Reject stage2_model.stage1_context = 'sampled' as unimplemented (issues.md Issue 1)

trainers.py unconditionally trains stage 2 against the ground-truth
stage-1 output (stage1_ctx = x1_s1.detach()), but 'sampled' was accepted
by validate_config, stored in config.toml and the checkpoint's
model_config, and silently trained identically to 'truth' — mislabeling
every downstream artifact for a run launched with
--stage2-stage1-context sampled. Mirrors the existing stop_token
validate_config pattern. User chose the immediate fix (reject loudly)
over the proper fix (actually implement sampled context), which is
scoped to Issue 16.

Also updates the _KNOWN_UNUSED reason for stage2_model.stage1_context
(added by Issue 5's consumed-keys audit) to reflect that the value is
now rejected rather than silently accepted, and drops the now-invalid
--stage2-stage1-context sampled case from test_stage2_only_knobs (a
full CLI invocation) — that flag's plumbing is still covered at the
overrides-dict level by test_overrides_from_flags_stage2_only_knobs.

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

giant/config.py

@@ -1256,6 +1256,14 @@ def validate_config(cfg: dict) -> None:
             "(standalone stage-2 evaluation only, never for rollout)"
         )
 
+    if _get_path(cfg, "stage2_model.stage1_context") == "sampled":
+        raise ValueError(
+            "stage2_model.stage1_context = 'sampled' is accepted by the schema "
+            "but not implemented — trainers.py always trains stage 2 against "
+            "the ground-truth stage-1 output; use 'truth' (default) instead "
+            "(see issues.md Issue 16 for the planned implementation)"
+        )
+
     if (
         _get_path(cfg, "stage2_model.n_sec.mode") == "truth"
         and _get_path(cfg, "stage1_model.active")

tests/test_cli_train_overrides.py

@@ -41,6 +41,10 @@ def test_stage_prefixed_generator_overrides_shared_mode(monkeypatch, tmp_path):
 
 
 def test_stage2_only_knobs(monkeypatch, tmp_path):
+    # --stage2-stage1-context is exercised separately at the overrides-dict
+    # level (test_overrides_from_flags_stage2_only_knobs in test_config.py):
+    # its only non-default value, "sampled", is rejected by validate_config
+    # (issues.md Issue 1), so it can't appear in a full CLI invocation here.
     cfg = _invoke_and_capture_cfg(
         monkeypatch,
         tmp_path,
@@ -53,15 +57,12 @@ def test_stage2_only_knobs(monkeypatch, tmp_path):
             "32",
             "--stage2-context-dim",
             "16",
-            "--stage2-stage1-context",
-            "sampled",
         ],
     )
     assert cfg["stage2_model"]["decoder"] == "one_shot"
     assert cfg["stage2_model"]["k_max"] == 8
     assert cfg["stage2_model"]["hidden_dim"] == 32
     assert cfg["stage2_model"]["context_dim"] == 16
-    assert cfg["stage2_model"]["stage1_context"] == "sampled"
     # untouched stage1 defaults
     assert cfg["stage1_model"]["hidden_dim"] == 256
 

tests/test_config.py

@@ -687,6 +687,15 @@ def test_validate_config_stop_token_not_implemented():
         assert "stop_token" in str(e)
 
 
+def test_validate_config_stage1_context_sampled_not_implemented():
+    cfg = _cfg_with(**{"stage2_model.stage1_context": "sampled"})
+    try:
+        gconfig.validate_config(cfg)
+        assert False, "expected ValueError"
+    except ValueError as e:
+        assert "sampled" in str(e)
+
+
 def test_validate_config_n_sec_truth_rejected_for_rollout_capable_checkpoint():
     """'n_sec.mode = "truth" is invalid for a rollout-capable checkpoint' —
     both stages active means giant rollout

tests/test_config_consumed_keys.py

@@ -55,8 +55,10 @@ _EXCLUDED_FILES = ("giant/model/_legacy.py",)
 _KNOWN_UNUSED = {
     "stage2_model.stage1_context": (
         "issues.md Issue 1 — trainers.py hardcodes stage1_ctx to the "
-        "ground-truth stage-1 output; 'sampled' is accepted and stored but "
-        "never read"
+        "ground-truth stage-1 output; 'sampled' is now rejected loudly by "
+        "validate_config (not silently accepted), but the key still isn't "
+        "read by any build/train consumer file since only 'truth' can pass "
+        "validation — see Issue 16 for the real implementation"
     ),
     "stage1_model.wgan.critic_hidden_dim": (
         "issues.md Issue 2 — build_critics always sizes the critic off the generator's own hidden_dim, never this key"