feat(fase1): spike D - verdictstore + expel retrieval [skip-docs]

docs/fase1-spikes/findings-d.md

@@ -0,0 +1,47 @@
+# Spike D findings — VerdictStore + ExpeL retrieval (B2)
+
+**Assumption:** retrieval surfaces a relevant *prior* verdict for a similar new proposal —
+the substrate for the framework's learning loop.
+
+## Result — CONFIRMED (deterministic, no endpoint)
+
+A minimal in-memory `VerdictStore` holds 12 synthetic verdicts seeded from the reference
+domain's cost codes, measure types, magnitudes, and decisions (B2's "10–20").
+
+**Similarity is structural, not textual (reviewer refinement #2):** a weighted score over
+*structured* fields — Jaccard on the affected cost-code set (0.60) + a `measure_type`
+match (0.25) + a magnitude-bucket match on the claimed saving (0.15). Raw description text
+is **deliberately ignored**.
+
+`retrieve(query, k)` is the guaranteed SC-D unit. The test is **non-tautological by
+construction**: the true match shares the structured fields with the query but uses
+*different wording*, while two decoys share the query's *surface text* but differ
+structurally (disjoint codes, different measure type, different magnitude bucket). The
+structural retriever returns the **true match as top-1** and ranks the surface-text decoys
+last — a text-matching retriever would be fooled. Ordering is **deterministic** (ties
+break by verdict id).
+
+**ExpeL injection:** a thin `ExpeLContextProvider` subclasses the real
+`agent_framework.ContextProvider` and, in `before_run`, injects the retrieved verdicts as
+few-shot instructions via `SessionContext.extend_instructions` — asserted against the
+introspected interface. The `retrieve` ranking remains the deliverable regardless of the
+MAF session surface.
+
+## Out of scope (Fase 2 option)
+
+The **embedding-based** similarity path is intentionally not built — it needs a live
+endpoint, is non-deterministic, and serves no SC for a throwaway spike. Structured-field
+similarity is sufficient to confirm B2. Embeddings (or a hybrid structured+embedding score)
+are a Fase 2 option for the durable VerdictStore.
+
+## Token use
+
+**0 — deterministic retrieval.** No model is called; similarity is pure arithmetic over
+structured fields. The ExpeL provider only *formats* retrieved verdicts into few-shot text
+— the actual model call that would consume tokens is a Fase 2 concern.
+
+## Implication for Fase 2
+
+The learning loop's retrieval is realizable with a simple, deterministic, structural
+similarity — good enough to surface relevant prior verdicts. Fase 2 can keep this as the
+baseline and add embeddings only if structured similarity proves insufficient on real data.

