feat(strategist) S1: Lead extension + InvestigationRound model
DESIGN_STRATEGIST.md §1. Foundation for the Phase 3 strategist loop. Lead now carries four annotations that let the orchestrator measure marginal yield per lead and dedupe strategist proposals: - proposed_by (agent that proposed it: "strategist", "filesystem", …) - motivating_hypothesis (hyp-id the lead is meant to corroborate/refute) - expected_evidence_type (edge type the lead's worker should produce) - round_number (0 = Phase 1 lead, ≥1 = strategist-proposed) add_lead idempotently dedupes strategist proposals on (motivating_hypothesis, expected_evidence_type, target_agent, source_id) to prevent the "strategist loops on the same lead" failure mode. New InvestigationRound dataclass records per-round provenance: before/ after hypothesis status snapshots, phenomena + edge count deltas, and the strategist's decision_rationale. ``new_phenomena_count``, ``new_edges_count``, ``status_flips`` are derived properties that the marginal_yield tool will use. start_investigation_round / complete_investigation_round / get_investigation_round / latest_round / leads_from_round complete the lifecycle. complete is idempotent on already-closed rounds. Lead.from_dict is forward-compat for state files written before this commit. InvestigationRound persists as a top-level list in graph_state.json (auto-save + load_state both wired). EvidenceGraph also gains graph.budgets and graph.run_start_monotonic fields that the budget_status view (S2) will read; orchestrator populates them in S5. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
BattleTag
@@ -9,6 +9,7 @@ import pytest
|
||||
|
||||
from evidence_graph import (
|
||||
EvidenceGraph, Phenomenon, Hypothesis, Lead, GroundingError,
|
||||
InvestigationRound,
|
||||
_compute_quality_score, _jaccard_similarity,
|
||||
prob_to_log_odds, log_odds_to_prob,
|
||||
)
|
||||
@@ -3036,3 +3037,180 @@ class TestOrchestratorMultiSource:
|
||||
assert Orchestrator._is_analysable(ok_media)
|
||||
assert not Orchestrator._is_analysable(no_path)
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Strategist loop foundation (DESIGN_STRATEGIST.md §1)
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
class TestLeadExtensionsForStrategist:
|
||||
"""Lead now carries strategist-loop annotations: proposed_by,
|
||||
motivating_hypothesis, expected_evidence_type, round_number. Old state
|
||||
files predate these fields — from_dict must accept them missing.
|
||||
"""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_add_lead_records_strategist_annotations(self):
|
||||
graph = EvidenceGraph()
|
||||
hid = await graph.add_hypothesis("h", "d")
|
||||
lid = await graph.add_lead(
|
||||
target_agent="filesystem",
|
||||
description="Check Safari bookmarks for device-switching evidence",
|
||||
proposed_by="strategist",
|
||||
motivating_hypothesis=hid,
|
||||
expected_evidence_type="supports",
|
||||
round_number=2,
|
||||
context={"source_id": "src-ios-chan"},
|
||||
)
|
||||
lead = next(l for l in graph.leads if l.id == lid)
|
||||
assert lead.proposed_by == "strategist"
|
||||
assert lead.motivating_hypothesis == hid
|
||||
assert lead.expected_evidence_type == "supports"
|
||||
assert lead.round_number == 2
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_strategist_lead_idempotency(self):
|
||||
"""Same (motivating_hyp, expected_type, target_agent, source_id) from
|
||||
the strategist should NOT create a duplicate while pending.
|
||||
"""
|
||||
graph = EvidenceGraph()
|
||||
hid = await graph.add_hypothesis("h", "d")
|
||||
first = await graph.add_lead(
|
||||
target_agent="filesystem", description="probe X",
|
||||
proposed_by="strategist", motivating_hypothesis=hid,
|
||||
expected_evidence_type="supports", round_number=1,
|
||||
context={"source_id": "src-A"},
|
||||
)
|
||||
again = await graph.add_lead(
|
||||
target_agent="filesystem", description="probe X (rephrased)",
|
||||
proposed_by="strategist", motivating_hypothesis=hid,
|
||||
expected_evidence_type="supports", round_number=2,
|
||||
context={"source_id": "src-A"},
|
||||
)
|
||||
assert first == again
|
||||
assert len(graph.leads) == 1
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_non_strategist_leads_not_deduped(self):
|
||||
"""Phase 1 worker leads (proposed_by != 'strategist') should NOT be
|
||||
deduped — agents can legitimately propose the same kind of lead
|
||||
multiple times from different contexts.
|
||||
"""
|
||||
graph = EvidenceGraph()
|
||||
hid = await graph.add_hypothesis("h", "d")
|
||||
a = await graph.add_lead(
|
||||
target_agent="filesystem", description="x", proposed_by="filesystem",
|
||||
motivating_hypothesis=hid, expected_evidence_type="supports",
|
||||
)
|
||||
b = await graph.add_lead(
|
||||
target_agent="filesystem", description="x", proposed_by="filesystem",
|
||||
motivating_hypothesis=hid, expected_evidence_type="supports",
|
||||
)
|
||||
assert a != b
|
||||
assert len(graph.leads) == 2
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_lead_from_old_state_file_loads_with_defaults(self, tmp_path):
|
||||
"""Forward-compat: a state file written before the strategist fields
|
||||
existed must still load. The new fields take their defaults.
