fix(resilience): apply 7 review fixes that didn't reach main

[koala73]

Summary

PRs #2660 and #2662 merged to feature branches, not main. When #2673 reimplemented handlers for main, it omitted the review fixes we applied during code review. This PR backports all 7 fixes.

Fixes

#	Issue	Impact
1	cronbach > 0 && guard skips negative alpha	Countries with internally inconsistent dimensions show lowConfidence: false
2	cachedFetchJson timeout 300ms for 15+ Redis RTTs	Score computation likely times out, returning overallScore: 0
3	Duplicate round() function	Imported from resilience-stats instead
4	runRedisPipeline swallows HTTP errors silently	Restored console.warn with status code
5	resilienceRanking not in health.js	No staleness monitoring for ranking cache
6	No seed-meta write in ranking handler	Health can't report freshness
7	Ranking + meta written independently	Now atomic via single pipeline

Test plan

• npm run typecheck + npm run typecheck:api
• npx tsx --test tests/resilience-handlers.test.mts tests/resilience-stats.test.mts tests/resilience-dimension-scorers.test.mts
• Verify health endpoint shows resilienceRanking after first Pro ranking request

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ready	Preview, Comment	Apr 5, 2026 1:58pm

[greptile-apps]

Greptile Summary

This PR backports 7 review fixes from feature branches that were omitted when PR #2673 reimplemented the resilience handlers for main. All fixes are targeted, well-scoped, and address real production bugs (silent lowConfidence: false on inconsistent data, silent HTTP errors, missing health monitoring, non-atomic cache writes).

Key changes:

• Fix Batch market and RSS fetching with progressive updates #1 (_shared.ts): computeLowConfidence now evaluates cronbach < 0.55 without a cronbach > 0 && guard, so negative alpha (internally inconsistent dimensions) correctly sets lowConfidence: true instead of silently masking the problem.
• Fix Add smart zoom visibility for dense map layers #2 (_shared.ts): cachedFetchJson now called with negativeTtlSeconds = 2000; previously a short value caused score computation to time out, returning overallScore: 0.
• Fix Add shareable map state URLs and Copy Link button #3 (_shared.ts + resilience-stats.ts): Duplicate round() removed; imported from resilience-stats as the canonical implementation.
• Fix Add related-asset cards to clusters and map highlighting for infrastructure #4 (redis.ts): runRedisPipeline restores console.warn with the HTTP status code on non-OK responses, preventing silent swallowing of Redis write failures.
• Fix Add pipeline flow disruption signals #5 (health.js): resilienceRanking added to STANDALONE_KEYS, SEED_META (key seed-meta:resilience:ranking, maxStaleMin: 720), and ON_DEMAND_KEYS so the health endpoint can track ranking cache freshness.
• Fix Fetch AIS signals from API instead of hardcoded data #6 (get-resilience-ranking.ts): Ranking handler now writes seed-meta:resilience:ranking alongside the ranking cache, so health can report staleness.
• Fix Add power grid stress & blackout alert layer #7 (get-resilience-ranking.ts): Ranking data and seed-meta are written atomically via a single runRedisPipeline call, preventing the health check from seeing a ranking key without a corresponding meta key.

Confidence Score: 4/5

Safe to merge — all 7 fixes are targeted bug fixes with no new functionality, and the logic is straightforward and well-understood

Score of 4 (not 5) because (a) the diff is not available for direct inspection so analysis is based on the current file state, and (b) fix #2 (negativeTtlSeconds change) cannot be fully verified without seeing the before state. All visible code paths look correct.

server/worldmonitor/resilience/v1/_shared.ts — the negativeTtlSeconds=2000 argument is unreachable (fetcher never returns null) and could mislead future readers; server/worldmonitor/resilience/v1/get-resilience-ranking.ts — meta key constant is defined locally rather than co-located with sibling constants in _shared.ts

