fix(classifier): match flashback/throwback after publisher brand prefix by koala73 · Pull Request #3480 · koala73/worldmonitor

koala73 · 2026-04-28T05:18:25Z

Summary

Brief 2026-04-28-0801 shipped "CBS News Radio flashback: D-Day, Invasion of Normandy in 1944" as severity=critical in slot Add shareable map state URLs and Copy Link button #3, two days after PR fix(classifier): two-layer historical-retrospective downgrade (L1 + L3) #3429 (f37216d04) shipped the historical-retrospective downgrade.
Root cause: HISTORICAL_PREFIX_RE in _classifier.ts was anchored with ^ so flashback / throwback only matched at title position 0. Publishers prepend their brand ("CBS News Radio", "BBC", "NPR") before the marker token, defeating the anchor. The keyword path then matched invasion unimpeded → critical.
Fix shipped in three commits on this branch:
1. 5ec9892 — initial relax of the prefix anchor to word-boundary.
2. f61fae5 — tightened to two-branch form after Greptile P2: anchored at start OR brand-prefix slot (Title-Case prefix words + marker + optional qualifier + colon). Adds 5 negative tests covering sentence-form occurrences ("Markets see flashback to 2008 crisis…"), so the L3b LLM-cache guard at list-feed-digest.ts:547 doesn't widen its blast radius.
3. fc1fcbf — bumps CLASSIFY_CACHE_PREFIX v4 → v5 in lockstep across all three sites (_shared.ts, list-feed-digest.ts doc-comment, ais-relay.cjs) so existing cache entries that promoted these titles to critical evict immediately rather than waiting for natural TTL.

- const HISTORICAL_PREFIX_RE = /^(?:science history|throwback|flashback)\s*:?/i;
+ const HISTORICAL_ANCHORED_PREFIX_RE =
+   /^(?:science history|throwback|flashback)\s*:?/i;
+ // No `i` flag — Title-Case enforcement on the brand prefix is load-bearing.
+ const HISTORICAL_BRAND_PREFIX_RE =
+   /^(?:[A-Z][\w'&-]*\s+){1,4}(?:[Tt]hrowback|[Ff]lashback)(?:\s+[A-Za-z]+)?\s*:/;

Test plan

Brief 2026-04-28-0801 surfaced "CBS News Radio flashback: D-Day, Invasion of Normandy in 1944" as severity=critical in slot #3, two days after PR #3429 shipped the historical-retrospective downgrade. Root cause: HISTORICAL_PREFIX_RE was anchored with `^` — `/^(?:science history|throwback|flashback)\s*:?/i` — so flashback / throwback only matched at title position 0. Publishers prepend their brand ("CBS News Radio", "BBC", "NPR") before the marker token, which defeated the anchor. The keyword path then matched `invasion` unimpeded → critical. Fix: word-boundary match for flashback / throwback (unambiguous markers regardless of position); science history stays anchored since it's an editorial-slot prefix that's only retrospective at the start of a headline. /(?:^science history\s*:?|\b(?:throwback|flashback)\b)/i Regression test added under the existing news-classifier-historical-downgrade suite covering CBS News Radio, BBC, and NPR brand-prefix forms. All 56 cases in that file plus the 56 cases across the historical-downgrade + LLM-cache-guard tests still pass.

vercel · 2026-04-28T05:18:29Z

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

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ready	Preview, Comment	Apr 28, 2026 5:46am

greptile-apps · 2026-04-28T05:20:51Z

Greptile Summary

This PR fixes a regex anchor bug in HISTORICAL_PREFIX_RE that caused brand-prefixed flashback/throwback headlines (e.g. "CBS News Radio flashback: D-Day…") to bypass the historical-retrospective downgrade and ship as critical. The fix replaces the ^-anchored match for throwback/flashback with a \b word-boundary match while keeping science history anchored, exactly scoping the change to the affected tokens.

Confidence Score: 4/5

Safe to merge; the fix is surgical and well-tested with only a minor speculative false-positive edge case at P2.

No P0 or P1 issues found. The one P2 comment flags that \bflashback\b is not strictly confined to retrospective-segment uses and a non-retrospective mid-sentence occurrence (e.g. PTSD coverage) alongside a CRITICAL keyword could cause a false downgrade — a scenario not currently covered by a negative test. The fix itself is correct for the stated regression and all existing plus new test cases pass.

