = "2026-08-19T05:00:00.802Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/e0/a65b64f34d556925faef2c4f14167d58c571bc15a3e1f2bba71138830562/lxml-6.1.2-cp312-cp312-manylinux_2_26_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:04cf9e3f4ee9cab9d9ba05401bef8668840fa9620fcd4d8e85a2d2fd0b0fa960", size = 5244532, upload-time = "2026-08-19T05:00:07.516Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/a9/471552e015e954fc9d960aa27c3d67ebf489683d03f033399a790417c67c/lxml-6.1.2-cp312-cp312-manylinux_2_28_i686.whl", hash = "sha256:9429d2371d406344ed1da5b5686d9412e74137c07b0171278368ff704f470ed5", size = 5358194, upload-time = "2026-08-19T05:00:22.747Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/0f/bc6248fbec2cc416f102b1267f1567e07510f6fa909bbe8cd2a22d6fb78e/lxml-6.1.2-cp312-cp312-manylinux_2_31_armv7l.whl", hash = "sha256:eff128ffdc093cc6317955934ad9751105d37ed8dbca3ff4ccd751af6be37185", size = 4704432, upload-time = "2026-08-19T05:00:51.115Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/3f/cec859f50e63f1fa338fab43d2362d7543e1237f2475960d8ab0769de0eb/lxml-6.1.2-cp312-cp312-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:ba58574d710b82ead7cbedea01cac3e110bc3ef82d4731519b74a2c11f7cf5e9", size = 5255038, upload-time = "2026-08-19T05:00:58.895Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/d9/2ced0cf2967115f92a1b8b3ae6bd18763abc3ebef88c98cf25145fda396c/lxml-6.1.2-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:52f6d4dff133c9778a24e9a2cfc1608930b15869866171aacc5131b5a418a003", size = 5054481, upload-time = "2026-08-19T05:01:10.096Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/f5/4f07386d3c88673daeec3b8cc09a2a4d39fa01c1fc49009791b0746d97fa/lxml-6.1.2-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:8807998c1023d1e9d60e02500f90e85a0752dbc0b670989806bba87b82dd5b42", size = 4785535, upload-time = "2026-08-19T05:01:18.909Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/5a/f4fe3ecbc189f48fba2547c5db5c940a10151d3e86b856a60a533a77e816/lxml-6.1.2-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:2170d0a280c877b6e2dc6738217db947be35dd8cf09ca458b355aa1bab2a9e70", size = 5655337, upload-time = "2026-08-19T05:01:41.324Z" },
+    { url = "https://files.pythonhosted.org/packages/92/c4/f586aa1bf27bfbace2dfdbb704da5c52f0bdece8ee440c8fb4946c940b2e/lxml-6.1.2-cp312-cp312-musllinux_1_2_riscv64.whl", hash = "sha256:c67f3c1278f942e97d8665c2a690324aaea5137de16f056583a21f0ac706177f", size = 5245778, upload-time = "2026-08-19T05:01:45.227Z" },
+    { url = "https://files.pythonhosted.org/packages/18/a1/677494bbaef4d6db5e4633af817414f478865850b55c03ae4bf70fa7b8ca/lxml-6.1.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:093fbf547d0f3ca02705381f795a050fbb58988be4aac7f79f99f280c4082313", size = 5267274, upload-time = "2026-08-19T05:01:57.687Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/71/b71425b8764d4cb7c92eb970483be7d5610dce2a6316242b5aaae7d260be/lxml-6.1.2-cp312-cp312-win32.whl", hash = "sha256:be365ce8d2d411cf2fb573747684b4fd470fa6224e0094d9d5a21155acc369d3", size = 3602563, upload-time = "2026-08-19T05:02:01.837Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/fb/909584e16d2148c1a252cc2c32dd99fe0e2682459c586d3d7a192e74a0ae/lxml-6.1.2-cp312-cp312-win_amd64.whl", hash = "sha256:b97153ca609b434b712ddfb92cd6af101a7045a7724c542258bd4727a344472f", size = 4005965, upload-time = "2026-08-19T05:02:07.157Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/8d/41207c9212caad0b52749e34739fb9bfab67486729f52a8fe9bd9266fee6/lxml-6.1.2-cp312-cp312-win_arm64.whl", hash = "sha256:7feb72424f19a893ae4f3373c7aae821b1aacb6076b708915c651f0683a97c49", size = 3666641, upload-time = "2026-08-19T05:02:11.3Z" },
+    { url = "https://files.pythonhosted.org/packages/61/2a/e9651f47a31a60b5cae031abc23391ed9aa30c8fc07571d1a38f58d6d770/lxml-6.1.2-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:351318f5c0eb7fcab5b4fdb507c6f88fb2c4b5e67784c7e5911448c91fffb5d4", size = 8590165, upload-time = "2026-08-19T04:58:40.489Z" },
+    { url = "https://files.pythonhosted.org/packages/61/87/a8098abaf35118767d1703b84c98940a5d833064e0eca39a00ecfe9840ab/lxml-6.1.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:c0edde95e4b4278dcc0175eda06dc8aa2631ad9f83ae5dbdbc4f0925e200b0b0", size = 4632474, upload-time = "2026-08-19T04:58:47.465Z" },
