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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ignored	Preview	Jul 6, 2026 10:34am

Greptile Summary

This PR introduces a textIncludesTerm word-boundary helper (regex-backed, 4000-entry compiled-regex cache) and replaces bare String.includes at nine call sites across six detector functions and the shared countTermMatches/computeMarketMatchScore paths, eliminating country-substring false positives that were corrupting forecast probabilities. Thirteen new regression tests (7 confirmed-red, 6 positive controls) are added to forecast-detectors.test.mjs.

• Incomplete fix in computeHeadlineRelevance: the domain-hint scoring loop (line 2130), title-token scoring loop (line 2134), and both counters in the requireSemantic guard (lines 2138\u20132139) still use bare .includes. The PR asserts this function is covered through countTermMatches, but that only handles region terms; the domain and title-token paths are independent and remain vulnerable to the same substring false positive.
• Cache eviction: the regex cache uses a full Map.clear() flush on overflow rather than single-entry eviction, which could cause a burst of regex recompilation if many distinct terms are encountered in one seed run.

Confidence Score: 3/5

The fix is real and correct for the nine patched sites, but the news-corroboration path in computeHeadlineRelevance retains the same bare .includes loops for domain hints and title tokens.

Four unpatched .includes calls remain in computeHeadlineRelevance for the same class of substring false positive the PR was written to fix. A headline about Tirana can still incorrectly score against an Iran prediction title-token through this path, and the requireSemantic guard counts are also affected.

scripts/seed-forecasts.mjs lines 2129–2140 — computeHeadlineRelevance domain-hint and title-token loops need the same textIncludesTerm swap applied in computeMarketMatchScore.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Incoming text + term"] --> B{"textIncludesTerm"}
    B --> C{"Cache hit?"}
    C -- yes --> D["Return cached RegExp"]
    C -- no --> E{"Cache full >= 4000?"}
    E -- yes --> F["cache.clear full flush"]
    E -- no --> G["Build word-boundary regex"]
    F --> G
    G --> H["Store in cache"]
    H --> D
    D --> I["re.test lowerText"]

    subgraph Fixed["Fixed 9 sites"]
        J["countTermMatches"]
        K["computeMarketMatchScore domain hints + title tokens"]
        L["detectPoliticalScenarios"]
        M["detectInfraScenarios"]
        N["detectSupplyChainScenarios"]
        O["detectMilitaryScenarios"]
    end

    subgraph Unfixed["Still bare includes"]
        P["computeHeadlineRelevance domain hints 2130+2138 title tokens 2134+2139"]
    end

    J --> B
    K --> B
    L --> B
    M --> B
    N --> B
    O --> B
    P -.-> Q["String.includes substring FP risk"]

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A["Incoming text + term"] --> B{"textIncludesTerm"}
    B --> C{"Cache hit?"}
    C -- yes --> D["Return cached RegExp"]
    C -- no --> E{"Cache full >= 4000?"}
    E -- yes --> F["cache.clear full flush"]
    E -- no --> G["Build word-boundary regex"]
    F --> G
    G --> H["Store in cache"]
    H --> D
    D --> I["re.test lowerText"]

    subgraph Fixed["Fixed 9 sites"]
        J["countTermMatches"]
        K["computeMarketMatchScore domain hints + title tokens"]
        L["detectPoliticalScenarios"]
        M["detectInfraScenarios"]
        N["detectSupplyChainScenarios"]
        O["detectMilitaryScenarios"]
    end

    subgraph Unfixed["Still bare includes"]
        P["computeHeadlineRelevance domain hints 2130+2138 title tokens 2134+2139"]
    end

    J --> B
    K --> B
    L --> B
    M --> B
    N --> B
    O --> B
    P -.-> Q["String.includes substring FP risk"]

_{Reviews (1): Last reviewed commit: "fix(forecast): word-boundary term matchi..." | Re-trigger Greptile}

Review findings addressed in 4813c35 (all three verified at head 5b617ec before fixing):

F1 (P1) reverse-lookup poisoning — fixed + worse than reported. getSearchTermsForRegion reverse lookup now uses word-boundary matching. Live probe before: Somalia => [Somalia, Mali, mali, bamako] — the substring hit not only injected Mali's terms, the loop's break displaced Somalia's own keywords entirely. After: Somalia => [Somalia, somalia, mogadishu], Nigeria => [Nigeria, nigeria, lagos, abuja]. Regression tests assert both no-leak and own-keywords-present, plus a Myanmar (Burma) parenthetical positive control and a Nigeria-vs-Niger-market calibration guard.

F2 (P1) computeHeadlineRelevance — fixed. All four raw .includes sites (domain-hint loop, title-token loop, both requireSemantic counters) now use textIncludesTerm. RED-first test: 'Award season kicks off in Tirana' with conflict hints + iran token scores 0 under requireSemantic; 'War fears grow as Iran mobilizes' still scores (positive control).

F3 (P2) cache eviction — fixed. FIFO oldest-key eviction instead of clear() at capacity.

Verification: forecast-detectors 206/206 (3 new tests confirmed RED pre-fix), forecast-trace-export 310/310, history/integrity/resilience/narrative-cache suites green, evaluate-forecast-benchmark 6/6.

Re-review findings addressed in d453a64 (both reproduced at head 4813c35 before fixing):

F1 (P1) plural regression — fixed. The term regex now accepts an optional s/es suffix on the text side: (?:^|[^a-z0-9])term(?:s|es)?(?:[^a-z0-9]|$). 'elections'/'attacks' match the singular hints again; the original false-positive geometry stays closed ('niger'+suffix still cannot match inside 'nigeria', 'mali' still cannot match inside 'somalia' — the leading/trailing boundary is still required, only a plural suffix on the SAME word is tolerated). RED-first tests: Nigeria pred + 'peaceful elections' market now calibrates at the exact 0.4/0.6 blend; 'Missile attacks intensify' scores under requireSemantic.

F2 (P1) compound-name reverse lookup — fixed. Exact name match now wins outright (the JSON has exact entries for all three repro cases); the containment fallback selects the LONGEST contained name instead of breaking on the first. Probes: DR Congo => [DR Congo, congo, kinshasa, drc], Guinea-Bissau => [Guinea-Bissau, guinea-bissau], Papua New Guinea => [Papua New Guinea, papua new guinea]; round-1 results unchanged (Somalia => [Somalia, somalia, mogadishu], Myanmar parenthetical control intact).

Verification: forecast-detectors 211/211 (5 new tests RED at prior head), forecast-trace-export 310/310, history/integrity/resilience/narrative suites green, evaluate-forecast-benchmark 6/6.

Round-3 finding addressed in fd1543d (reproduced at head d453a64 first: the wares market calibrated the Iran conflict forecast 0.70→0.54).

P2 global 'es' suffix — fixed. The suffix is now conditional on the term's ending: sibilants (s/x/z/ch/sh) allow (?:es)? (gas→gases, tax→taxes), everything else plain s? (attack→attacks, election→elections). 'war' can no longer match 'wares'. RED-first tests: wares market yields domainHits 0 and no calibration; greenhouse-gases positive control keeps the sibilant path.

Verification: forecast-detectors 214/214, trace-export 310/310, adjacent suites green, benchmark 6/6.