fix(intelligence): publish CII risk scoring methodology + de-dup country constants (#3725) by koala73 · Pull Request #3780 · koala73/worldmonitor

koala73 · 2026-05-18T06:00:25Z

Summary

/api/intelligence/risk-scores emits per-country Composite Instability Index (CII) scores whose editorial weights — baselineRisk, eventMultiplier, and the strategic-risk roll-up coefficients — were undisclosed magic numbers living next to each other in get-risk-scores.ts. There was no way for an API consumer to detect coefficient drift between deploys or audit which values produced a given score.

This PR closes the disclosure gap and folds in #3722 (strict subset — also asked for methodology publication).

Changes

Proto — added CiiScore.methodology_version (string) and CiiScore.event_multiplier (double) so every score carries its provenance on the wire. Regenerated via make generate.
Editorial coefficients module — server/worldmonitor/intelligence/v1/_risk-config.ts now owns CII_FORMULA_VERSION, the strategic-risk positional decay (0.15), scale floor (15), scale factor (0.7), and top-N window (5), each with a rationale comment and a "bump the version + update the doc + CHANGELOG" change protocol header.
Published methodology — docs/methodology/cii-risk-scores.md covers the four CII components (U/C/S/I) qualitatively, lists the per-country baselineRisk/eventMultiplier table, documents the strategic-risk band derivation, and explicitly flags drift between the server tables and the frontend CURATED_COUNTRIES table.
De-dup note — the server's BASELINE_RISK / EVENT_MULTIPLIER are a near-copy of CURATED_COUNTRIES in src/config/countries.ts, but they have drifted in 12 cells across 7 codes (AF, EG, IQ, JP, KR, LB, QA). Server values remain authoritative on the wire (we did not silently reconcile either side); the divergence is now visible in the methodology doc with an explicit Drift vs CURATED_COUNTRIES column. If a reviewer wants the server to adopt frontend values (or vice versa), that is a follow-up — call it out and bump CII_FORMULA_VERSION.
UI tooltip — the components.cii.infoTooltip locale string in src/locales/en.json now points to /docs/methodology/cii-risk-scores. Only en.json is updated; the translation team will pick up the other locales in their normal cadence.
Tests — extended tests/cii-scoring.test.mts from 24 → 35 tests:
- every CiiScore carries methodologyVersion === CII_FORMULA_VERSION and a finite eventMultiplier > 0
- staticBaseline matches CURATED_COUNTRIES for the 8 representative no-drift codes
- computeStrategicRisks output is in [STRATEGIC_RISK_SCALE_FLOOR=15, 100] and equals exactly 15 when all top-N CII scores are 0
- positional-decay weights match the documented [1.00, 0.85, 0.70, 0.55, 0.40]
- drift guard: docs/methodology/cii-risk-scores.md must list every CURATED_COUNTRIES and every TIER1_COUNTRIES code, and must reference the current CII_FORMULA_VERSION string — catches "added a country but forgot the doc" and "changed the version but forgot the doc"

Closes #3725
Closes #3722 (duplicate — #3725 is a strict superset)

Test plan

npm run typecheck — clean
npm run typecheck:api — clean
npm run test:data — 9107 / 9107 passing (24 → 35 in cii-scoring.test.mts)
make generate — proto regen clean, new fields visible in src/generated/server/.../service_server.ts
Lint on changed files — no new findings (pre-existing noExcessiveCognitiveComplexity on computeCIIScores is unchanged)
Pre-push hooks pass (TS check, edge-function bundle check, MDX/MD lint, version sync, Unicode safety)

Review round 2 — `f564e98`

Folded in CHANGES-REQUESTED items from the round-1 review (no blockers, all should-fixes):

Cache prefix bumped v1 → v2 — risk:scores:sebuf:v1 and risk:scores:sebuf:stale:v1 now :v2 across every reader/writer (15 source files + 2 test files). The proto change adds REQUIRED methodologyVersion + eventMultiplier; old :v1 payloads in Redis violated the new shape on read. Seed-meta key (seed-meta:risk:scores:sebuf) is intentionally unchanged — that's freshness tracking, not payload.
event_multiplier upper bound — added (buf.validate.field).double.lte = 10 on proto/.../intelligence.proto. De-facto editorial max is ~3.0 (KP); 10 gives ~3.3x headroom while still catching obviously-wrong values. Refreshed openapi schemas via make generate (maximum: 10 visible in docs/api/IntelligenceService.openapi.{yaml,json}).
i18n tooltip split — the methodology link is now its own key components.cii.methodologyLink (sibling to infoTooltip). en.json infoTooltip reverted to pre-Undisclosed algorithmic bias: hardcoded country event multipliers (up to 10x) and arbitrary formula coefficients in CII risk scores #3725 wording; the new key shipped to all 21 locale files (English text used as stop-gap in the 20 non-en locales — translators can localize independently of the long-form tooltip). CIIPanel.ts concatenates both at render time. Sibling _methodologyLink_translatorNote in en.json flags the new key for translation pipelines (JSON has no comment syntax).
Drift test hardened — tests/cii-scoring.test.mts previously checked only row existence (\|\s${code}\s\|). New test parses each row's numeric columns and asserts BASELINE_RISK[code] + EVENT_MULTIPLIER[code] match the methodology doc exactly. Required exporting both tables from get-risk-scores.ts. Adds 1 test (9107 → 9108).
Drift reconciliation deferred to follow-up — opened #3789 to track the 7-country baselineRisk / eventMultiplier drift (AF, EG, IQ, JP, KR, LB, QA). Linked from docs/methodology/cii-risk-scores.md § 2.

