* feat(resilience): add score confidence intervals via batch Monte Carlo
Weekly cron perturbs domain weights ±10% across 100 draws per country,
stores p05/p95 in Redis. Score handler reads intervals and includes
them in the API response as ScoreInterval { p05, p95 }.
Proto field 14 (score_interval) added to GetResilienceScoreResponse.
* chore: regenerate proto types and OpenAPI docs for ScoreInterval
* fix(resilience): add seed-meta + lock + fix interval cache + percentile formula
1. Write seed-meta:resilience:intervals for health monitoring
2. Add distributed lock to prevent concurrent cron overlap
3. Move scoreInterval read outside 6h score cache boundary
4. Fix percentile index from floor to ceil-1 (nearest-rank)
* fix(health): add resilience:intervals to health + seed-health registries
* fix(seed): skip seed-meta on no-op runs + register intervals in health check
* feat(resilience): populate dataVersion field from seed-meta timestamp
Sets dataVersion to the ISO date of the most recent static bundle
seed, making the data vintage visible to API consumers.
* fix(resilience): bump score cache to v7 for dataVersion field addition
* feat(resilience): baseline vs stress scoring engine
Splits the resilience index into structural capacity (baselineScore)
and active disruption (stressScore) using the dimension type tags from
RESILIENCE_DIMENSION_TYPES (baseline/stress/mixed).
overallScore = baselineScore * (1 - stressFactor) where stressFactor
is clamped to [0, 0.5]. Mixed dimensions contribute to both scores.
Proto fields 10-12 added (baseline_score, stress_score, stress_factor).
Widget updated to display baseline/stress breakdown.
Cache keys bumped v4 -> v5 for atomic rollout.
* fix(resilience): bump history key to v2 for baseline/stress formula change
The overallScore formula changed from domain-weighted-sum to
baselineScore * (1 - stressFactor). Old history entries are
incomparable, causing fake change30d drops of -20 to -30 points.
Versioned history key starts a clean series.
* refactor(resilience): remove Cronbach alpha, add imputationShare confidence
Remove cronbach_alpha from proto (field 5 reserved) and all response
builders. Replace with imputationShare (field 9): the fraction of
weighted score from imputed (not observed) data.
lowConfidence now triggers on averageCoverage < 0.55 or
imputationShare > 0.40, replacing the unstable Cronbach-based gate.
weightedBlend() return type extended with observedWeight/imputedWeight
for provenance tracking through the scoring pipeline.
* fix(resilience): version cache key + fix IMF proxy imputation classification
1. Bump resilience score cache key to v2 to avoid serving stale
cached responses missing imputationShare after deploy.
2. Add explicit `imputed` flag to WeightedMetric so proxy data
(real IMF inflation with lower certaintyCoverage) is classified
as observed, not imputed. Only synthetic absence-based scores
count toward imputationShare.
* feat(resilience): Phase 1 §3.2+§3.3 — full-country sanctions counts + grey-out ranking
§3.2 — Switch sanctions from top-12 to full country counts
- RESILIENCE_SANCTIONS_KEY: 'sanctions:pressure:v1' → 'sanctions:country-counts:v1'
- New key is a plain ISO2→entryCount map covering ALL countries (no top-12 truncation)
- Replaces compound pressure formula with normalizeSanctionCount() piecewise scale:
0=100, 1-10=90-75, 11-50=75-50, 51-200=50-25, 201+=25→0
- IMPUTE.ofacSanctions removed (country-counts covers all countries; no absent-country
imputation needed for sanctions)
§3.3 — Grey-out criteria for ranking
- Proto: ResilienceRankingItem.overall_coverage (field 5) + GetResilienceRankingResponse.greyed_out
- GREY_OUT_COVERAGE_THRESHOLD = 0.40: countries below this are excluded from ranking
but still appear on choropleth in "insufficient data" style
- buildRankingItem() now computes overallCoverage from domain/dimension data
- getResilienceRanking() splits items into ranked (≥0.40) + greyedOut (<0.40)
Tests updated for new sanctions format; overall score anchor updated (67.56).
* fix(resilience): fix ranking cache guard for all-greyed-out + stale shape cases
Two cache bugs:
1. Empty-items guard: `cached?.items?.length` fails when every country falls
below GREY_OUT_COVERAGE_THRESHOLD (items=[], greyedOut=[…]). The cache was
written correctly but never served, causing unnecessary rewarming on every
request for sparse-data deployments.