No files require special attention; the P2 concern is in _classifier.ts at the new HISTORICAL_PREFIX_RE.

Important Files Changed

Filename	Overview
server/worldmonitor/news/v1/_classifier.ts	Regex loosened for `throwback`/`flashback` from `^`-anchored to `\b` word-boundary anywhere in title; `science history` anchor preserved. Fix is logically correct for the stated regression; minor P2 concern that `\bflashback\b` can match non-retrospective mid-sentence uses.
tests/news-classifier-historical-downgrade.test.mts	Three new `hasHistoricalMarker` cases (CBS, BBC, NPR brand-prefix forms) and one full `classifyByKeyword` integration case covering the exact failing title from brief 2026-04-28-0801; all cleanly organised under existing describe blocks.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Incoming headline\ne.g. 'CBS News Radio flashback: D-Day...'"] --> B["classifyByKeyword(title)"]
    B --> C["hasHistoricalMarker(title)"]
    C --> D{"HISTORICAL_PREFIX_RE\n/(?:^science history\\s*:?|\\b(?:throwback|flashback)\\b)/i"}
    D -- "match (e.g. 'flashback' word boundary)" --> E["isRetrospective = true"]
    D -- "no match" --> F["Check other markers\n(HISTORICAL_PHRASE_RE, FULL_DATE_RE, …)"]
    F --> G{"Any marker?"}
    G -- yes --> E
    G -- no --> H["isRetrospective = false"]
    B --> I["matchKeywords(lower, CRITICAL_KEYWORDS)\ne.g. 'invasion' → match"]
    E --> J{"isRetrospective?"}
    H --> J
    I --> J
    J -- true --> K["Return info / keyword-historical-downgrade\nconfidence 0.85"]
    J -- false --> L["Return critical / keyword\nconfidence 0.9"]

    style K fill:#90EE90
    style L fill:#FF6961

_{Reviews (1): Last reviewed commit: "fix(classifier): match flashback/throwba..." | Re-trigger Greptile}

greptile-apps · 2026-04-28T05:20:54Z

+const HISTORICAL_PREFIX_RE =
+  /(?:^science history\s*:?|\b(?:throwback|flashback)\b)/i;


\bflashback\b can fire on non-retrospective medical/narrative uses

flashback can appear in current-event headlines outside of the retrospective-segment context — e.g. "Veteran's flashback triggers meltdown on base" or "Abuse survivor's flashback described in court". If any CRITICAL/HIGH keyword also appears (here meltdown), the title would be downgraded to info, suppressing a real alert.

The risk is low given how rarely such phrasing appears alongside CRITICAL keywords in hard-news syndication, but the assertion in the comment that flashback is "always retrospective by definition" is slightly overstated. A negative regression test covering a non-retrospective mid-sentence flashback (e.g. "Veteran flashback triggers meltdown on base" → level critical) would document the accepted boundary explicitly and catch any future broadening of the pattern.

Good catch. Tightened in f61fae5.

Replaced the bare word-boundary form with two explicit branches:

ANCHORED (original behavior): marker at title position 0 — Throwback Thursday:, Flashback:, Science history:.

BRAND-PREFIX: 1-4 Title-Case prefix words + marker + optional qualifier word + colon. Catches CBS News Radio flashback:, BBC Throwback Thursday:, NPR Flashback Friday: while rejecting sentence-form occurrences.

const HISTORICAL_ANCHORED_PREFIX_RE = /^(?:science history|throwback|flashback)\s*:?/i; // No `i` flag — Title-Case enforcement on the brand prefix is load-bearing. const HISTORICAL_BRAND_PREFIX_RE = /^(?:[A-Z][\w'&-]*\s+){1,4}(?:[Tt]hrowback|[Ff]lashback)(?:\s+[A-Za-z]+)?\s*:/;

The brand-prefix branch deliberately requires (a) Title-Case prefix words and (b) an editorial-slot colon after the marker. Both gates together reject the realistic false-positive shapes you were worried about for the L3b LLM-cache guard.