+    { url = "https://files.pythonhosted.org/packages/93/cc/fe74d1def7f4fb967c4a825608a074d4dbdbb871b0d6bd59c6ed07d67868/lxml-6.1.2-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:a8326e24ae6c3a6bfb03fa8b4793f9a5d804c125228aa067f652b0428e31b87c", size = 4936196, upload-time = "2026-08-19T04:59:03.477Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/ad/b96e6ca926e26726a99aa643602aac7411ecc1731ddb1b25af8cc57edfcd/lxml-6.1.2-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:7c534ed898413f439b048130011e99a4245ee13d62d431f6b4f7f2484d02a93a", size = 5093290, upload-time = "2026-08-19T04:59:17.498Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/84/616f5d3b7cd086fcfba3e5add6fccda67f976c1c753ae9ed7bbd317cb9be/lxml-6.1.2-cp313-cp313-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:2e37fe49fe2d5aa40a2cb1cc8176673ad7de0d124e6f4a509d9318f5979c7871", size = 4998767, upload-time = "2026-08-19T04:59:28.385Z" },
+    { url = "https://files.pythonhosted.org/packages/80/88/d5b453a8d083483c9442ad7f5ac5c560796022eb5c80d60b65d75e449236/lxml-6.1.2-cp313-cp313-manylinux_2_26_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:9b52ea73a37fc64aa3357ff8607801d46dd170506d3cf8253a91a1d91639d4f9", size = 5626717, upload-time = "2026-08-19T04:59:40.045Z" },
+    { url = "https://files.pythonhosted.org/packages/71/45/31e5aa4d4bae024908ba1d03480c7425cf027a28b7e5c88d1b7202bd80cc/lxml-6.1.2-cp313-cp313-manylinux_2_26_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:e8b9a92652e75e7731309ea51db5dee892eef414ce70a6ec3441e5d36bf5189f", size = 5232330, upload-time = "2026-08-19T04:59:46.175Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/5a/2627912420df8b2d31ba3014da5539f15ec85add01d42048864ffefda516/lxml-6.1.2-cp313-cp313-manylinux_2_28_i686.whl", hash = "sha256:9088da25ecd609965f838d89fda0465a905b48f4dd90331db9845518f2177372", size = 5347054, upload-time = "2026-08-19T04:59:52.762Z" },
+    { url = "https://files.pythonhosted.org/packages/16/86/54ac0f529b22a8f12313726dd49e12961bb46471d9028cc28d2a29408f0b/lxml-6.1.2-cp313-cp313-manylinux_2_31_armv7l.whl", hash = "sha256:0349321a0537d4fdbebb2af06dd1b64676132c72e2ae250de8cdb58f8c43019c", size = 4707275, upload-time = "2026-08-19T05:00:04.836Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/42/ffcdc6e4519be90df907cdae7e88409efb25d823ae4de8846f737dae1884/lxml-6.1.2-cp313-cp313-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:b20440e578d269c5e8a722ab602ddd0f0cedb8b080006b3f936da9991a593d3b", size = 5240071, upload-time = "2026-08-19T05:00:19.604Z" },
+    { url = "https://files.pythonhosted.org/packages/68/49/5b1d7ab35f013f1127ec48f3108319f58b65b00d5cb26f215adbe86eadfb/lxml-6.1.2-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:7766e525282dd38fd89567311323e441996eb958e8e816d16b38f782e3aecd2a", size = 5050356, upload-time = "2026-08-19T05:00:27.968Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/57/1cf049d054189b55c8fe8012269234f6602256949b69cd3ba80608a88219/lxml-6.1.2-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:9221442682c27417f10fe11184ea4cce174b25ab52465570b1f3ee3f85f320fa", size = 4780394, upload-time = "2026-08-19T05:00:39.047Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/ad/064488a8fa60e639fd773e421a18bf17541d02a95fbf36238ad7c65f69d4/lxml-6.1.2-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:75530642d8471327e691ab9b0513a5f9c77f38871014ceda40f51bb51765c0a1", size = 5645854, upload-time = "2026-08-19T05:03:42.697Z" },
+    { url = "https://files.pythonhosted.org/packages/85/bb/120e56f3cf1c149bb3b014278fb86d0a6dd552403981081f0ee0a0a57be7/lxml-6.1.2-cp313-cp313-musllinux_1_2_riscv64.whl", hash = "sha256:678e35f1cbca98f55107511ee21a60568535c950f3c2371819bd64504c980d20", size = 5231132, upload-time = "2026-08-19T05:03:45.466Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/2c/7d49aab893c128671a3276580074cce4c002896145b8dd2893da79633bca/lxml-6.1.2-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:5c2bae42b3a09f977330a08f4a8fe72aec58c4bdb89069d3fe7272a71d885881", size = 5256076, upload-time = "2026-08-19T05:03:48.092Z" },
+    { url = "https://files.pythonhosted.org/packages/72/28/ddea3aa1fa9acfd384fe34d4a2a93eecc07541dd2d922fa9b140c60d8014/lxml-6.1.2-cp313-cp313-win32.whl", hash = "sha256:5848f3de6a8de8a93cff9f068134393ff5fa69ac2a04399f7d49cd67c61c348c", size = 3602177, upload-time = "2026-08-19T05:03:50.571Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/7a/96bac167538748cae2544335855f812fa33e49a9a67bc8b8520dcbd592bd/lxml-6.1.2-cp313-cp313-win_amd64.whl", hash = "sha256:6cb0c87421946030b92b558be416852780a912454e3dcba0998e4497c9c588d5", size = 4004117, upload-time = "2026-08-19T05:03:53.074Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/24/9498fa3c84135956e5ef55ea4d8bd11e999e381f7f210fb6f8c6a980ef03/lxml-6.1.2-cp313-cp313-win_arm64.whl", hash = "sha256:648861c19b775b89ebefa14586f85090b10163367476d77f242c4131c835ce73", size = 3665412, upload-time = "2026-08-19T05:03:55.621Z" },
