fix(evidence): preserve legacy formula migration boundaries

harness/tht/evidence/formula_store.py

@@ -8,11 +8,12 @@ from __future__ import annotations
 
 import hashlib
 import re
+from dataclasses import dataclass
 from pathlib import Path
 from typing import Literal
 
 import yaml
-from pydantic import BaseModel
+from pydantic import BaseModel, ValidationError
 
 from tht.evidence.canonical import CuratedEvidence
 
@@ -56,6 +57,16 @@ class ConceptFormula(BaseModel):
         return cls.model_validate({**meta, "sql": body.strip("\n")})
 
 
+@dataclass(frozen=True)
+class LegacyFormulaMigrationFailure:
+    """A reviewed legacy formula that must be resolved manually before publication."""
+
+    code: Literal["legacy_formula_requires_manual_review"]
+    legacy_path: str
+    formula: ConceptFormula
+    problems: tuple[str, ...]
+
+
 def _next_path(root: Path, slug: str) -> Path:
     """First free <slug>-<n>.sql.md path under root (n starts at 1)."""
     root.mkdir(parents=True, exist_ok=True)
@@ -110,7 +121,7 @@ def legacy_formula_to_curated(
     formula: ConceptFormula,
     *,
     legacy_path: str,
-) -> CuratedEvidence | None:
+) -> CuratedEvidence | LegacyFormulaMigrationFailure | None:
     """Convert one reviewed legacy formula into its deterministic curated counterpart.
 
     Drafts and model-generated formulas have no global publication status. Their
@@ -123,23 +134,38 @@ def legacy_formula_to_curated(
     )
     source_sha256 = hashlib.sha256(formula.dump().encode("utf-8")).hexdigest()
     source_file = legacy_path if legacy_path.startswith("source/") else f"source/{legacy_path}"
-    return CuratedEvidence.model_validate({
-        "schema_version": 1,
-        "id": f"evidence:{formula._slug}",
-        "title": formula.concept[:1].upper() + formula.concept[1:],
-        "kind": "formula",
-        "purposes": ["schema_linking", "sql_generation"],
-        "applies_to": {"concepts": [formula.concept], "columns": formula.columns},
-        "language": "it",
-        "provenance": {
-            "source_file": source_file,
-            "source_sha256": f"sha256:{source_sha256}",
-            "supporting_excerpts": source_notes,
-        },
-        "review_items": [],
-        "payload": {
-            "concept": formula.concept,
-            "columns": formula.columns,
-            "sql": formula.sql,
-        },
-    })
+    # The legacy path is the immutable identity of this unit during migration. Keeping
+    # its full digest avoids a duplicate public ID when the same concept has reviewed
+    # competing formulas, while retaining the readable concept slug as the prefix.
+    legacy_identity = hashlib.sha256(legacy_path.encode("utf-8")).hexdigest()
+    try:
+        return CuratedEvidence.model_validate({
+            "schema_version": 1,
+            "id": f"evidence:{formula._slug}-{legacy_identity}",
+            "title": formula.concept[:1].upper() + formula.concept[1:],
+            "kind": "formula",
+            "purposes": ["schema_linking", "sql_generation"],
+            "applies_to": {"concepts": [formula.concept], "columns": formula.columns},
+            "language": "it",
+            "provenance": {
+                "source_file": source_file,
+                "source_sha256": f"sha256:{source_sha256}",
+                "supporting_excerpts": source_notes,
+            },
+            "review_items": [],
+            "payload": {
+                "concept": formula.concept,
+                "columns": formula.columns,
+                "sql": formula.sql,
+            },
+        })
+    except ValidationError as error:
+        return LegacyFormulaMigrationFailure(
+            code="legacy_formula_requires_manual_review",
+            legacy_path=legacy_path,
+            formula=formula,
+            problems=tuple(sorted(
+                ".".join(str(part) for part in issue["loc"])
+                for issue in error.errors()
+            )),
+        )

harness/tht/evidence/session.py

@@ -64,11 +64,17 @@ def _formula_proposals(decisions: list["DecisionRecord"], linking: "SchemaLinkin
     A proposal remains in the persisted session artifact until a separate curator
     imports, reviews, and publishes it in the workspace repository.
     """
-    approved = {
-        decision.detail: decision.seq
+    retracted = {
+        decision.retracts
         for decision in decisions
-        if decision.type == "concept_formula_approved" and decision.detail
+        if decision.type == "decision_retracted" and decision.retracts is not None
     }
+    latest_f4_decision = {}
+    for decision in sorted(decisions, key=lambda item: item.seq):
+        if decision.seq in retracted or decision.subject != "phase:4" or not decision.detail:
+            continue
+        if decision.type in {"concept_formula_approved", "concept_formula_rejected"}:
+            latest_f4_decision[decision.detail] = decision
     proposals = []
     for formula in linking.concept_formulas:
         if not isinstance(formula, dict):
@@ -83,6 +89,9 @@ def _formula_proposals(decisions: list["DecisionRecord"], linking: "SchemaLinkin
         evidence_id = formula.get("evidence_id", formula.get("id", ""))
         if isinstance(evidence_id, str) and evidence_id.startswith("evidence:"):
             continue
+        decision = latest_f4_decision.get(concept)
+        if decision is None or decision.type != "concept_formula_approved":
+            continue
         proposals.append({
             "schema_version": 1,
             "kind": "formula_proposal",
@@ -91,7 +100,7 @@ def _formula_proposals(decisions: list["DecisionRecord"], linking: "SchemaLinkin
             "columns": columns,
             "sql": sql,
             "sources": formula.get("sources", []),
-            "decision_seq": approved.get(concept),
+            "decision_seq": decision.seq,
         })
     return proposals