spikes/d_verdictstore.py

@@ -0,0 +1,145 @@
+"""Spike D — VerdictStore + ExpeL retrieval (B2).
+
+De-risks the assumption that retrieval surfaces a relevant *prior* verdict for a similar
+new proposal — the substrate for the framework's learning loop.
+
+**Similarity is structural, not textual (reviewer refinement #2):** a weighted score over
+*structured* fields — Jaccard on the affected cost-code set + a `measure_type` match + a
+magnitude-bucket match on the claimed saving — **never** raw description text. This makes
+the retrieval non-tautological: a true match with different wording beats decoys that
+merely share surface text.
+
+`retrieve()` is the guaranteed SC-D deliverable (always tested). A thin
+`ExpeLContextProvider` wraps it for few-shot injection via the real `ContextProvider`
+hook. The embedding-based similarity path is intentionally **out of scope** for a
+throwaway spike (a Fase 2 option — see findings-d).
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from agent_framework import ContextProvider
+
+# Weights: the affected cost-code overlap dominates, then measure type, then magnitude.
+_W_CODES, _W_MEASURE, _W_MAGNITUDE = 0.60, 0.25, 0.15
+_MAGNITUDE_BUCKETS = [(0.0, 1e5), (1e5, 5e5), (5e5, 1e6), (1e6, float("inf"))]
+
+
+@dataclass(frozen=True)
+class ProposalFeatures:
+    """The *structured* features retrieval ranks on. ``description`` is surface text and
+    is deliberately NOT part of the similarity score."""
+
+    affected_codes: frozenset[str]
+    measure_type: str
+    claimed_saving_nok: float
+    description: str = ""
+
+
+@dataclass(frozen=True)
+class Verdict:
+    """One historical expert verdict in the store."""
+
+    id: str
+    proposal_features: ProposalFeatures
+    decision: str  # "approved" | "rejected"
+    rationale: str
+
+
+def _magnitude_bucket(value: float) -> int:
+    for i, (low, high) in enumerate(_MAGNITUDE_BUCKETS):
+        if low <= value < high:
+            return i
+    return len(_MAGNITUDE_BUCKETS) - 1
+
+
+def _jaccard(a: frozenset[str], b: frozenset[str]) -> float:
+    union = a | b
+    return len(a & b) / len(union) if union else 1.0
+
+
+def similarity(query: ProposalFeatures, candidate: ProposalFeatures) -> float:
+    """Weighted structural similarity in [0, 1] — text is ignored by design."""
+    codes = _jaccard(query.affected_codes, candidate.affected_codes)
+    measure = 1.0 if query.measure_type == candidate.measure_type else 0.0
+    magnitude = 1.0 if _magnitude_bucket(query.claimed_saving_nok) == _magnitude_bucket(
+        candidate.claimed_saving_nok
+    ) else 0.0
+    return _W_CODES * codes + _W_MEASURE * measure + _W_MAGNITUDE * magnitude
+
+
+@dataclass
+class VerdictStore:
+    """Minimal in-memory store of historical verdicts."""
+
+    verdicts: list[Verdict]
+
+    def retrieve(self, query: ProposalFeatures, k: int) -> list[Verdict]:
+        """Return the top-``k`` verdicts by structural similarity. Deterministic: ties
+        break by verdict id, so ordering is stable across runs."""
+        if k <= 0:
+            raise ValueError(f"k must be positive, got {k}")
+        ranked = sorted(
+            self.verdicts,
+            key=lambda v: (-similarity(query, v.proposal_features), v.id),
+        )
+        return ranked[:k]
+
+
+class ExpeLContextProvider(ContextProvider):
+    """Wraps `VerdictStore.retrieve` for ExpeL few-shot injection.
+
+    Asserted only against the introspected `ContextProvider` interface
+    (`before_run` + `SessionContext.extend_instructions`); the `retrieve` ranking remains
+    the guaranteed SC-D deliverable regardless of the MAF session surface.
+    """
+
+    def __init__(self, store: VerdictStore, query: ProposalFeatures, *, k: int = 3) -> None:
+        super().__init__(source_id="expel-verdictstore")
+        self._store = store
+        self._query = query
+        self._k = k
+
+    def format_fewshot(self) -> str:
+        hits = self._store.retrieve(self._query, self._k)
+        body = "\n".join(f"- [{v.id}] {v.decision}: {v.rationale}" for v in hits)
+        return f"Relevant prior verdicts (ExpeL few-shot):\n{body}"
+
+    async def before_run(self, *, agent: object, session: object, context: object, state: dict) -> None:
+        context.extend_instructions([self.format_fewshot()])  # type: ignore[attr-defined]
+
+
+def seed_store() -> VerdictStore:
+    """Seed 12 synthetic verdicts spanning the reference domain's cost codes, measure
+    types, magnitudes, and decisions (B2 store of 10–20)."""
+    rows = [
+        ("V01", {"05.2", "03.1"}, "scope_reduction", 180_000, "approved", "asphalt + base course trimmed within feasible range"),
+        ("V02", {"05.2"}, "rate_renegotiation", 60_000, "approved", "renegotiated asphalt unit rate"),
+        ("V03", {"07.4"}, "material_substitution", 120_000, "rejected", "granite kerb substitution unsafe"),
+        ("V04", {"09.1"}, "scope_reduction", 240_000, "approved", "fewer LED masts on low-traffic stretch"),
+        ("V05", {"02.3", "03.1"}, "scope_reduction", 350_000, "rejected", "soil replacement is load-bearing, cannot cut"),
+        ("V06", {"21.2"}, "rate_renegotiation", 90_000, "approved", "blasting rate renegotiated"),
+        ("V07", {"22.4"}, "material_substitution", 700_000, "rejected", "fiber shotcrete spec is mandated"),
+        ("V08", {"88.2"}, "scope_reduction", 150_000, "approved", "concrete repair area re-measured smaller"),
+        ("V09", {"87.3"}, "material_substitution", 130_000, "approved", "alternative membrane qualified"),
+        ("V10", {"05.2", "03.1"}, "rate_renegotiation", 200_000, "approved", "combined paving rate discount"),
+        ("V11", {"01.1"}, "scope_reduction", 95_000, "rejected", "rigging is fixed cost, no scope to cut"),
+        ("V12", {"31.3"}, "scope_reduction", 110_000, "approved", "drainage length reduced after survey"),
+    ]
+    return VerdictStore(
+        verdicts=[
+            Verdict(
+                id=vid,
+                proposal_features=ProposalFeatures(
+                    affected_codes=frozenset(codes),
+                    measure_type=mtype,
+                    claimed_saving_nok=saving,
+                    description=desc,
+                ),
+                decision=decision,
+                rationale=desc,
+            )
+            for vid, codes, mtype, saving, decision, desc in rows
+        ]
+    )

