DESIGN_STRATEGIST.md §2.5. The strategist's two write actions.
propose_lead validates motivating_hypothesis exists in the graph,
validates expected_evidence_type is a real edge type, validates
source_id refers to a real source in the case — fast specific
errors so the strategist gets fixable feedback rather than a
generic crash. On success, calls graph.add_lead with proposed_by=
"strategist" and round_number=graph.current_strategist_round so
the round-completion code can collect this round's leads.
declare_investigation_complete sets graph.strategist_complete_requested
which the orchestrator inspects after each strategist run to decide
whether to break the loop. reason must come from a closed enum so
the audit log is consistent.
EvidenceGraph gains two transient run-context fields:
current_strategist_round — set by orchestrator at start of round
strategist_complete_requested — flipped by declare_complete
These are intentionally NOT persisted — they're per-run flags, not
graph state.
Both tools required to be in InvestigationStrategist.mandatory_record_
tools (added in S4) so the agent's forced-retry mechanism kicks in if
it returns without taking a documented decision.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
DESIGN_STRATEGIST.md §2. Four read-only view tools the strategist uses
to ground its decision each round.
graph_overview() — hypotheses table (log_odds, conf, edges_in,
distinct_sources, recent_flip), sources table,
pending leads. distinct_sources is the
critical signal: a hypothesis with 23 edges
but only 1 distinct_source has fragile cross-
source independence and is a candidate for
a corroboration-seeking lead.
source_coverage(src) — per-source ✓/✗ against an expected-artefact
catalogue. Catalogue is heuristic hints,
NOT a forced checklist. Footer reminds the
strategist to investigate ✗ items only when
an active hypothesis depends on them — this
is the "应试能力存在但不被绑死" guardrail.
marginal_yield(N) — new phenomena / edges / status flips per
recent round. Two consecutive zero-yield
rounds = strong signal to declare complete.
budget_status() — usage vs caps (tool_calls, rounds, wall
clock). Pacing warnings at 70% / 90%.
tools/strategy.py also exports EXPECTED_ARTEFACTS, a per-source-type
table of (name, detector, value_for) entries. Detectors are
substring patterns on tool name + args; the matcher resolves at
call time against graph.tool_invocations. Catalogue covers iOS /
Android / Windows disk / media-collection / archive source types.
All four tools registered in tool_registry, listed as read-only in
llm_client.READ_ONLY_TOOLS for parallel execution. They go through
the invocation-logging wrapper so the strategist's reads are
themselves auditable (the wrapper does NOT cache them — graph
state changes between calls).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