+    { url = "https://files.pythonhosted.org/packages/27/b5/728b0578791b397ace8d1b101c8b3fe10f36043542f7bb85f82d8bdc3f50/lxml-6.1.2-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:d50a44113fe6800dcc8a859332b823a4735b1e6ae1b0063882e4cca569ec3e29", size = 8609651, upload-time = "2026-08-19T04:58:42.42Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/6b/49209fa6225c15c48a30061f03d3aba75e3c19634813b88bf83b88c525ed/lxml-6.1.2-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:fa813b0247d0543a563b993ac3dba6168eef59e3a61448432cf5453300c2412b", size = 4639588, upload-time = "2026-08-19T04:59:01.501Z" },
+    { url = "https://files.pythonhosted.org/packages/20/86/80bae4e8bc2eed9d6f017701a3d86fdea56936218efa738911d0b76aa7f4/lxml-6.1.2-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:d858e718b94033ab4b67e4a58fe3114c65bae01ae2314a62fb39ae8897ed4324", size = 4964846, upload-time = "2026-08-19T04:59:08.59Z" },
+    { url = "https://files.pythonhosted.org/packages/70/ce/4782caee7a22959c1ac67cb46495e03912c22a4ba7d20c163496a519e815/lxml-6.1.2-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:3e3b666f57a5d81562f38c766c762416b0f6eb58a00590546911514b48412abd", size = 5099288, upload-time = "2026-08-19T04:59:18.649Z" },
+    { url = "https://files.pythonhosted.org/packages/32/21/f120967cc43b54e05512dff0c39726b832c836195d30f41f88733ef36ac8/lxml-6.1.2-cp314-cp314-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:26ff164c6629e5c4d11c9e55d5ea3d6eed0be2a420eee1f55cbce6e2c23e231a", size = 5036837, upload-time = "2026-08-19T04:59:47.217Z" },
+    { url = "https://files.pythonhosted.org/packages/61/ba/8005e9f47598e3ec5c18312c77f94e889580027616678848405c6aeba5de/lxml-6.1.2-cp314-cp314-manylinux_2_26_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:962c12b51d0b164f12569af225dea57568477e24a845b96eaccbef6c07e4cc03", size = 5658569, upload-time = "2026-08-19T04:59:54.078Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/ba/add33b3c7ce51462cf7a4637bcfec2eaa258364d6015b989dd7d1216e6a6/lxml-6.1.2-cp314-cp314-manylinux_2_26_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:47e367dfe341521426692819803e260d0673899c0ff611f14af978d725e2c999", size = 5246003, upload-time = "2026-08-19T04:59:59.764Z" },
+    { url = "https://files.pythonhosted.org/packages/05/b3/a43012748fb861c914c5eac1c1a3bad44282e767499cd02280d4d1edf092/lxml-6.1.2-cp314-cp314-manylinux_2_28_i686.whl", hash = "sha256:92c2b366028ac01e90399e6d17734ce6e4f4aeddd8ba75fbaf80ea11d6c6d645", size = 5354047, upload-time = "2026-08-19T05:00:21.657Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/cb/813021d9a445713b8d758b9e5eae2ed392cd598d9f119d9b053b37c2ab93/lxml-6.1.2-cp314-cp314-manylinux_2_31_armv7l.whl", hash = "sha256:7e81fc065ede5d58dd0bf0912025aee1bd04c52c2affd61fdb93226a97ce2fc6", size = 4704382, upload-time = "2026-08-19T05:00:47.067Z" },
+    { url = "https://files.pythonhosted.org/packages/17/c9/1155299f4577bebf3c280497534a73e4b8ad8cab3b96074731ad10949d4e/lxml-6.1.2-cp314-cp314-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:633ac039cb32366dd5935868e041e385875c017b8cd54ea56aeee3fe29ca5935", size = 5258530, upload-time = "2026-08-19T05:01:14.893Z" },
+    { url = "https://files.pythonhosted.org/packages/25/6e/d76e58384b378b877e140e25b9a9835da00035f81ff70cbe943a3749bf27/lxml-6.1.2-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:2f3194777c0d05945ac91d8594be25d2679d1d826e01e1fc90bae568ff3a547b", size = 5089919, upload-time = "2026-08-19T05:01:33.602Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/b7/898013c0f8891481d0624ab3bd5dd8c8ff827232dfee2a5d1f8bf970a7cc/lxml-6.1.2-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:1133bd969f2bfcc6b0c0cf7cdf5f2631e62b23fa2471ee8bd44f6ab73554ee9a", size = 4741972, upload-time = "2026-08-19T05:01:38.18Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/47/efb53c4d7b655831c03317a450d9da439b0829c61f34d9d4fe7c863445d6/lxml-6.1.2-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:1edca8f4a92b94e873093df959f141d388f2141fcad0c47598442fb4730ef57a", size = 5683241, upload-time = "2026-08-19T05:02:00.731Z" },
