# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## What this is

Tatami ("Tool for Arranging Tasty Appointments, Meetings & Invitations") generates a "running dinner" masterplan: given a list of participants (with home addresses), it groups them into hosting groups for starter/main/dessert courses, assigns which groups visit which hosts for each course, and orders things to minimize travel time (by bike, via the Google Maps Routes API), finishing at a shared after-party location.

The actively developed package is `src/tatami/`. `running_dinner/running_dinner.py` is a legacy, pre-package standalone script (German-language field names, brute-force search over team pairings) kept for reference — it is not wired into the `tatami` package and uses a different/older Google Maps API (the legacy Distance Matrix API vs. the new Routes API used in `traveltimes.py`). Don't assume code or conventions from `running_dinner.py` apply to `src/tatami/`.

## Commands

This project uses `uv` for dependency management (Python >=3.13).

- Install deps: `uv sync`
- Run the masterplan script: `uv run python -m tatami.tatami_masterplan` (requires `test-config.csv` in the working directory and `GOOGLE_MAPS_API_KEY` set; optionally set `GOOGLE_SHEETS_CREDENTIALS_FILE` in `.env` + `spreadsheet_id` in the saved plan to also export to a shared Google Sheet — see `.env.example`). The first run computes a plan and saves it to `masterplan.json` (override via `PLAN_FILE`); later runs load that file instead of recomputing, so it's the place to hand-edit groups/courses/contacts/`spreadsheet_id` between runs.
- Lint: `uv run ruff check`
- Format: `uv run ruff format`
- Type check: `uv run mypy --allow-redefinition src/` (mypy is configured to treat untyped imports as errors except where ignored)
- Test: `uv run pytest` (offline suite, HTTP mocked); `uv run pytest -m e2e` for the opt-in live-API tests
- Pre-commit runs ruff check, ruff format, and mypy automatically (see `.pre-commit-config.yaml`); install hooks with `uv run pre-commit install` if working interactively.

## Architecture

The pipeline (see `src/tatami/tatami_masterplan.py` `__main__` block) is:

1. **Load participants** — `load_csv_to_participants` reads a tab-separated CSV (`name`, `address`, `phone`, `kitchen_size`, `allergies`) into `Participant` objects (`src/tatami/classes.py`).
2. **Fetch travel times** — `traveltimes.get_participant_distance_matrix` calls the Google Routes API (`GOOGLE_MAPS_API_KEY` env var required) to build a full pairwise duration matrix between all participant addresses plus the after-party address, indexed by participant UUID.
3. **Build masterplan** — `compute_masterplan_groups` (the live-object core; `get_masterplan` is a thin wrapper around it that returns plain dicts instead):
   - Splits participants into `hosts` (one per group, `len(participants)//6` groups of 3 courses each) and `semi_hosts` (non-hosting members assigned round-robin into existing groups), ranked by each participant's `get_after_party_time` (kitchen size penalty + distance to after-party).
   - Reduces the full distance matrix to just host-to-host distances (`reduce_distance_matrix`), adding each host's kitchen-size penalty into their row.
   - Runs `run_simulated_annealing` / `simulated_annealing` (Boltzmann-style annealing over `itertools.permutations` of group order — note this is brute-force over all permutations per iteration, so it only scales to a small number of groups) to find a low-travel-time ordering of groups.
   - `assign_courses` assigns each group a course (starter/main/dessert cycling) and, via `get_courses`, determines which other groups host it for each course. The rotation is a resolvable "Latin-square"/transversal design (`_rotation_hosts`): slots form a `k x 3` grid (`k = n // 3` groups per course), each course is a parallel class partitioning all groups into transversal tables of three (one starter/main/dessert each), so for any `n >= 9` no two groups ever meet more than once. `n = 3`/`6` are combinatorially impossible and fall back to a degenerate same-row rotation. (This replaced an earlier fixed `+1`/`-4` cyclic offset that produced repeat meetings for group counts like 9.)
4. `get_masterplan` returns two lists of plain dicts (`group.dict()`, `participant.dict()`) suitable for serialization; `compute_masterplan_groups` returns the live `Group`/`Participant` objects, which is what the `Plan`/sheet export step needs (`.hosts`, `.get_guests(...)`).
5. **Wrap in a `Plan` and save** (`src/tatami/plan.py`) — `__main__` bundles the computed `groups` + `after_party_group` with the event config (`course_times`, `organizer_contacts`, `info_text`, `spreadsheet_id`) into a `Plan` and calls `plan.save(PLAN_FILE)`. On the next invocation, if that file exists, `Plan.load()` reads it back instead of recomputing — this is the save/reload/edit path: hand-edit `masterplan.json` (move a member between groups, change a course, fill in `spreadsheet_id`, ...) and rerun to pick up the edit without hitting the Routes API again.
6. **Export to Google Sheets (optional)** — if the loaded/built `Plan.spreadsheet_id` is set, `__main__` calls `sheets_export.export_masterplan_to_sheet` to populate a pre-existing, pre-shared spreadsheet with an Overview tab and one tab per group. This is opt-in and never sends anything directly to participants — the organizer still shares the sheet link manually.