Added 5 negative tests covering them:

Markets see flashback to 2008 crisis as bonds tumble → false (no colon)

Stocks suffer flashback to March 2020 crash → false (no colon)

Tesla stock throwback after split → false (no colon)

AI flashback to 2023 boom: Nvidia earnings beat → false (qualifier slot overruns; colon not adjacent)

markets see flashback: bonds tumble → false (lowercase first word fails Title-Case gate)

All existing positive tests still pass (61/61 across news-classifier-historical-downgrade.test.mts + news-classifier-llm-historical-guard.test.mts).

Reviewer P2 on PR #3480: widening flashback/throwback to bare word-boundary match also broadens the L3b LLM-cache guard at list-feed-digest.ts:547, which force-demotes ANY cached LLM hit to info when hasHistoricalMarker is true. A current-event headline that incidentally uses "flashback" / "throwback" as a comparison word ("Markets see flashback to 2008 crisis as bonds tumble") would be wrongly suppressed even though the marker isn't acting as a retrospective framing token. Replace the broad word-boundary form with two explicit branches: 1. ANCHORED: marker at title position 0 (original behavior — "Throwback Thursday:", "Flashback:", "Science history:"). 2. BRAND-PREFIX: 1-4 Title-Case prefix words + marker + optional qualifier word + colon. Catches the brand-prefixed forms ("CBS News Radio flashback:", "BBC Throwback Thursday:", "NPR Flashback Friday:") while rejecting sentence-form uses. The brand-prefix branch deliberately requires (a) Title-Case prefix words (rejects "markets see flashback…"), AND (b) an editorial-slot colon after the marker (rejects "Markets See Flashback To 2008 Crisis As Bonds Tumble"). The Title-Case gate uses a non-i regex so [A-Z] enforces actual capitalization rather than getting case-folded. Adds 5 negative tests for the realistic false-positive cases the reviewer was worried about — "Markets see flashback to 2008 crisis", "Stocks suffer flashback to March 2020 crash", "Tesla stock throwback after split", "AI flashback to 2023 boom: Nvidia earnings beat", and a lowercase-leading sentence form. All existing positive tests still pass (61/61 in the historical- downgrade + LLM-cache-guard suite).

Evicts cache entries that landed under the pre-publisher-prefix-fix classifier — brand-prefixed retrospective titles like "CBS News Radio flashback: D-Day, Invasion of Normandy in 1944" had been promoted to severity=critical via the `invasion` keyword and persisted in the classify cache. The new brand-prefix branch in _classifier.ts (PR #3480 commits 5ec9892 + f61fae5) re-rules those rows on next touch; v5 forces immediate eviction rather than waiting for natural TTL on every poisoned entry. Three lockstep sites updated atomically per the cache-prefix-bump- propagation-scope learning: - server/worldmonitor/intelligence/v1/_shared.ts:20 (canonical writer) - server/worldmonitor/news/v1/list-feed-digest.ts:511 (doc-comment) - scripts/ais-relay.cjs:3172 (relay inline helper) The news-classify-cache-prefix-audit static-analysis test enforces lockstep — passes after this change.

