Part of #5097. P2. MiroFish theater-limited simulation. Verified 2026-07-09.

Context / de-risking: the sim does not rewrite published pred.probability — decorations (simulationAdjustment/simPathConfidence/demotedBySimulation) are consumed only by ForecastPanel.ts (display) + deep-path acceptance + next-run priorWorldState seeding. Merge math itself is correct, clamped, idempotent, and race-safe. So these are lower-urgency correctness/completeness items, not Brier-corrupting.

Accuracy

• A1 — demotion-gate threshold wrong. applyPostSimulationRescore gate uses acceptanceScore < 0.65 (:17722) with a comment deriving "max negative −0.15". But invalidator (−0.12) and stabilizer (−0.15) are separate loops that can both fire (−0.27 combined), so the threshold should be 0.77. Paths in [0.65, 0.77) that a combined hit would push below 0.50 silently skip a legitimate demotion (no_actionable_paths). Fix: threshold 0.50 + 0.27 = 0.77 + fix the derivation comment.
• A2 (SUSPECTED) — broad substring keyword matching in matchesBucket/matchesChannel can over-fire adjustments (e.g. bare 'military'/'attack').

Completeness (silent-by-design gaps)

• Geo-group dedup can drop a higher-ranked theater for a lower-ranked one (hard cap 3, first-per-geo-group): Add smart zoom visibility for dense map layers #2/Add shareable map state URLs and Copy Link button #3 candidates sharing a group with Batch market and RSS fetching with progressive updates #1 are never simulated.
• 3 archetypes not enforced — round parser passes with 1–2 of {escalation, containment, market_cascade}.
• Round-2 failure → positive-only theater — stabilizers/invalidators exist only in round 2, so a r2-failed theater can only receive + adjustments → systematic upward bias for degraded theaters.
• timingMarkers produced but never consumed by scoring — display-only. Notable: epic(forecast): verified probabilistic forecasting — resolution engine, public Brier track record, learned calibration, LLM ensemble (MiroFish++ roadmap) #4930 Bet 6 assumes they can be bound to feeds for resolution; they aren't wired into scoring today.

Related: #5097, #4930 (Bet 6).