+    { url = "https://files.pythonhosted.org/packages/da/0a/0ff36a584cbba14a71326ee8a5300694400f0b97927d1f90a87d95b17d4a/lxml-6.1.2-cp314-cp314-musllinux_1_2_riscv64.whl", hash = "sha256:8512b3775d68994dd1d6d533161e0a214f2ad9c634659d34a99c98e86c6c3d68", size = 5245892, upload-time = "2026-08-19T05:02:06.108Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/9e/303717a1aa56d4bd775c91936717d3c9e8d999a8e8b68b00979c4c1f93d0/lxml-6.1.2-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:a5005c0c9e4d749a76a2ff8bd5918a8bb248df8e08e73a55654b9f79c9cd1e2b", size = 5269528, upload-time = "2026-08-19T05:02:09.883Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/c2/2ae7cb97089eb86bf0689516db3cf280a007b6145853d2a0235a1f01683d/lxml-6.1.2-cp314-cp314-win32.whl", hash = "sha256:e17e2c30e27f56da5551e7a425888b45f013e940b99ab07d125a1c33f77a4605", size = 3662743, upload-time = "2026-08-19T05:03:02.513Z" },
+    { url = "https://files.pythonhosted.org/packages/77/13/a3d483230a09201e211ceb1aa208b1374d27d23b8b180d74dba14b30f6b3/lxml-6.1.2-cp314-cp314-win_amd64.whl", hash = "sha256:87e9673cd8a3445024fe38e7f91b55fa3428437eec9b7a7ff7d81979520c0d2d", size = 4073942, upload-time = "2026-08-19T05:03:04.864Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/f1/c1445d4b6ad7c51e39d4e2ebbf015a4880f5b297a4ab0e77e4d0e5b70110/lxml-6.1.2-cp314-cp314-win_arm64.whl", hash = "sha256:878e7c8ada8f92c52f13f35a2ab98ef0adf7fd0211d164fc2af589e4c3cfed63", size = 3749235, upload-time = "2026-08-19T05:03:07.239Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/eb/598c76f4ce19a67c635e86a46d880cc854f308f39a6f1fdf13bbb01813ec/lxml-6.1.2-cp314-cp314t-macosx_10_15_universal2.whl", hash = "sha256:94162456ed0a64fb1c06915df5bd06af4675ae3966d6048fcb73b0906e0e0222", size = 8860315, upload-time = "2026-08-19T05:02:14.39Z" },
+    { url = "https://files.pythonhosted.org/packages/da/c7/1f9fac7b566a86ad0da13dcc0259164266469c0ad86744c740ccd5c2a081/lxml-6.1.2-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:4b0fa7109b1d0bc1747d8241a0853e135eefb1c978685241b544c46937383efd", size = 4755176, upload-time = "2026-08-19T05:02:18.705Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/1b/cfda9307388d496e7eeb7493d9455896b8137ed95f51f3d6ae6ddcc14a47/lxml-6.1.2-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:604f4778632588d7c000e7e19430639dc12fca58b5b6e99edffba7631725ef0e", size = 4979444, upload-time = "2026-08-19T05:02:21.262Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/71/f732c8919c45b7f29acf443288c6e90036877a67bfeeb1acceb0fffa011b/lxml-6.1.2-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5a096d6a5f96b776a5b020cb45c17c545effd2a3b6639e6fa97bc95537600923", size = 5115887, upload-time = "2026-08-19T05:02:23.62Z" },
+    { url = "https://files.pythonhosted.org/packages/30/00/121d52b944f41e33ea86c62875f902d24982842dc7231ab154ac5a6c6593/lxml-6.1.2-cp314-cp314t-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6454d184d556eaf4cb3d6f69e405d21602d6fdcf08b8d57796824275986c6595", size = 5032418, upload-time = "2026-08-19T05:02:26.114Z" },
+    { url = "https://files.pythonhosted.org/packages/70/19/cadb73c7fe48c7563dc8ab62ea53d5b920c8911bfb808507a6daa82e78d2/lxml-6.1.2-cp314-cp314t-manylinux_2_26_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:8b68f2548259bb04e0b3d5df0c397abe8b0080f5e1ffe4019fb7a8bf01a9339e", size = 5603304, upload-time = "2026-08-19T05:02:28.694Z" },
+    { url = "https://files.pythonhosted.org/packages/13/32/9de126a14d5a5db8c371c5ec869178417db226707b62a47273a95ae6df7f/lxml-6.1.2-cp314-cp314t-manylinux_2_26_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:6c9cc4b6532abe154dbdebb42aaba8d52c852919591e45067f5b7d46a0405e88", size = 5228938, upload-time = "2026-08-19T05:02:30.99Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/9b/22dd9e843629ed04652591fb220eb2bf2394d97be3be377d60d8083405d7/lxml-6.1.2-cp314-cp314t-manylinux_2_28_i686.whl", hash = "sha256:57188e441ab24f906bd5a5c14eb55363ab51aa6c0de549f3dd320043721cc118", size = 5317790, upload-time = "2026-08-19T05:02:33.301Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/2c/b12a1dc121f81c280635c721c7bcaa341441fcbe37397f60b8915048aece/lxml-6.1.2-cp314-cp314t-manylinux_2_31_armv7l.whl", hash = "sha256:d0bfd719c254bbe60ea022cff0e6ffb799a6fa7d4d72852cebe0257957b32d68", size = 4646468, upload-time = "2026-08-19T05:02:35.504Z" },
+    { url = "https://files.pythonhosted.org/packages/57/41/fd87a41edc531e7969c25ab1d6b52b5b041eb108b88f6394d6afb4374396/lxml-6.1.2-cp314-cp314t-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:be6f87cd224254a8f81324e34cc655508b83f1d70458a1a39857ad2aa9925852", size = 5240607, upload-time = "2026-08-19T05:02:37.805Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/30/713ba813b6e6673c6dc34733746516017efcd17949b767b154cc50bccf20/lxml-6.1.2-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:074a88f70a7360a4a0c5be5d898062cd26f898c25b459efb1bdd43ae700c5a1a", size = 5086495, upload-time = "2026-08-19T05:02:40.099Z" },
