perf(dedup): eliminate per-pair array allocation in compareLocations

[guyghost]

Context

As part of a performance pass over the live scan hot path (scanner.ts → buildDeterministicMissions → deduplicateMissionsDetailed → scoreMission), I surveyed 10 candidate optimizations. This PR lands the highest-impact, lowest-effort one.

The win

compareLocations runs on every candidate pair during deduplication. It allocated a merged array every call just to do a single .some() membership check against a 3-entry constant set:

// before — allocates per pair
[...aTokens, ...bTokens].some((token) => REMOTE_LOCATION_TOKENS.has(token))

Replaced with a zero-allocation helper that iterates the tiny constant needles set with O(1) Set.has checks:

// after — zero allocations
hasAnyToken(aTokens, REMOTE_LOCATION_TOKENS) ||
hasAnyToken(bTokens, REMOTE_LOCATION_TOKENS)

Semantics are identical (OR of remote-token-in-A or remote-token-in-B ≡ remote-token-in-A∪B). This matches the file's existing allocation-avoidance style (intersectionSize, inclusion-exclusion in jaccardSimilarity).

Verification

• ✅ vitest run tests/unit/scoring/dedup.test.ts — 22/22 pass (added a regression test covering the remote-context fallback with disjoint location token sets)
• ✅ pnpm --filter @pulse/extension typecheck — clean
• ✅ pnpm --filter @pulse/extension lint — 0 errors (8 pre-existing warnings in an untouched file)
• ✅ Pre-push ci:check gate passed (format + lint + typecheck + test + build)

The other 9 improvements identified (not in this PR)

Listed for visibility / future work, roughly ordered by impact-to-effort:

1. relevance.ts — profile Set rebuilt per mission. rawStackScore rebuilds a Set from the profile's keywords on every mission during scoring. Hoist the Set construction out of the per-mission loop.
2. feed.svelte.ts — per-keystroke text re-join. recomputeFilteredMissions re-joins all mission text on every search keystroke. Memoize per-mission searchable text.
3. relevance.ts — normalizeWeights recomputed per mission. The weights object is rebuilt for every mission; it only depends on the profile, not the mission.
4. location-matching.ts — hasTokenMatch uses Array.includes. Linear scan over arrays that could be Sets for O(1) lookups in the location matcher.
5. dedup.ts — computeMissionScore(existing) recomputed inside candidate loop. Re-score the kept mission once outside the inner loop instead of per pair.
6. location-matching.ts — repeated tokenization in findMetroArea. The same location string is tokenized multiple times; tokenize once and reuse.
7. db.ts — batch IndexedDB reads. Several scan-path reads that could be combined into a single cursor / getAllKeys traversal.
8. semantic-scorer.ts — cache key construction. String concatenation for cache keys on every lookup; could precompute.
9. scanner.ts — redundant Array.from / spread on results. A couple of intermediate array materializations before dedup that could be streamed.

Note: trackStage / createDefaultPipelineStages() / runPipeline() in pipeline.ts were intentionally excluded — per application-tracking.model.md, that code is not part of the live scan runtime (dead code), so optimizing it would be wasted effort.

---

🤖 Generated with Copilot

---

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
pulse	Ready	Preview	Aug 12, 2026 7:13am

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
pulse-dashboard	Skipped		Aug 12, 2026 7:13am

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[Copilot]

Pull request overview

This PR optimizes the deduplication hot path in the extension’s pure scoring core by removing a per-candidate-pair array allocation inside compareLocations, replacing it with an allocation-free helper that checks for “remote-context” tokens.

Changes:

• Introduces hasAnyToken(tokens, needles) to perform zero-allocation membership checks across token sets.
• Updates compareLocations to use hasAnyToken instead of spreading/merging token arrays.
• Adds a unit regression test to ensure disjoint-location missions still dedupe when one location implies remote context.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
apps/extension/src/lib/core/scoring/dedup.ts	Replaces per-pair merged-array allocation in compareLocations with an allocation-free helper.
apps/extension/tests/unit/scoring/dedup.test.ts	Adds regression coverage for the remote-context fallback when location token sets are disjoint.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.