### Core domain model (`src/tatami/classes.py`, `src/tatami/plan.py`)

`Participant` and `Group` are pydantic `BaseModel`s (so they support `model_dump()`/`model_dump_json()`/`model_validate_json()` directly); `Plan` wraps the whole thing for persistence.

- `Participant`: a person with an address, phone, kitchen size (0–10, used as a "willingness/suitability to host" proxy via `get_penalty`, which adds travel-time-equivalent minutes for smaller kitchens), and allergies.
- `Group`: a hosting unit with a `main_member` (a property resolved from `main_member_uuid` against `members` — used as the group's representative location for all distance lookups; other members' addresses are not used for travel calculations), a `course`, and a `hosts` property (the groups that host *this* group across the evening, kept sorted starter→main→dessert via `sort_hosts`). `get_total_time` sums travel + penalty across this group's full route (its hosts, then the after-party).
  - `hosts` references *other* `Group`s and these references are genuinely cyclic (a group's hosts can host it back), so they can't be embedded directly in JSON. The live `hosts` list is a private, non-persisted attribute set via `set_hosts()`/`add_host()`; the persisted field is `host_uuids` (kept in sync automatically). After `Plan.load()`, `Group.resolve_hosts()` re-links `hosts` from `host_uuids` against the sibling groups in the same `Plan` — call it yourself if you ever construct `Group`s outside of a `Plan` and need `.hosts` populated from `host_uuids`.
- `Plan` (`plan.py`): bundles `groups`, `after_party_group`, and the event-wide config (`course_times`, `organizer_contacts`, `info_text`, `spreadsheet_id`). `Plan.save(path)` / `Plan.load(path)` round-trip the whole thing to/from JSON; `plan.participants` is a derived property (flattened, deduplicated `group.members` across all groups + the after party), not a separately stored field, so editing a participant's data in a group's `members` is the single source of truth.
- Distance/time lookups throughout the codebase are keyed by `Participant.uuid` (host groups are addressed via `main_member.uuid`), not by name — when adding new matrix operations, index by uuid for consistency with `traveltimes.py` and `classes.py`.

### Travel times (`src/tatami/traveltimes.py`)

- Wraps the Google Routes API `computeRouteMatrix` endpoint. Raises at import time if `GOOGLE_MAPS_API_KEY` is unset.
- `get_distance_matrix` returns a square `DataFrame` of `pd.Timedelta` (or float meters) indexed/columned by integer position; `get_participant_distance_matrix` relabels both axes to participant UUIDs.
- `reduce_distance_matrix` is the bridge between the full participant-level matrix and the group-level matrix used by the annealing step.

### Sheet export (`src/tatami/sheets_export.py`)

- Wraps `gspread` (Google Sheets API). Unlike `traveltimes.py`, validation of `GOOGLE_SHEETS_CREDENTIALS_FILE` happens lazily inside `load_sheets_client`, not at import time — this feature is optional/opt-in, so importing the module must not require the env var.
- `export_masterplan_to_sheet` expects an existing `gspread.Spreadsheet` (organizer pre-creates it and shares Editor access with the service account's `client_email` once) and rewrites it idempotently: an "Overview" tab listing every participant, plus one tab per group named after all of that group's members joined with `&` (deduplicated; see `_group_titles` — deliberately not a single member's name, since a group can have several members) showing that group's course, route (their `hosts`, sorted starter→main→dessert, plus the after-party), and their guests for the course they host (`Group.get_guests`). Course start times are a fixed dict passed in by the caller (`COURSE_TIMES` in `tatami_masterplan.py`) — the dinner runs on a synchronized schedule, not on travel-time-derived timing.
- The Overview tab also gets three caller-supplied sections below the group assignments, each rendered as-is (no computation) and each skippable: a "Meal Times" table built straight from `course_times`, a "Support Contacts" table from the optional `organizer_contacts: list[tuple[str, str]]` argument, and a free-text "Info" block from the optional `info_text: str` argument (split into one row per line). All three are configured in `tatami_masterplan.py` (`COURSE_TIMES`, `ORGANIZER_CONTACTS`, `INFO_TEXT`) and just passed through — `sheets_export.py` has no event-specific content baked in.
- Tabs left over from a previous run with a different group count are deleted so reruns don't accumulate stale tabs.