+    { url = "https://files.pythonhosted.org/packages/33/f8/6532ce0fecd9c326d06b08274ee075cc28dbc9f5e9285355db8504689114/lxml-6.1.2-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:9031f5f01452681abf39fdd65f84a70cb01a7572a1bbf570042e826b1232d07b", size = 4758801, upload-time = "2026-08-19T05:02:45.434Z" },
+    { url = "https://files.pythonhosted.org/packages/74/b6/5a1f7833ebaa0dd33c28f6f9755ec6ff3891bf63f097634b44e6da1bb65e/lxml-6.1.2-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:cfeac14425fc7a6fca7864b774d4ee63547926158f4a18c67d77b2c9a948acf1", size = 5626977, upload-time = "2026-08-19T05:02:48.092Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/20/6ae0fc1b45e20877cdcfb1168ceeaf9abb0fba5ed36bd639a260e7b2101e/lxml-6.1.2-cp314-cp314t-musllinux_1_2_riscv64.whl", hash = "sha256:8ec111ff8067325f85c08aa9c2b26179ec0537bb89c003fde31127139f85f82d", size = 5235036, upload-time = "2026-08-19T05:02:50.726Z" },
+    { url = "https://files.pythonhosted.org/packages/47/b4/2bc7b37fbb990ccfb7d30393660741592177224a94e07d842c8da70638e8/lxml-6.1.2-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:48e912f37c99a297175ba955f55a47c0e1c834b506ef162e52a6e4fe276e6e45", size = 5252270, upload-time = "2026-08-19T05:02:53.454Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/0b/07fb8e1dee29a78e2c5fa5c6c914218be76a6406baff27907429566e90ec/lxml-6.1.2-cp314-cp314t-win32.whl", hash = "sha256:7c444c3a6e8e75334879980eed96568f0e12064c8b1913424eac1805e976736b", size = 3902666, upload-time = "2026-08-19T05:02:55.607Z" },
+    { url = "https://files.pythonhosted.org/packages/58/ab/3371527bd9820aae6f511697c93032ed197b0d8dab0f17818f18d3099637/lxml-6.1.2-cp314-cp314t-win_amd64.whl", hash = "sha256:7f35ba7667004ecdafebbe08da7c9fa06ee6195275bb7ef7a29ee1901e69519c", size = 4401011, upload-time = "2026-08-19T05:02:57.899Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/bb/e6de9b2546a4e6df4fb52fb18921906a8b7a041aba06570995759a4d6d8b/lxml-6.1.2-cp314-cp314t-win_arm64.whl", hash = "sha256:d117f39b28ab8a330a74abdbe61c2255b51973b238db25fd6c2448de1eb2a02d", size = 3823384, upload-time = "2026-08-19T05:03:00.371Z" },
+    { url = "https://files.pythonhosted.org/packages/0e/83/7ff98683e14a148191278728d11ba782c3d5137886d49fd95ab4036efa1b/lxml-6.1.2-cp315-cp315-macosx_10_15_universal2.whl", hash = "sha256:1e3c67b817867c484794d7fe0d73045d7d0c67460c78a0a1249a9e92266e6a0e", size = 8609183, upload-time = "2026-08-19T04:58:32.19Z" },
+    { url = "https://files.pythonhosted.org/packages/24/39/c39f05e8240e98009dd3d4ceb248319d0f36467babc5f90a909ed0c5b68a/lxml-6.1.2-cp315-cp315-macosx_10_15_x86_64.whl", hash = "sha256:d3e97ac4353cca3fbbfa829bc0c6a913771573d1c6d46932d4335c46f2b7796a", size = 4639898, upload-time = "2026-08-19T04:58:39.017Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/bf/25e26b089510940a0777ab334357874569255e50930224c8159cd649e754/lxml-6.1.2-cp315-cp315-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:827438bf6c8292d22a409bb7990d7cffce410f33e7664e46ca74d2ecc26975ef", size = 5037527, upload-time = "2026-08-19T04:58:46.224Z" },
+    { url = "https://files.pythonhosted.org/packages/65/6d/aed3a58a3d662f7367a537fabe8c549f1446dbd043719e0ae8cd53f47819/lxml-6.1.2-cp315-cp315-manylinux_2_26_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c470d192e27f97842a068cf12a1c1296b20ca716c56a9249715c6654bc192d19", size = 5661918, upload-time = "2026-08-19T04:59:02.534Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/ca/706d32b6957c0c2e005a9833e8fc528449196b38d5cfcf9e0fd86a96fb00/lxml-6.1.2-cp315-cp315-manylinux_2_26_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ef0b8ba6e13597f681b2b4924ca9c4e8c88420bf0e21d9a9006c757f2fc39d1f", size = 5249359, upload-time = "2026-08-19T05:04:01.956Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/e9/445ff43f56fcffa06f6f3a7189920c216f3eacef68ef834d4111cdbd86ba/lxml-6.1.2-cp315-cp315-manylinux_2_31_armv7l.whl", hash = "sha256:65c32ddc5d0750129c7b119fb57d48192b76d334c21e6b690d19dfb06b34af79", size = 4704548, upload-time = "2026-08-19T05:04:04.57Z" },
+    { url = "https://files.pythonhosted.org/packages/69/78/20b8b7e79a1b1d9cd4465c332d62962858562b446692f16a27068fa54b85/lxml-6.1.2-cp315-cp315-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:0aa07065497f191ad26c4b587ce5dbb5a7105285a3789aafd0661750e8bac537", size = 5261170, upload-time = "2026-08-19T05:04:07.336Z" },
