feat(WO72): gates require an anchor (protocol + convention or no gate)

recipes/ara-diac.yaml

@@ -14,6 +14,7 @@ stages:
     route: legacy:interscript-train/continue_oracle_train_hf.py
     params: {epochs: 2}
 gates:
+  anchor: run-047  # byt5-large first-alt gate line
   convention: row-mean
   quality: {wer_max: 3.5607}  # run-047 first-alt line; stretch 2.70 (QCRI frontier parity)
   parity: {mode: single-row, agree_min: 0.98, three_way: true}

recipes/fas-diac.yaml

@@ -9,6 +9,7 @@ stages:
     route: legacy:interscript-train/train_fas_diac_hf.py
     params: {epochs: 3, bs: 16, accum: 2, gate: 0.0052}
 gates:
+  anchor: TODO.sota/47-fas-diac-release.md (docs split, corpus-level)  # the 0.52% docs line
   convention: corpus-level
   quality: {cer_max: 0.0052}
   parity: {mode: single-row, agree_min: 0.98, three_way: true}

recipes/urd-diac.yaml

@@ -11,6 +11,7 @@ stages:
   - stage: decode_policy
     params: {rep_penalty: 1.2, no_repeat_ngram: 3}
 gates:
+  anchor: run-066  # urd-diac-base gate pass
   convention: row-mean
   quality: {cer_max: 0.045}
   parity: {mode: single-row, agree_min: 0.98, three_way: true}

recipes/urd-g2p.yaml

@@ -8,6 +8,7 @@ stages:
   - stage: sft
     route: legacy:interscript-train/train_urd_g2p_hf.py
 gates:
+  anchor: run-059  # urd-g2p-base gate pass
   convention: row-mean
   quality: {cer_max: 0.17}  # torch greedy 0.16 (artifact ≡ torch under the WO51 single-row contract)
   parity: {mode: single-row, agree_min: 0.98, three_way: true}

src/pipeline/contracts.py

@@ -116,9 +116,18 @@ class GateSpec(_Strict):
     quality: dict[str, float]
     convention: Literal["row-mean", "corpus-level"]  # the fas scar: a gate
     # number without its aggregation convention is not a gate
+    anchor: str  # the protocol/ledger section this gate line is anchored to
     parity: ParityGate
     license: bool = True
 
+    @field_validator("anchor")
+    @classmethod
+    def _anchor_nonempty(cls, v: str) -> str:
+        if not v.strip():
+            raise ValueError("gates: anchor must name the protocol/ledger "
+                             "section the gate line is anchored to")
+        return v
+
     @field_validator("quality")
     @classmethod
     def _metrics(cls, v: dict[str, float]) -> dict[str, float]:

tests/fixtures/recipes/urd-diac.yaml

@@ -11,6 +11,7 @@ stages:
   - stage: decode_policy
     params: {rep_penalty: 1.2, no_repeat_ngram: 3}
 gates:
+  anchor: run-047  # fixture anchor
   convention: row-mean
   quality: {cer_max: 0.045}
   parity: {mode: single-row, agree_min: 0.98, three_way: true}

tests/test_audit_fixes.py

@@ -63,7 +63,7 @@ def _tmp_registry(tmp_path: Path, license_value: str) -> Path:
         "gates": {
             "quality": {"cer_max": 0.1},
             "parity": {"agree_min": 0.98},
-            "convention": "row-mean",
+            "convention": "row-mean", "anchor": "run-047",
         },
     }
     (tmp_path / "corpora").mkdir()

tests/test_pipeline_contracts.py

@@ -45,6 +45,7 @@ def test_stage_needs_route_or_params() -> None:
         "model": "byt5-small",
         "stages": [{"stage": "sft"}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                    "parity": {"agree_min": 0.98}},
     }
     with pytest.raises(ValueError, match="route"):
@@ -59,6 +60,7 @@ def test_no_llm_teachers_for_diacritization() -> None:
         "model": "byt5-small",
         "stages": [{"stage": "distill", "route": "legacy:llm_teacher.py"}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                    "parity": {"agree_min": 0.98}},
     }
     with pytest.raises(ValueError, match="llm"):
@@ -73,6 +75,7 @@ def test_corpus_ref_pin_format() -> None:
         "model": "byt5-small",
         "stages": [{"stage": "sft", "params": {"epochs": 1}}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                    "parity": {"agree_min": 0.98}},
     }
     with pytest.raises(ValueError, match="pin"):
@@ -113,6 +116,7 @@ def test_recipe_refs_resolve_into_registry(tmp_path: Path) -> None:
         "model": "byt5-small",
         "stages": [{"stage": "sft", "params": {"epochs": 1}}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                    "parity": {"agree_min": 0.98}},
     }
     import yaml  # noqa: PLC0415
@@ -139,6 +143,7 @@ def test_unverified_license_is_blocking(tmp_path: Path) -> None:
         "model": "byt5-small",
         "stages": [{"stage": "sft", "params": {"epochs": 1}}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                    "parity": {"agree_min": 0.98}},
     }
     import yaml  # noqa: PLC0415
@@ -192,6 +197,7 @@ def test_restricted_but_read_license_blocks(tmp_path: Path) -> None:
         "model": "byt5-small",
         "stages": [{"stage": "sft", "params": {"epochs": 1}}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                    "parity": {"agree_min": 0.98}},
     }
     (tmp_path / "corpora").mkdir()
@@ -206,3 +212,28 @@ def test_load_corpus_roundtrip() -> None:
     c = load_corpus(FIXTURES / "corpora" / "urd-dia.yaml")
     assert c.id == "urd-dia"
     assert "train" in c.splits and "test" in c.splits
+
+
+def test_gate_block_requires_anchor_and_convention() -> None:
+    base = {
+        "id": "xx-anchor",
+        "task": "transliteration",
+        "corpora": {"train": "xx-src@v1", "test": "xx-src@v1#test"},
+        "model": "byt5-small",
+        "stages": [{"stage": "sft", "params": {"epochs": 1}}],
+    }
+    no_anchor = {
+        **base,
+        "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "parity": {"agree_min": 0.98}},
+    }
+    with pytest.raises(ValueError, match="anchor"):
+        RecipeSpec.model_validate(no_anchor)
+
+    no_convention = {
+        **base,
+        "gates": {"quality": {"cer_max": 0.1}, "anchor": "run-047",
+                   "parity": {"agree_min": 0.98}},
+    }
+    with pytest.raises(ValueError, match="convention"):
+        RecipeSpec.model_validate(no_convention)

tests/test_pipeline_plan_gate.py

@@ -114,6 +114,7 @@ def test_license_gate_blocks_publish(tmp_path: Path) -> None:
         "model": "byt5-small",
         "stages": [{"stage": "sft", "params": {"epochs": 1}}],
         "gates": {"quality": {"cer_max": 0.1}, "convention": "row-mean",
+                   "anchor": "run-047",
                   "parity": {"agree_min": 0.98}},
     }
     (tmp_path / "corpora").mkdir()

tests/test_pipeline_registry.py

@@ -33,6 +33,7 @@ def _recipe_body() -> dict:
         "gates": {
             "quality": {"cer_max": 0.1},
             "convention": "row-mean",
+            "anchor": "run-047",
             "parity": {"agree_min": 0.98},
         },
     }