…ix (koala73#3480) * fix(classifier): match flashback/throwback after publisher brand prefix Brief 2026-04-28-0801 surfaced "CBS News Radio flashback: D-Day, Invasion of Normandy in 1944" as severity=critical in slot #3, two days after PR koala73#3429 shipped the historical-retrospective downgrade. Root cause: HISTORICAL_PREFIX_RE was anchored with `^` — `/^(?:science history|throwback|flashback)\s*:?/i` — so flashback / throwback only matched at title position 0. Publishers prepend their brand ("CBS News Radio", "BBC", "NPR") before the marker token, which defeated the anchor. The keyword path then matched `invasion` unimpeded → critical. Fix: word-boundary match for flashback / throwback (unambiguous markers regardless of position); science history stays anchored since it's an editorial-slot prefix that's only retrospective at the start of a headline. /(?:^science history\s*:?|\b(?:throwback|flashback)\b)/i Regression test added under the existing news-classifier-historical-downgrade suite covering CBS News Radio, BBC, and NPR brand-prefix forms. All 56 cases in that file plus the 56 cases across the historical-downgrade + LLM-cache-guard tests still pass. * fix(classifier): tighten flashback/throwback to editorial-slot syntax Reviewer P2 on PR koala73#3480: widening flashback/throwback to bare word-boundary match also broadens the L3b LLM-cache guard at list-feed-digest.ts:547, which force-demotes ANY cached LLM hit to info when hasHistoricalMarker is true. A current-event headline that incidentally uses "flashback" / "throwback" as a comparison word ("Markets see flashback to 2008 crisis as bonds tumble") would be wrongly suppressed even though the marker isn't acting as a retrospective framing token. Replace the broad word-boundary form with two explicit branches: 1. ANCHORED: marker at title position 0 (original behavior — "Throwback Thursday:", "Flashback:", "Science history:"). 2. BRAND-PREFIX: 1-4 Title-Case prefix words + marker + optional qualifier word + colon. Catches the brand-prefixed forms ("CBS News Radio flashback:", "BBC Throwback Thursday:", "NPR Flashback Friday:") while rejecting sentence-form uses. The brand-prefix branch deliberately requires (a) Title-Case prefix words (rejects "markets see flashback…"), AND (b) an editorial-slot colon after the marker (rejects "Markets See Flashback To 2008 Crisis As Bonds Tumble"). The Title-Case gate uses a non-i regex so [A-Z] enforces actual capitalization rather than getting case-folded. Adds 5 negative tests for the realistic false-positive cases the reviewer was worried about — "Markets see flashback to 2008 crisis", "Stocks suffer flashback to March 2020 crash", "Tesla stock throwback after split", "AI flashback to 2023 boom: Nvidia earnings beat", and a lowercase-leading sentence form. All existing positive tests still pass (61/61 in the historical- downgrade + LLM-cache-guard suite). * chore(classifier): bump CLASSIFY_CACHE_PREFIX v4 → v5 Evicts cache entries that landed under the pre-publisher-prefix-fix classifier — brand-prefixed retrospective titles like "CBS News Radio flashback: D-Day, Invasion of Normandy in 1944" had been promoted to severity=critical via the `invasion` keyword and persisted in the classify cache. The new brand-prefix branch in _classifier.ts (PR koala73#3480 commits 5ec9892 + f61fae5) re-rules those rows on next touch; v5 forces immediate eviction rather than waiting for natural TTL on every poisoned entry. Three lockstep sites updated atomically per the cache-prefix-bump- propagation-scope learning: - server/worldmonitor/intelligence/v1/_shared.ts:20 (canonical writer) - server/worldmonitor/news/v1/list-feed-digest.ts:511 (doc-comment) - scripts/ais-relay.cjs:3172 (relay inline helper) The news-classify-cache-prefix-audit static-analysis test enforces lockstep — passes after this change.

greptile-apps Bot reviewed Apr 28, 2026

View reviewed changes

vercel Bot deployed to Preview April 28, 2026 05:20 View deployment

vercel Bot deployed to Preview April 28, 2026 05:35 View deployment

vercel Bot deployed to Preview April 28, 2026 05:46 View deployment

koala73 merged commit 07202db into main Apr 28, 2026
11 checks passed

koala73 deleted the fix/classifier-publisher-prefix-flashback branch April 28, 2026 05:58

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(classifier): match flashback/throwback after publisher brand prefix#3480

fix(classifier): match flashback/throwback after publisher brand prefix#3480
koala73 merged 3 commits into
mainfrom
fix/classifier-publisher-prefix-flashback

koala73 commented Apr 28, 2026 •

edited

Loading

Uh oh!

vercel Bot commented Apr 28, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Apr 28, 2026

Uh oh!

greptile-apps Bot Apr 28, 2026

Uh oh!

koala73 Apr 28, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		const HISTORICAL_PREFIX_RE =
		/(?:^science history\s*:?\|\b(?:throwback\|flashback)\b)/i;

Conversation

koala73 commented Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

vercel Bot commented Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greptile-apps Bot commented Apr 28, 2026

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Flowchart

Uh oh!

greptile-apps Bot Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

koala73 Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

koala73 commented Apr 28, 2026 •

edited

Loading

vercel Bot commented Apr 28, 2026 •

edited

Loading