+    { url = "https://files.pythonhosted.org/packages/54/ca/84a0e1148bf511e12e0d99732a4e136a3bf1b91622f0a1b197796e2ff984/lxml-6.1.2-cp315-cp315-musllinux_1_2_aarch64.whl", hash = "sha256:cde6b8db7d2e5135129eb5e74b7b44dd2053aa767cd5023541fccedddc262453", size = 5090576, upload-time = "2026-08-19T05:04:09.968Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/f8/1ef6fc7070bed8753315f2e4ea66bc0d37620e1444d014db7f0267b8faaf/lxml-6.1.2-cp315-cp315-musllinux_1_2_armv7l.whl", hash = "sha256:b28842b30c4bc2e6afe137d98a5d2071a62589471e76d053bea55b0e53298af9", size = 4744614, upload-time = "2026-08-19T05:04:12.717Z" },
+    { url = "https://files.pythonhosted.org/packages/87/f6/3a4824cd1c1b81d996d2d75bbd176ba13fbe9b5d89489290d93ff9558486/lxml-6.1.2-cp315-cp315-musllinux_1_2_ppc64le.whl", hash = "sha256:11f529062255209a421ae4de5b1bb36b2f0a2e1a700745e675a4bf4084d13c00", size = 5685792, upload-time = "2026-08-19T05:04:15.367Z" },
+    { url = "https://files.pythonhosted.org/packages/64/9a/f133bf16a67149e00ca5d8a8f1ae662c30a86c303aa242693b67f8e19856/lxml-6.1.2-cp315-cp315-musllinux_1_2_riscv64.whl", hash = "sha256:f8b89b3be75a37509602b03f9cfa1a28298d4eed4625748148307aeb907901b7", size = 5248972, upload-time = "2026-08-19T05:04:18.491Z" },
+    { url = "https://files.pythonhosted.org/packages/50/63/273e7e8a73a5d183d8552dfdaa131dfda0292ddab7bcddc5a66a0ae525d8/lxml-6.1.2-cp315-cp315-musllinux_1_2_x86_64.whl", hash = "sha256:1a2331da06dd55a8184985306eb2afd72d708283ce7e85d67bba77317b785060", size = 5271809, upload-time = "2026-08-19T05:04:21.448Z" },
+    { url = "https://files.pythonhosted.org/packages/49/eb/614117c36a28909e79ff7cdec87008f0bd996478f35cf72309189cf398b1/lxml-6.1.2-cp315-cp315-win32.whl", hash = "sha256:442766b326d9892585a64e8c6c4b5ab81d0e6c0538c9f0fc11a84dc101a5d97f", size = 3662854, upload-time = "2026-08-19T05:05:07.141Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/e9/06aee6107cf8e7b870f10f82539f366cba10dc6053144cca80e838caf8c8/lxml-6.1.2-cp315-cp315-win_amd64.whl", hash = "sha256:a7fd1dd6faa3df9dcd8f1765237362cd885ca62cdf77a7c5f5ea383ae5b6048b", size = 4074590, upload-time = "2026-08-19T05:05:09.697Z" },
+    { url = "https://files.pythonhosted.org/packages/84/bf/dad9b6baf9b26d79584834e15cef2a5dd0a13c7b1df08831e8f18244b494/lxml-6.1.2-cp315-cp315-win_arm64.whl", hash = "sha256:054175250531a5fb102d485743ff16412279c93add12385b3b1c3d7b16d8deaa", size = 3749336, upload-time = "2026-08-19T05:05:12.334Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/9d/cd0c43d45e2eb52df7735c6558f24054ca633499191899b0cb9040fbbc3c/lxml-6.1.2-cp315-cp315t-macosx_10_15_universal2.whl", hash = "sha256:84a2a46b93b789d8acb44cfcb3d967ce9dbe29884ddb93fbb1a33f0e0c8fcd86", size = 8857688, upload-time = "2026-08-19T05:04:24.648Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/26/27093dc1a9edbdd8a54652f237a387f7e63ec0192efe708bc2576d8a1383/lxml-6.1.2-cp315-cp315t-macosx_10_15_x86_64.whl", hash = "sha256:4aced3284e0353c798b060fe2c175eb81410e99b9a7e2ae6951be5333732b111", size = 4754422, upload-time = "2026-08-19T05:04:27.645Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/ee/502f7c93507f57eb496744a64da8f4ca86855cf88e48d14584342f1bfd92/lxml-6.1.2-cp315-cp315t-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:47c92dc5167de16e27ace8332454f12ba172dcab04f7a78a9eae14e2e41b6a41", size = 5033396, upload-time = "2026-08-19T05:04:30.054Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/72/c4cbbe72f951650f2afe43a70e51687e111d82b9bec46e3310ea76419d46/lxml-6.1.2-cp315-cp315t-manylinux_2_26_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:40366c23a938008a3bedfcfd80709b3a857c188b4d710b083e978ef5d2c1c715", size = 5615298, upload-time = "2026-08-19T05:04:32.752Z" },
+    { url = "https://files.pythonhosted.org/packages/14/83/a3df966d6d7b6513e9dfb6fbfb041c0619642170359c1b36ab20a83e59eb/lxml-6.1.2-cp315-cp315t-manylinux_2_26_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1c4c6dc1b2485aaa4adfb6ed754f90dddcb2b96a66bbebc9e1ac242b5ce5e818", size = 5236282, upload-time = "2026-08-19T05:04:35.762Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/85/8692ec8173c9f8d295735b9bf410d202317e7b3ed11141e80a30f421f409/lxml-6.1.2-cp315-cp315t-manylinux_2_31_armv7l.whl", hash = "sha256:3a698fad6f122a9b3e2dc2fb598c1de7329c74a67c7a334c9109a440de2508e5", size = 4650647, upload-time = "2026-08-19T05:04:38.396Z" },
