refactor(evidence): unify formulas with typed evidence

harness/tests/test_evidence_formula_migration.py

@@ -0,0 +1,26 @@
+"""Migration boundary: legacy formulas become curated evidence or session proposals."""
+
+from tht.evidence import formula_store
+from tht.evidence.formula_store import ConceptFormula
+
+
+def test_reviewed_formula_migration_has_deterministic_provenance_hash():
+    formula = ConceptFormula(
+        concept="fascia pediatrica",
+        columns=["clinical.patient.birth_date"],
+        sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        status="reviewed",
+        sources=["Regola clinica approvata dal gruppo pediatrico."],
+    )
+
+    first = formula_store.legacy_formula_to_curated(
+        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+    )
+    second = formula_store.legacy_formula_to_curated(
+        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+    )
+
+    assert first is not None
+    assert second is not None
+    assert first.provenance.source_sha256 == second.provenance.source_sha256
+    assert first.provenance.source_sha256.startswith("sha256:")

harness/tests/test_formula.py

@@ -8,7 +8,13 @@ part of the schema-linking artifact. The store is frontmatter-YAML + SQL body.
 """
 
 
+from datetime import UTC, datetime
+
+from tht.decisions import DecisionRecord
+from tht.evidence import formula_store
 from tht.evidence.formula_store import ConceptFormula, retrieve_formula, save_formula
+from tht.evidence.session import project_session
+from tht.session.models import SchemaLinking
 
 
 def test_formula_retrieval_by_concept(tmp_path):
@@ -82,3 +88,63 @@ def test_concept_formula_default_status(tmp_path):
     f = ConceptFormula(concept="x", columns=["c"], sql="SELECT 1")
     assert f.status == "draft"  # not yet reviewed
     assert f.sources == []
+
+
+def test_reviewed_legacy_formula_becomes_curated_formula_with_stable_provenance():
+    formula = ConceptFormula(
+        concept="fascia pediatrica",
+        columns=["clinical.patient.birth_date"],
+        sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        status="reviewed",
+        sources=["Regola clinica approvata dal gruppo pediatrico."],
+    )
+
+    migrated = formula_store.legacy_formula_to_curated(
+        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+    )
+
+    assert migrated is not None
+    assert migrated.id == "evidence:fascia-pediatrica"
+    assert migrated.title == "Fascia pediatrica"
+    assert migrated.kind == "formula"
+    assert migrated.payload.concept == formula.concept
+    assert migrated.payload.columns == ("clinical.patient.birth_date",)
+    assert migrated.payload.sql == formula.sql
+    assert migrated.provenance.source_file == "source/formulas/fascia-pediatrica-1.sql.md"
+    assert migrated.provenance.supporting_excerpts == tuple(formula.sources)
+    assert migrated.review_items == ()
+    assert formula_store.legacy_formula_to_curated(
+        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+    ).id == migrated.id
+
+
+def test_session_formula_proposal_is_versioned_and_is_not_published_evidence(tmp_path):
+    linking = SchemaLinking(
+        question="Conta i pazienti pediatrici",
+        concept_formulas=[{
+            "concept": "fascia pediatrica",
+            "columns": ["clinical.patient.birth_date"],
+            "sql": "CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+            "sources": ["Sintetizzata nella sessione"],
+        }],
+    )
+    decisions = [DecisionRecord(
+        seq=9,
+        ts=datetime(2026, 8, 25, tzinfo=UTC),
+        type="concept_formula_approved",
+        subject="phase:4",
+        detail="fascia pediatrica",
+    )]
+
+    projected = project_session(decisions, linking, tmp_path / "evidence")
+
+    assert projected == [{
+        "schema_version": 1,
+        "kind": "formula_proposal",
+        "publication": "session_only",
+        "concept": "fascia pediatrica",
+        "columns": ["clinical.patient.birth_date"],
+        "sql": "CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        "sources": ["Sintetizzata nella sessione"],
+        "decision_seq": 9,
+    }]

harness/tests/test_formula_wiring.py

@@ -5,8 +5,16 @@ Completes the formula layer beyond the store: the `auto` status the spec require
 and the guarantee that load_evidence_dir does NOT choke on *.sql.md formula files when
 they live under the evidence root.
 """
+import json
+from types import SimpleNamespace
+
+from typer.testing import CliRunner
+
+from tht.cli import app, search_cmd
+from tht.evidence import formula_store
 from tht.evidence.formula_store import ConceptFormula, save_formula, search_formulas
 from tht.evidence.model import EvidenceDoc, load_evidence_dir
+from tht.evidence.search import EvidenceSearchOutcome
 