Important Files Changed

Filename	Overview
api/health.js	Correctly registers resilienceRanking in STANDALONE_KEYS, SEED_META (maxStaleMin: 720), and ON_DEMAND_KEYS for health staleness monitoring
server/_shared/redis.ts	Restores console.warn with HTTP status code in runRedisPipeline, preventing silent swallowing of Redis HTTP errors
server/_shared/resilience-stats.ts	Exports round() as the canonical rounding helper for the resilience subsystem
server/worldmonitor/resilience/v1/_shared.ts	Imports round from resilience-stats (dedup), fixes computeLowConfidence to flag negative cronbach alpha, and calls cachedFetchJson with negativeTtlSeconds=2000
server/worldmonitor/resilience/v1/get-resilience-ranking.ts	Adds seed-meta write for health monitoring and atomically writes ranking + meta via single pipeline when all country scores are available

Sequence Diagram

sequenceDiagram
    participant Client
    participant RankingHandler as get-resilience-ranking.ts
    participant SharedLib as _shared.ts
    participant Redis as Redis (Upstash)
    participant Health as api/health.js

    Client->>RankingHandler: getResilienceRanking()
    RankingHandler->>Redis: getCachedJson('resilience:ranking')
    Redis-->>RankingHandler: null (cache miss)
    RankingHandler->>SharedLib: listScorableCountries()
    SharedLib->>Redis: getCachedJson(RESILIENCE_STATIC_INDEX_KEY)
    Redis-->>SharedLib: country list
    SharedLib-->>RankingHandler: ["US", "DE", ...]
    RankingHandler->>SharedLib: getCachedResilienceScores(countryCodes)
    SharedLib->>Redis: runRedisPipeline([GET score:US, GET score:DE, ...])
    Redis-->>SharedLib: cached scores
    SharedLib-->>RankingHandler: Map<code, score>
    Note over RankingHandler: warmMissingResilienceScores() if needed
    RankingHandler->>SharedLib: ensureResilienceScoreCached(missing[])
    SharedLib->>Redis: cachedFetchJson (score computation)
    Redis-->>SharedLib: stored
    RankingHandler->>RankingHandler: sortRankingItems()
    Note over RankingHandler: Fix #7: atomic write
    RankingHandler->>Redis: runRedisPipeline([\n  SET resilience:ranking EX 21600,\n  SET seed-meta:resilience:ranking EX 604800\n])
    RankingHandler-->>Client: { items: [...] }
    Health->>Redis: STRLEN resilience:ranking
    Health->>Redis: GET seed-meta:resilience:ranking
    Redis-->>Health: { fetchedAt, count }
    Health->>Health: seedAge > 720min? → STALE_SEED

Loading

Comments Outside Diff (1)

1. server/worldmonitor/resilience/v1/_shared.ts, line 194 (link)

   negativeTtlSeconds = 2000 is unreachable

   The fetcher passed to cachedFetchJson always returns a fully-populated GetResilienceScoreResponse object and never returns null, so the negativeTtlSeconds parameter is never exercised. The cachedFetchJson function caches NEG_SENTINEL only when the fetcher returns null; since that path is unreachable here, this value has no effect at runtime.

   This is harmless, but it may cause confusion for the next reader who wonders why 2000 seconds was chosen. A brief comment clarifying intent would help:

_{Reviews (1): Last reviewed commit: "fix: revert cachedFetchJson arg change (..." | Re-trigger Greptile}

[greptile-apps]

Meta key constant not co-located with sibling constants

RESILIENCE_RANKING_META_KEY and RESILIENCE_RANKING_META_TTL_SECONDS are defined locally here, but all related resilience cache constants (RESILIENCE_RANKING_CACHE_KEY, RESILIENCE_RANKING_CACHE_TTL_SECONDS, etc.) live in _shared.ts. The meta key string 'seed-meta:resilience:ranking' is also hardcoded independently in api/health.js (line 257). If the key ever changes, both sites must be updated manually.

Consider moving these to _shared.ts and exporting them:

// in _shared.ts
export const RESILIENCE_RANKING_META_KEY = 'seed-meta:resilience:ranking';
export const RESILIENCE_RANKING_META_TTL_SECONDS = 7 * 24 * 60 * 60;

Then import in the handler and reference from health.js as documentation.