tests/spikes/test_d_verdictstore.py

@@ -0,0 +1,110 @@
+"""Spike D tests — non-tautological top-K retrieval (B2).
+
+The true match shares the *structured* similarity fields with the query but uses different
+description text; the decoys share surface description text but differ structurally. A
+text-matching retriever would be fooled by the decoys — a structural one is not.
+Pattern: tests/test_reference_domain.py.
+"""
+
+import pytest
+
+from spikes.d_verdictstore import (
+    ExpeLContextProvider,
+    ProposalFeatures,
+    Verdict,
+    VerdictStore,
+    seed_store,
+)
+
+_QUERY = ProposalFeatures(
+    affected_codes=frozenset({"05.2", "03.1"}),
+    measure_type="scope_reduction",
+    claimed_saving_nok=220_000,  # bucket [100k, 500k)
+    description="asphalt base course reduction near school",
+)
+
+
+def _store_with_true_match_and_decoys() -> tuple[VerdictStore, str]:
+    true_match = Verdict(
+        id="TRUE",
+        proposal_features=ProposalFeatures(
+            affected_codes=frozenset({"05.2", "03.1"}),  # same codes
+            measure_type="scope_reduction",  # same measure type
+            claimed_saving_nok=200_000,  # same magnitude bucket
+            description="zzz totally unrelated wording alpha beta",  # DIFFERENT text
+        ),
+        decision="approved",
+        rationale="prior scope reduction on the same codes was approved",
+    )
+    # Decoys share the query's SURFACE text but differ structurally (disjoint codes,
+    # different measure type, different magnitude bucket).
+    decoy_low = Verdict(
+        id="DECOY-LOW",
+        proposal_features=ProposalFeatures(
+            affected_codes=frozenset({"09.1"}),
+            measure_type="rate_renegotiation",
+            claimed_saving_nok=50_000,  # bucket [0, 100k)
+            description="asphalt base course reduction near school",  # same words as query
+        ),
+        decision="rejected",
+        rationale="surface-text decoy",
+    )
+    decoy_high = Verdict(
+        id="DECOY-HIGH",
+        proposal_features=ProposalFeatures(
+            affected_codes=frozenset({"21.2"}),
+            measure_type="material_substitution",
+            claimed_saving_nok=700_000,  # bucket [500k, 1M)
+            description="asphalt base course reduction extra words",
+        ),
+        decision="rejected",
+        rationale="surface-text decoy",
+    )
+    # Order deliberately not putting the true match first.
+    return VerdictStore(verdicts=[decoy_low, true_match, decoy_high]), "TRUE"
+
+
+def test_retrieve_finds_structural_match_over_text_decoys() -> None:
+    store, true_id = _store_with_true_match_and_decoys()
+    hits = store.retrieve(_QUERY, k=3)
+    assert hits[0].id == true_id  # structural match ranks #1 despite different wording
+    assert true_id in {h.id for h in hits[:1]}  # within top-K (top-1 here)
+
+
+def test_retrieve_is_deterministic() -> None:
+    store, _ = _store_with_true_match_and_decoys()
+    assert [h.id for h in store.retrieve(_QUERY, k=3)] == [h.id for h in store.retrieve(_QUERY, k=3)]
+
+
+def test_retrieve_rejects_non_positive_k() -> None:
+    store, _ = _store_with_true_match_and_decoys()
+    with pytest.raises(ValueError):
+        store.retrieve(_QUERY, k=0)
+
+
+def test_seed_store_has_10_to_20_verdicts() -> None:
+    store = seed_store()
+    assert 10 <= len(store.verdicts) <= 20
+
+
+class _RecordingContext:
+    """Duck-typed stand-in for SessionContext — records injected instructions without
+    depending on the (private) SessionContext internals."""
+
+    def __init__(self) -> None:
+        self.instructions: list[str] = []
+
+    def extend_instructions(self, items: list[str]) -> None:
+        self.instructions.extend(items)
+
+
+async def test_expel_provider_injects_retrieved_fewshot() -> None:
+    store, true_id = _store_with_true_match_and_decoys()
+    provider = ExpeLContextProvider(store, _QUERY, k=2)
+    # The deterministic deliverable: the few-shot text carries the retrieved true match.
+    assert true_id in provider.format_fewshot()
+    # The injection hook (real ContextProvider.before_run signature) extends instructions.
+    ctx = _RecordingContext()
+    await provider.before_run(agent=None, session=None, context=ctx, state={})
+    assert len(ctx.instructions) == 1
+    assert true_id in ctx.instructions[0]