 
 def test_status_auto_is_valid():
@@ -35,3 +43,77 @@ def test_load_evidence_dir_skips_formula_files(tmp_path):
     ids = [d.id for d in docs]
     assert ids == ["ev1"]  # the .sql.md formula file is skipped, no crash
     assert all(isinstance(d, EvidenceDoc) for d in docs)
+
+
+def test_unreviewed_legacy_formulas_cannot_become_curated_evidence():
+    for status in ("auto", "draft"):
+        formula = ConceptFormula(
+            concept="fascia pediatrica",
+            columns=["clinical.patient.birth_date"],
+            sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+            status=status,
+        )
+        assert formula_store.legacy_formula_to_curated(
+            formula, legacy_path="formulas/pediatric-1.sql.md",
+        ) is None
+
+
+def test_formula_search_uses_typed_evidence_with_a_formula_constraint():
+    class Searcher:
+        vector_generation = "gen:" + "a" * 32
+
+        def __init__(self):
+            self.calls = []
+
+        def search(self, embedding, **kwargs):
+            self.calls.append((embedding, kwargs))
+            return [SimpleNamespace(
+                id="fragment:formula", similarity=0.9, content="formula excerpt",
+                title="Fascia pediatrica", metadata={
+                    "evidence_id": "evidence:fascia-pediatrica",
+                    "evidence_kind": "formula",
+                    "document_id": "doc:formula",
+                    "ordinal": 0,
+                    "source_uri": "file:///curated/formula/fascia-pediatrica.md",
+                    "provenance": {},
+                },
+            )]
+
+    class Embedder:
+        def embed_query(self, query):
+            return [0.25]
+
+    searcher = Searcher()
+    outcome = search_cmd.search_formula_evidence(
+        "fascia pediatrica", searcher=searcher, embedder=Embedder(), top=3,
+    )
+
+    assert outcome.status == "available"
+    assert [result.evidence_id for result in outcome.results] == ["evidence:fascia-pediatrica"]
+    assert searcher.calls[0][1]["metadata_filter"]["required_kinds"] == ["formula"]
+
+
+def test_formula_search_json_is_pristine_while_human_output_warns_about_legacy_store(monkeypatch):
+    monkeypatch.setattr(search_cmd, "_load_config_or_exit", lambda _path: SimpleNamespace(embeddings=object()))
+    monkeypatch.setattr(search_cmd, "workspace_id_for_config", lambda _cfg, _path: "workspace-a")
+    monkeypatch.setattr(search_cmd, "search_formula_evidence", lambda *args, **kwargs: EvidenceSearchOutcome(
+        "available", "gen:" + "a" * 32,
+    ))
+    monkeypatch.setattr("tht.cli.vector_cmd.require_vector_cfg", lambda _cfg: None)
+    monkeypatch.setattr("tht.cli.vector_cmd.open_searcher", lambda _cfg: object())
+    monkeypatch.setattr("tht.cli.vector_cmd.make_embedder", lambda _cfg: object())
+    monkeypatch.setattr("tht.evidence.active_searcher", lambda *args, **kwargs: object())
+    monkeypatch.setattr("tht.evidence.validate_corpus_workspace", lambda _cfg, _workspace: None)
+
+    json_result = CliRunner().invoke(
+        app, ["search", "find", "fascia pediatrica", "--kind", "formula", "--top", "1", "--json"],
+    )
+    human_result = CliRunner().invoke(
+        app, ["search", "find", "fascia pediatrica", "--kind", "formula", "--top", "1"],
+    )
+
+    assert json_result.exit_code == 0
+    assert json.loads(json_result.stdout) == []
+    assert "ATTENZIONE" not in json_result.stdout
+    assert human_result.exit_code == 0
+    assert "ATTENZIONE" in human_result.output

harness/tests/test_pi_skill_projection.py

@@ -9,7 +9,7 @@ from tht.pi_skill_projection import (
     render_projection,
 )
 
-BASELINE_SHA256 = "bb6daa6fe83d22f5e701025c5334d72ec9eb35349e90d24cb9d7f6290d0fecfe"
+BASELINE_SHA256 = "62bfa0dbc1179b43b2d80dc48155a6119421488a1ffc160e9c664f1fe280ce52"
 
 
 def test_modular_pi_skill_renders_the_byte_identical_approved_projection():
@@ -23,6 +23,7 @@ def test_modular_pi_skill_renders_the_byte_identical_approved_projection():
         ("{{DISAMBIGUATION_REWRITING_INSTRUCTIONS}}", "disambiguation/phase-3.md"),
         ("{{MEMORY_SOLVED_SEARCH_F4}}", "memory/solved-search-f4.md"),
         ("{{DISAMBIGUATION_SCHEMA_GROUNDING}}", "disambiguation/schema-grounding.md"),
+        ("{{EVIDENCE_FORMULA_PROPOSALS}}", "evidence/formula-proposals.md"),
         ("{{MEMORY_SOLVED_SEARCH_F6}}", "memory/solved-search-f6.md"),
         ("{{MEMORY_SOLVED_SEARCH_F7}}", "memory/solved-search-f7.md"),
         ("{{MEMORY_PROMOTION_F8}}", "memory/phase-8-promotion.md"),