+    { url = "https://files.pythonhosted.org/packages/11/e7/dbe3cece28a5bf82997a091d9dbb0fc49e725a5fa86550897ee2cf6412e6/lxml-6.1.2-cp315-cp315t-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:14879fa5eb2b793c040bbfcb62011aa3015c65d6c9875e063ea98ce2029d51fb", size = 5243387, upload-time = "2026-08-19T05:04:41.247Z" },
+    { url = "https://files.pythonhosted.org/packages/99/a9/81a2d27640db0d27200b2f32339a54e74c36d58feb5ad528b87d52a59ecc/lxml-6.1.2-cp315-cp315t-musllinux_1_2_aarch64.whl", hash = "sha256:b631174cd2e4d9f8a94ef17f911c6ded10ede93b5e7860dee7bbf85961d321e9", size = 5092624, upload-time = "2026-08-19T05:04:43.919Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/f4/0b0304c70c087f618d95b0306738b070bd556afd09c2c92589b78dbe5eb0/lxml-6.1.2-cp315-cp315t-musllinux_1_2_armv7l.whl", hash = "sha256:ceafa5e0536c62a5cd9f65327fa0b57d6f0b0e3435daf2c98a78d0dde7ecbae1", size = 4758742, upload-time = "2026-08-19T05:04:46.615Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/ec/f9fc45f1d01b632b673e11880e75292dff9953db9f426d1a38201b8eb5f5/lxml-6.1.2-cp315-cp315t-musllinux_1_2_ppc64le.whl", hash = "sha256:7c482e87cc86bed78a50462560675bc2c348ef72c47596f9b933346d5a8e920e", size = 5649540, upload-time = "2026-08-19T05:04:49.777Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/0b/d65e0458c2bcce0df68d5cc29ad0006e76446f02d9e50caf188fd1fb8bae/lxml-6.1.2-cp315-cp315t-musllinux_1_2_riscv64.whl", hash = "sha256:b1c0d2dde8a50520efc51644587f0fc4810e3af7d3e029d7af0be93bf39e2b5c", size = 5234869, upload-time = "2026-08-19T05:04:52.972Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/62/1fee828238badd3bfe9544f5cc9ce6ded421ef38e9634030445dedd78b36/lxml-6.1.2-cp315-cp315t-musllinux_1_2_x86_64.whl", hash = "sha256:dd7ea3fa47154b9fff90591b961e41b3718bd7fcd5bc2d9bb47e9845c8ace088", size = 5259992, upload-time = "2026-08-19T05:04:56.028Z" },
+    { url = "https://files.pythonhosted.org/packages/20/18/35fb14dd6baccbffa6daeb2369802f04a94e3f73db3c7bb405dbab009729/lxml-6.1.2-cp315-cp315t-win32.whl", hash = "sha256:87534cec6ea325435e4adf2326b0cf3110eee9a47abf73652eb155db639c08c6", size = 3901151, upload-time = "2026-08-19T05:04:58.671Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/b6/07530896ca062bc3d2f09d5cb8a48e799c05b12c496205db03159ba13b6c/lxml-6.1.2-cp315-cp315t-win_amd64.whl", hash = "sha256:4e220a9c297e5d36895d489a08c9a3f1f6193b6414e702c5fb751e4a3767f8d0", size = 4395355, upload-time = "2026-08-19T05:05:01.651Z" },
+    { url = "https://files.pythonhosted.org/packages/31/a8/237d8de1d77085cfd41d0c6049a044d8d01886f3afb7f1eda2f43d900a96/lxml-6.1.2-cp315-cp315t-win_arm64.whl", hash = "sha256:f16a407766bac51c65d605b06d900821751a79aa20e12185f273f14a17180e7b", size = 3822823, upload-time = "2026-08-19T05:05:04.63Z" },
+]
+
 [[package]]
 name = "markdown-it-py"
 version = "4.2.0"
@@ -1518,6 +1719,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/0f/54/68a0978d1ef8502b8492099beaa6e7a0c1b32e3b5d4f677f5810cb08711c/pandas-3.0.3-cp314-cp314t-win_arm64.whl", hash = "sha256:b2c95f8bfc1ee412bf482605d7bfd30c12d1d26bd59fdd91efeef1d4718decb1", size = 9466464, upload-time = "2026-05-11T18:54:22.754Z" },
 ]
 
+[[package]]
+name = "parse"
+version = "1.22.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a4/f2/0b504486c2a5564798607d3860e48ed19c6443d5e9cc3ec61cc6b8b4ef58/parse-1.22.1.tar.gz", hash = "sha256:d3a4740ec3da338e2b258b2d69741b731eadfddca59e24a14bc4ee5fce38c911", size = 36970, upload-time = "2026-05-26T03:44:52.624Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6f/c5/7c16e99869e1f422629092cfd23e3b58e461988c3f9c36fd3624bb4142e6/parse-1.22.1-py2.py3-none-any.whl", hash = "sha256:20f0925a46f06602485ac90d751764d0697fd8455aaa97489ba8953a4b66de32", size = 20925, upload-time = "2026-05-26T03:44:51.156Z" },
+]
+
 [[package]]
 name = "parso"
 version = "0.8.7"
@@ -1956,6 +2166,20 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/9d/7a/d968e294073affff457b041c2be9868a40c1c71f4a35fcc1e45e5493067b/pytest_cov-7.1.0-py3-none-any.whl", hash = "sha256:a0461110b7865f9a271aa1b51e516c9a95de9d696734a2f71e3e78f46e1d4678", size = 22876, upload-time = "2026-03-21T20:11:14.438Z" },
 ]
 
+[[package]]
+name = "python-daemon"
+version = "2.1.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "docutils" },
+    { name = "lockfile" },