|
||||
"""
|
||||
legacy = {
|
||||
"id": "lead-legacy01",
|
||||
"target_agent": "filesystem",
|
||||
"description": "old-style lead",
|
||||
"priority": 5,
|
||||
"context": {},
|
||||
"status": "pending",
|
||||
"hypothesis_id": None,
|
||||
}
|
||||
lead = Lead.from_dict(legacy)
|
||||
assert lead.proposed_by == ""
|
||||
assert lead.motivating_hypothesis == ""
|
||||
assert lead.round_number == 0
|
||||
|
||||
|
||||
class TestInvestigationRound:
|
||||
"""The InvestigationRound provenance node + start/complete lifecycle."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_round_lifecycle_captures_before_and_after(self):
|
||||
graph = EvidenceGraph()
|
||||
h1 = await graph.add_hypothesis("h1", "d")
|
||||
h2 = await graph.add_hypothesis("h2", "d")
|
||||
rid = await graph.start_investigation_round(1)
|
||||
r = graph.get_investigation_round(rid)
|
||||
assert r is not None
|
||||
assert r.round_number == 1
|
||||
assert r.hypothesis_status_snapshot_before == {h1: "active", h2: "active"}
|
||||
assert r.phenomena_count_before == 0
|
||||
assert r.completed_at == ""
|
||||
|
||||
pid, _ = await graph.add_phenomenon(
|
||||
"fs", "filesystem", "found something", "interp", source_tool="t",
|
||||
)
|
||||
await graph.update_hypothesis_confidence(h1, pid, "direct_evidence", "")
|
||||
|
||||
closed = await graph.complete_investigation_round(
|
||||
rid, strategist_action="propose_leads",
|
||||
decision_rationale="found new evidence for h1",
|
||||
)
|
||||
assert closed is not None
|
||||
assert closed.completed_at != ""
|
||||
assert closed.hypothesis_status_snapshot_after[h1] == "supported"
|
||||
assert closed.hypothesis_status_snapshot_after[h2] == "active"
|
||||
assert closed.new_phenomena_count == 1
|
||||
assert closed.new_edges_count == 1
|
||||
assert closed.status_flips == 1
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_complete_round_is_idempotent(self):
|
||||
graph = EvidenceGraph()
|
||||
rid = await graph.start_investigation_round(1)
|
||||
first = await graph.complete_investigation_round(rid)
|
||||
await graph.add_phenomenon("fs", "x", "y", "z", source_tool="t")
|
||||
second = await graph.complete_investigation_round(rid)
|
||||
assert first is second
|
||||
assert first.phenomena_count_after == 0
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_leads_from_round_filters_correctly(self):
|
||||
graph = EvidenceGraph()
|
||||
hid = await graph.add_hypothesis("h", "d")
|
||||
await graph.add_lead(
|
||||
target_agent="filesystem", description="r1 lead",
|
||||
proposed_by="strategist", motivating_hypothesis=hid,
|
||||
expected_evidence_type="supports", round_number=1,
|
||||
)
|
||||
await graph.add_lead(
|
||||
target_agent="filesystem", description="r2 lead",
|
||||
proposed_by="strategist", motivating_hypothesis=hid,
|
||||
expected_evidence_type="supports", round_number=2,
|
||||
context={"source_id": "src-different"},
|
||||
)
|
||||
await graph.add_lead(
|
||||
target_agent="ios_artifact", description="phase 1 finding",
|
||||
proposed_by="filesystem", round_number=0,
|
||||
)
|
||||
r1 = graph.leads_from_round(1)
|
||||
r2 = graph.leads_from_round(2)
|
||||
r0 = graph.leads_from_round(0)
|
||||
assert len(r1) == 1 and r1[0].description == "r1 lead"
|
||||
assert len(r2) == 1 and r2[0].description == "r2 lead"
|
||||
assert len(r0) == 1 and r0[0].proposed_by == "filesystem"
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_round_persistence_round_trip(self, tmp_path):
|
||||
"""Investigation rounds must survive save/load."""
|
||||
path = tmp_path / "state.json"
|
||||
graph = EvidenceGraph(persist_path=path)
|
||||
hid = await graph.add_hypothesis("h", "d")
|
||||
rid = await graph.start_investigation_round(1)
|
||||
await graph.complete_investigation_round(
|
||||
rid, decision_rationale="probe complete",
|
||||
)
|
||||
loaded = EvidenceGraph.load_state(path)
|
||||
assert len(loaded.investigation_rounds) == 1
|
||||
r = loaded.investigation_rounds[0]
|
||||
assert r.id == rid
|
||||
assert r.decision_rationale == "probe complete"
|
||||
assert hid in r.hypothesis_status_snapshot_before
|
||||
|
||||
|
||||
Reference in New Issue
Block a user