feat(simulation): confidence-weighted adjustments + simulationSignal trace lane

[koala73]

Summary

• Confidence-weighted bonuses: +0.08 and +0.04 simulation bonuses now scale by simPathConfidence from SimulationTopPath.confidence. Missing/null/zero confidence falls back to 1.0 (conservative — old artifacts not penalized). Negative adjustments (-0.12/-0.15) remain flat (structural, not sim-confidence-dependent).
• simulationSignal trace lane: applySimulationMerge attaches a compact simulationSignal object (backed, adjustmentDelta, channelSource, demoted, simPathConfidence) to each ExpandedPath when adjustment ≠ 0. Written to R2 trace artifacts via existing writeForecastTraceArtifacts.
• simPathConfidence observability: Added simPathConfidence field to SimulationAdjustmentDetail so adjustments per path are fully inspectable in trace output.
• Named-actor gap documented: Macro-financial theaters with role-based stateSummary.actors ("Commodity traders") have actorOverlapCount=0 by design — only named geo-political actor theaters (maritime/security) fire the +0.04 bonus.

Closes out the last two "Next (planned)" items from docs/internal/wm-mirofish-gap.md.

ForecastPanel UI chip deferred: simulationSignal lives in R2 trace artifacts. Surfacing it as a UI chip requires a proto field addition + buf generate + prediction-to-path plumbing. Tracked for a follow-up PR.

Testing

• T-N1..T-N8: confidence weighting unit tests (full 1.0, partial 0.72 with overlap, 0.35 without, undefined fallback, zero fallback, flat negatives, simPathConfidence observability, unweighted negatives)
• T-O1..T-O4: simulationSignal tests (backed + not-demoted, absent on zero adjustment, demoted on threshold crossing, simPathConfidence propagated)
• 267 passing, 0 failing

Post-Deploy Monitoring & Validation

• What to monitor: R2 trace artifacts for simulation-outcome-*.json — check simulationSignal fields on paths with non-zero simulationAdjustment
• Expected healthy behavior: Paths with bucketChannelMatch=true have simulationSignal.backed=true, adjustmentDelta matches simulationAdjustment, simPathConfidence < 1.0 when sim output includes fractional confidence
• Failure signal: simulationSignal absent on a path that has non-zero simulationAdjustment (would indicate applySimulationMerge early-exit regression)
• No production runtime impact — no Redis writes or API surface changes; all changes are in trace artifact content

---

🤖 Generated with Claude Sonnet 4.6 (200K context) via Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ready	Preview, Comment	Mar 29, 2026 6:04pm

[greptile-apps]

Greptile Summary

This PR introduces two tightly scoped improvements to the simulation merge pipeline: confidence-weighted bonuses (scaling +0.08/+0.04 by simPathConfidence from the matched SimulationTopPath) and a compact simulationSignal trace object written to each ExpandedPath when a non-zero adjustment is applied. Negative adjustments (-0.12/-0.15) intentionally remain flat. The changes are purely trace/observability additions with no Redis or API surface impact.

• Confidence weighting — simConf is derived from bucketChannelMatch.confidence with a safe fallback: any missing/null/zero/non-finite value collapses to 1.0, preserving backward-compat with older LLM artifacts.
• simulationSignal attachment — written by applySimulationMerge only when adjustment !== 0; carries backed, adjustmentDelta, channelSource, demoted, and simPathConfidence for downstream trace inspection.
• SimulationAdjustmentDetail.simPathConfidence — extends the existing detail record for full per-path observability in R2 trace artifacts.
• Tests (T-N1..T-N8, T-O1..T-O4) — comprehensive coverage of weighting, all fallback branches, flat negatives, and signal lifecycle including the zero-adjustment absent-signal case.
• Minor JSDoc mismatch — the SimulationSignal.backed comment describes it as indicating "non-zero adjustment applied", but the implementation sets it only for adjustment > 0; negative-adjustment paths will have backed=false. The comment should be updated to reflect the actual semantics.

Confidence Score: 5/5

Safe to merge — no API surface or Redis changes; only minor JSDoc inaccuracies remain.

All findings are P2 (style/documentation). The logic for confidence weighting, fallbacks, flat negatives, and signal lifecycle is correct and thoroughly tested. The remaining JSDoc comment on SimulationSignal.backed is the only open item and does not affect runtime behavior.

scripts/seed-forecasts.types.d.ts — SimulationSignal.backed JSDoc needs a wording update to match adjustment > 0 semantics.