+    { name = "setuptools" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/b2/fb/a280d65f81e9d69989c8d6c4e0bb18d7280cdcd6d406a2cc3f4eb47d4402/python-daemon-2.1.2.tar.gz", hash = "sha256:261c859be5c12ae7d4286dc6951e87e9e1a70a882a8b41fd926efc1ec4214f73", size = 76176, upload-time = "2016-10-26T10:16:01.909Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f4/59/816004688f8e8602526553cd96226f34657ce4a86daa2240c3eebb0568a3/python_daemon-2.1.2-py2.py3-none-any.whl", hash = "sha256:53da55aec3bb67b576e13a8091a2181f99b395c2eec32a5a0d91d347a5c420a7", size = 19425, upload-time = "2016-10-26T10:16:05.441Z" },
+]
+
 [[package]]
 name = "python-dateutil"
 version = "2.9.0.post0"
@@ -2248,6 +2472,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/89/d3/726bd88f0eece09ddf431bea4c9191c18e7a8d070b854eb0014d447712ee/sentry_sdk-2.66.1-py3-none-any.whl", hash = "sha256:86002793161d9a95ef04bdd8d442e9bfece5d989b755f05d6360215094a7aff6", size = 505555, upload-time = "2026-07-22T12:26:52.71Z" },
 ]
 
+[[package]]
+name = "setuptools"
+version = "84.0.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/6d/44/f5da03a8ef95d369145c5bb53050e7877c9f3d312e128605fd9504829143/setuptools-84.0.0.tar.gz", hash = "sha256:f4695c21257f0d9b537ec2692c941d02ee143b7cc1276941349a546573b2ef73", size = 1168449, upload-time = "2026-08-08T18:27:58.365Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/95/9c/c510029fc6ef33a6275cd2c5d3cecd6613dfd6aa401d57c54f1c18852ccf/setuptools-84.0.0-py3-none-any.whl", hash = "sha256:51a52592b3b99e102b609654876bd65f19f999935166d1352678931132b0c670", size = 818216, upload-time = "2026-08-08T18:27:56.719Z" },
+]
+
 [[package]]
 name = "shellingham"
 version = "1.5.4"
@@ -2266,6 +2499,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/b7/ce/149a00dd41f10bc29e5921b496af8b574d8413afcd5e30dfa0ed46c2cc5e/six-1.17.0-py2.py3-none-any.whl", hash = "sha256:4721f391ed90541fddacab5acf947aa0d3dc7d27b2e1e8eda2be8970586c3274", size = 11050, upload-time = "2024-12-04T17:35:26.475Z" },
 ]
 
+[[package]]
+name = "smmap"
+version = "5.0.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/1f/ea/49c993d6dfdd7338c9b1000a0f36817ed7ec84577ae2e52f890d1a4ff909/smmap-5.0.3.tar.gz", hash = "sha256:4d9debb8b99007ae47165abc08670bd74cb74b5227dda7f643eccc4e9eb5642c", size = 22506, upload-time = "2026-03-09T03:43:26.1Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c1/d4/59e74daffcb57a07668852eeeb6035af9f32cbfd7a1d2511f17d2fe6a738/smmap-5.0.3-py3-none-any.whl", hash = "sha256:c106e05d5a61449cf6ba9a1e650227ecfb141590d2a98412103ff35d89fc7b2f", size = 24390, upload-time = "2026-03-09T03:43:24.361Z" },
+]
+
 [[package]]
 name = "stack-data"
 version = "0.6.3"
@@ -2303,6 +2545,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/9b/24/84ce997e8ae6296168a74d0d9c4dde572d90fb23fd7c0b219c30ff71e00e/tbb-2021.13.1-py3-none-win_amd64.whl", hash = "sha256:cbf024b2463fdab3ebe3fa6ff453026358e6b903839c80d647e08ad6d0796ee9", size = 286908, upload-time = "2024-08-07T15:09:05.677Z" },
 ]
 
+[[package]]
+name = "tenacity"
+version = "8.5.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a3/4d/6a19536c50b849338fcbe9290d562b52cbdcf30d8963d3588a68a4107df1/tenacity-8.5.0.tar.gz", hash = "sha256:8bc6c0c8a09b31e6cad13c47afbed1a567518250a9a171418582ed8d9c20ca78", size = 47309, upload-time = "2024-07-05T07:25:31.836Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d2/3f/8ba87d9e287b9d385a02a7114ddcef61b26f86411e121c9003eb509a1773/tenacity-8.5.0-py3-none-any.whl", hash = "sha256:b594c2a5945830c267ce6b79a166228323ed52718f30302c1359836112346687", size = 28165, upload-time = "2024-07-05T07:25:29.591Z" },
+]
+
 [[package]]
 name = "threadpoolctl"
 version = "3.6.0"
@@ -2596,6 +2847,20 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/bd/6e/95b0e537de1f4d4301f76f944642c6da50d1511cc7b3d64dc418a66c7509/wcwidth-0.8.1-py3-none-any.whl", hash = "sha256:f453740b1e4a4f3291faa37944c555d71056c4da08d59809b307ef4feba695c8", size = 323092, upload-time = "2026-06-08T05:57:21.413Z" },
 ]
 
+[[package]]
+name = "webdavclient3"
+version = "3.14.7"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "lxml" },
+    { name = "python-dateutil" },
+    { name = "requests" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/67/d8/ca3981053ed553363322f71745f543186b93439b6417f5d6ca91d4b4fec7/webdavclient3-3.14.7.tar.gz", hash = "sha256:6c04252b579bc015cec78081480c63eadf1030f382768248777c6203f059b3f5", size = 30836, upload-time = "2026-02-06T17:54:15.506Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/89/5e/0b1c2f494d03c4acbc44567fa68b954cd0fa3f21eb3f9528011da371f9b1/webdavclient3-3.14.7-py3-none-any.whl", hash = "sha256:a904381da8e3ae77b4ca9e11e05058d91a07704254d71c193c797f7c2fb15025", size = 22887, upload-time = "2026-02-06T17:54:14.068Z" },
+]
+
 [[package]]
 name = "xxhash"
 version = "3.7.0"