refactor(evidence): unify formulas with typed evidence

harness/tht/evidence/formula_store.py

@@ -1,17 +1,12 @@
-"""SQL concept->formula evidence store (spec D14b, §4.7).
+"""Legacy ConceptFormula reader and one-way migration into Curated Evidence.
 
-A concept (e.g. 'fascia pediatrica', 'ablazione') maps to a reusable SQL formula
-(a CASE WHEN ...) that derives it from physical columns. These are reviewable
-units: the gate surfaces a candidate formula, the reviewer approves or rejects it
-(decision types concept_formula_approved / concept_formula_rejected), and approved
-formulas travel with the schema-linking artifact.
-
-Storage: one file per formula, frontmatter YAML + SQL body (same shape as
-EvidenceDoc.parse). Directory layout: <root>/formulas/<slug>-<n>.sql.md.
-Retrieve is by concept (may return several, e.g. competing drafts vs reviewed).
+The ``formulas/*.sql.md`` store is retained only for the migration window. Runtime
+lookup uses typed, published ``kind=formula`` Evidence instead. A session reviewer
+may still approve a formula locally; that is a proposal, not publication.
 """
 from __future__ import annotations
 
+import hashlib
 import re
 from pathlib import Path
 from typing import Literal
@@ -19,6 +14,8 @@ from typing import Literal
 import yaml
 from pydantic import BaseModel
 
+from tht.evidence.canonical import CuratedEvidence
+
 FORMULAS_SUBDIR = "formulas"
 _SUFFIX_RE = re.compile(r"^(.*?)-(\d+)\.sql\.md$")
 
@@ -104,8 +101,45 @@ def retrieve_formula(root: Path | str, concept: str) -> list[ConceptFormula]:
 
 
 def search_formulas(root: Path | str, query: str) -> list[ConceptFormula]:
-    """Formulas whose concept contains `query` (case-insensitive). Used by
-    `tht search find --kind formula` (D14b retrieval, §4.7.2): the reviewer searches a
-    concept term and gets the candidate formulas to approve before they reach the CTE."""
+    """Read legacy formulas for migration tooling only (case-insensitive concept match)."""
     q = query.strip().lower()
     return [f for f in _load_all(Path(root)) if q in f.concept.lower()]
+
+
+def legacy_formula_to_curated(
+    formula: ConceptFormula,
+    *,
+    legacy_path: str,
+) -> CuratedEvidence | None:
+    """Convert one reviewed legacy formula into its deterministic curated counterpart.
+
+    Drafts and model-generated formulas have no global publication status. Their
+    caller must project them as session-local Formula proposals instead.
+    """
+    if formula.status != "reviewed":
+        return None
+    source_notes = tuple(formula.sources) or (
+        "Legacy formula migrated without a recorded provenance note.",
+    )
+    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,
+        },
+    })

harness/tht/evidence/session.py

@@ -55,7 +55,45 @@ def project_session(
             "esito": "accettata" if decision.type == "evidence_accepted" else "scartata",
             "decision_seq": decision.seq,
         }
-    return list(entries.values())
+    return list(entries.values()) + _formula_proposals(decisions, linking)
+
+
+def _formula_proposals(decisions: list["DecisionRecord"], linking: "SchemaLinking") -> list[dict]:
+    """Project locally approved F4 formulas without representing them as Evidence.
+
+    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
+        for decision in decisions
+        if decision.type == "concept_formula_approved" and decision.detail
+    }
+    proposals = []
+    for formula in linking.concept_formulas:
+        if not isinstance(formula, dict):
+            continue
+        concept = formula.get("concept")
+        sql = formula.get("sql")
+        columns = formula.get("columns")
+        if not isinstance(concept, str) or not isinstance(sql, str) or not isinstance(columns, list):
+            continue
+        # A referenced published Formula Evidence is already represented by its
+        # Evidence receipt/citation, so it must not be reintroduced as a proposal.
+        evidence_id = formula.get("evidence_id", formula.get("id", ""))
+        if isinstance(evidence_id, str) and evidence_id.startswith("evidence:"):
+            continue
+        proposals.append({
+            "schema_version": 1,
+            "kind": "formula_proposal",
+            "publication": "session_only",
+            "concept": concept,
+            "columns": columns,
+            "sql": sql,
+            "sources": formula.get("sources", []),
+            "decision_seq": approved.get(concept),
+        })
+    return proposals
 
 
 @dataclass(frozen=True)