Important Files Changed

Filename	Overview
scripts/seed-forecasts.mjs	Core logic changes: confidence-weighted bonuses (+0.08/+0.04 × simConf) replace flat bonuses, and simulationSignal trace object is attached to each adjusted path; fallback logic (undefined/null/zero → 1.0) and negative-flat behavior are correct.
scripts/seed-forecasts.types.d.ts	New SimulationSignal interface and additions to ExpandedPath/SimulationAdjustmentDetail; the backed field JSDoc comment is misleading (says "non-zero adjustment applied" but the impl only sets it true for adjustment > 0).
tests/forecast-trace-export.test.mjs	T-N1..T-N8 and T-O1..T-O4 cover confidence weighting, fallback paths, flat negatives, and simulationSignal lifecycle; all assertions are logically consistent with the implementation.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[applySimulationMerge called] --> B{theaterResults present?}
    B -- No --> Z[return null simulationEvidence]
    B -- Yes --> C[for each ExpandedPath]
    C --> D{simResult found for path?}
    D -- No --> C
    D -- Yes --> E[computeSimulationAdjustment]
    E --> F{bucketChannelMatch?}
    F -- Yes --> G[rawConf = match.confidence]
    G --> H{isFinite and > 0?}
    H -- No --> I[simConf = 1.0 fallback]
    H -- Yes --> J[simConf = min 1 rawConf]
    I & J --> K[adjustment += 0.08 x simConf]
    K --> L{actorOverlap >= 2?}
    L -- Yes --> M[adjustment += 0.04 x simConf]
    L -- No --> N
    M --> N{invalidator hit?}
    F -- No --> N
    N -- Yes --> O[adjustment -= 0.12 flat]
    N -- No --> P{stabilizer hit?}
    O --> P
    P -- Yes --> Q[adjustment -= 0.15 flat]
    P -- No --> R[return adjustment + details]
    Q --> R
    R --> S{adjustment == 0?}
    S -- Yes --> C
    S -- No --> T[write simulationSignal to path]
    T --> U{wasAccepted and score < threshold?}
    U -- Yes --> V[path.demotedBySimulation = true]
    U -- No --> W{was rejected and score >= threshold?}
    W -- Yes --> X[path.promotedBySimulation = true]
    W -- No --> C
    V & X --> C

Loading

_{Reviews (1): Last reviewed commit: "feat(simulation): confidence-weighted ad..." | Re-trigger Greptile}

[greptile-apps]

backed JSDoc contradicts the implementation

The comment reads "Non-zero simulation adjustment was applied (positive = promoted, negative = weakened)", which implies backed would be true for any non-zero adjustment, including negative ones. But the implementation in applySimulationMerge sets backed: adjustment > 0, so paths that receive a negative adjustment (e.g. an invalidator hit) will have backed=false even though the signal is present and has a non-zero adjustmentDelta.

A future consumer checking simulationSignal.backed to determine "did simulation touch this path?" would get a false negative for all weakened paths. Aligning the comment with the actual semantics avoids that confusion:

Suggested change

	/** Non-zero simulation adjustment was applied (positive = promoted, negative = weakened). */
	/** True when adjustment > 0 — simulation positively supports this path (promoted or reinforced). False for paths weakened by simulation. */
	backed: boolean;

[greptile-apps]

Redundant unary + on parseFloat return value

parseFloat(...) already returns a number (or NaN), so the leading + cast is a no-op in both places. It doesn't affect correctness, but it adds visual noise and suggests the author may have intended Number(...) or +(str) on a string that wasn't already numeric. Consider removing the outer + for clarity:

Suggested change

	adjustment += +parseFloat((0.08 * simConf).toFixed(3));
	details.bucketChannelMatch = true;
	details.simPathConfidence = simConf;
	const simActors = new Set((Array.isArray(bucketChannelMatch.keyActors) ? bucketChannelMatch.keyActors : []).map(normalizeActorName));
	const overlap = candidateActors.filter((a) => simActors.has(a));
	details.actorOverlapCount = overlap.length;
	// Overlap bonus fires only when both sides have named geo-political actors.
	// Macro-financial theaters with role-based stateSummary.actors (e.g. "Commodity traders",
	// "Central banks") will have actorOverlapCount=0 — this is expected, not a bug.
	if (overlap.length >= 2) {
	adjustment += 0.04;
	adjustment += +parseFloat((0.04 * simConf).toFixed(3));
	adjustment += parseFloat((0.08 * simConf).toFixed(3));
	details.bucketChannelMatch = true;
	details.simPathConfidence = simConf;
	const simActors = new Set((Array.isArray(bucketChannelMatch.keyActors) ? bucketChannelMatch.keyActors : []).map(normalizeActorName));
	const overlap = candidateActors.filter((a) => simActors.has(a));
	details.actorOverlapCount = overlap.length;
	// Overlap bonus fires only when both sides have named geo-political actors.
	// Macro-financial theaters with role-based stateSummary.actors (e.g. "Commodity traders",
	// "Central banks") will have actorOverlapCount=0 — this is expected, not a bug.
	if (overlap.length >= 2) {
	adjustment += parseFloat((0.04 * simConf).toFixed(3));
	}

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}