Fix: `cached != null && (items.length > 0 || greyedOut.length > 0)`
2. Stale-shape test: agent-written cache test stored a payload without
`greyedOut` or `overallCoverage`, locking in pre-PR shape. Updated to the
correct post-deploy shape so the test reflects actual cached content.
Cache key was already bumped to resilience:ranking:v2 (forces fresh compute
on first post-deploy request, avoiding old-shape responses in production).
* fix(resilience): consume greyedOut on choropleth; version ranking cache key
- Add 'insufficient_data' level to ResilienceChoroplethLevel and RESILIENCE_CHOROPLETH_COLORS
- Extend buildResilienceChoroplethMap to accept optional greyedOut array
- Thread greyedOut through DeckGLMap.setResilienceRanking, MapContainer.setResilienceRanking (with replay), and data-loader.ts
- Add 'Insufficient data' tooltip guard for greyed-out countries in DeckGLMap
- Bump RESILIENCE_RANKING_CACHE_KEY to resilience:ranking:v2 to invalidate stale schema-mismatched cache entries
- Update api/health.js probe key to match
* fix(resilience): include greyedOut in seed-meta count to avoid false health alert
seed-meta:resilience:ranking was written with count=response.items.length,
which excludes greyedOut countries. In an all-greyed-out deployment, count=0
causes api/health.js to report the ranking as EMPTY_DATA/critical even though
the cached payload is valid (items:[], greyedOut:[…]).
Fix: count = items.length + greyedOut.length — total scoured countries
regardless of ranking eligibility.
* test(resilience): pin all-greyed-out cache-hit regression
Adds the missing test case: cached payload with items=[] and greyedOut=[…]
must be served from cache without triggering score rewarming.
Previously, `cached?.items?.length` was falsy for this shape, making the
guard ineffective. The fix (items.length > 0 || greyedOut.length > 0) was
correct but unpinned — this test locks it in.
* feat(resilience): add service proto and stub handlers
Add the worldmonitor.resilience.v1 proto package, generated client/server artifacts, edge routing, and zero-state handler stubs so the domain is deployable before the seed and scoring layers land.
Validation:
- PATH="/Users/lucaspassos/go/bin:/Users/lucaspassos/.codex/tmp/arg0/codex-arg06nbVvG:/Users/lucaspassos/.antigravity/antigravity/bin:/Users/lucaspassos/.local/bin:/Users/lucaspassos/.codeium/windsurf/bin:/Users/lucaspassos/Library/Python/3.12/bin:/Library/Frameworks/Python.framework/Versions/3.12/bin:/usr/local/bin:/System/Cryptexes/App/usr/bin:/usr/bin:/bin:/usr/sbin:/sbin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/local/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/appleinternal/bin:/opt/pkg/env/active/bin:/opt/pmk/env/global/bin:/Library/Apple/usr/bin:/opt/homebrew/bin:/Applications/Codex.app/Contents/Resources" make generate
- PATH="/Users/lucaspassos/go/bin:/Users/lucaspassos/.codex/tmp/arg0/codex-arg06nbVvG:/Users/lucaspassos/.antigravity/antigravity/bin:/Users/lucaspassos/.local/bin:/Users/lucaspassos/.codeium/windsurf/bin:/Users/lucaspassos/Library/Python/3.12/bin:/Library/Frameworks/Python.framework/Versions/3.12/bin:/usr/local/bin:/System/Cryptexes/App/usr/bin:/usr/bin:/bin:/usr/sbin:/sbin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/local/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/appleinternal/bin:/opt/pkg/env/active/bin:/opt/pmk/env/global/bin:/Library/Apple/usr/bin:/opt/homebrew/bin:/Applications/Codex.app/Contents/Resources" npx tsx --test tests/route-cache-tier.test.mjs tests/edge-functions.test.mjs
- npm run typecheck (fails on upstream Dodo/Clerk baseline)
- npm run typecheck:api (fails on upstream vitest baseline)
- npm run test:data (fails on upstream dodopayments-checkout baseline via tests/runtime-config-panel-visibility.test.mjs)
* fix(resilience): add countryCode validation to get-resilience-score
Throw ValidationError when countryCode is missing instead of silently
returning a zero-state response with an empty string country code.
* fix(resilience): validate countryCode format and mark required in spec
- Trim whitespace and reject non-ISO-3166-1 alpha-2 codes to prevent
cache pollution from malformed aliases (e.g. 'USA', ' us ', 'foobar')
- Add required: true to proto QueryConfig so generated OpenAPI spec
matches runtime validation behavior
- Regenerated OpenAPI artifacts via make generate
---------
Co-authored-by: Elie Habib <elie.habib@gmail.com>