Verification (round 2)

make generate — clean, openapi reflects maximum: 10 on eventMultiplier
npm run typecheck — clean
npm run typecheck:api — clean
npm run test:data — 9108 / 9108 passing (+1 from new drift-guard test)
npm run lint — no NEW warnings on touched files (pre-existing complexity warnings on computeCIIScores, buildOilStocksCover, useOptionalChain lints unchanged)

mintlify · 2026-05-18T06:00:29Z

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
WorldMonitor	🟢 Ready	View Preview	May 18, 2026, 6:01 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

vercel · 2026-05-18T06:00:30Z

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

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ready	Preview, Comment	May 18, 2026 8:27am

greptile-apps · 2026-05-18T06:05:02Z

Greptile Summary

This PR publishes the previously opaque CII scoring methodology by adding methodology_version and event_multiplier fields to every CiiScore on the wire, extracting all editorial coefficients into a new _risk-config.ts module, and publishing a docs/methodology/cii-risk-scores.md note that documents the formula, per-country tables, and drift against the frontend CURATED_COUNTRIES table.

Proto + generated code: two new fields (methodology_version, event_multiplier) added to CiiScore (fields 8–9); generated TypeScript stubs regenerated via make generate.
_risk-config.ts: extracts CII_FORMULA_VERSION, positional-decay, scale-floor/factor, and top-N window into documented, versioned constants with a change protocol that requires bumping the version string and updating the methodology doc atomically.
Tests (24 → 35): adds disclosure-field assertions, strategic-risk floor checks, positional-decay weight verification, and a drift-guard that fails loudly if a country is added to either table without updating the methodology doc.

Confidence Score: 3/5

The methodology publication and new proto fields are well-structured, but the unchanged Redis cache keys mean that a deployed instance may serve cached responses missing the newly guaranteed fields for up to an hour after rollout.

The core correctness concern is the stale Redis cache: both risk:scores:sebuf:v1 (10-min TTL) and risk:scores:sebuf:stale:v1 (1-hour TTL) will serve pre-deploy payloads that have no methodologyVersion or eventMultiplier. Clients reading the new fields get proto defaults ("" and 0.0) during that window, breaking the invariant the PR sets out to guarantee. The active cache resolves in 10 minutes, but the stale path persists for a full hour. Bumping both keys to v2 is a one-line fix.

server/worldmonitor/intelligence/v1/get-risk-scores.ts — the RISK_CACHE_KEY and RISK_STALE_CACHE_KEY constants need to be bumped before or alongside deploy.

Important Files Changed

Filename	Overview
server/worldmonitor/intelligence/v1/get-risk-scores.ts	Adds `eventMultiplier` and `methodologyVersion` to every `CiiScore` output and imports coefficients from `_risk-config.ts`; cache keys are not bumped, so cached responses will serve proto-default values for the new fields until TTL expiry.
server/worldmonitor/intelligence/v1/_risk-config.ts	New module extracting all editorial CII coefficients with change-protocol comments; minor comment inaccuracy on `STRATEGIC_RISK_POSITIONAL_DECAY` (claims weight reaches zero at position 6, actually 0.10) contradicts the adjacent `STRATEGIC_RISK_TOP_N` JSDoc.
proto/worldmonitor/intelligence/v1/intelligence.proto	Adds `methodology_version` (field 8) and `event_multiplier` (field 9) to `CiiScore`; pre-existing `gte = 0` on `dynamic_score` is inconsistent with the now-published formula where `dynamicScore = composite − baseline` can be negative for low-baseline countries.
docs/methodology/cii-risk-scores.md	Well-structured methodology note covering all four CII components, the per-country baseline/multiplier table with explicit drift annotations, and the strategic-risk roll-up formula. Matches the code formulas and is enforced by the new drift-guard tests.
tests/cii-scoring.test.mts	Extends test suite from 24 to 35 cases; new tests verify `methodologyVersion`, `eventMultiplier`, positional-decay weights, strategic-risk floor, and drift-guard assertions that tie the methodology doc to live country tables.
src/locales/en.json	Adds a methodology link to the CII tooltip; change is scoped to English locale only, with other locales deferred to translation team cadence.

Sequence Diagram

sequenceDiagram
    participant Client
    participant Handler as getRiskScores handler
    participant Cache as Redis Cache
    participant ACLED
    participant Aux as Auxiliary Sources

    Client->>Handler: GET /api/intelligence/risk-scores

    Handler->>Cache: getCachedJson('risk:scores:sebuf:v1')
    alt cache hit (up to 600s TTL)
        Cache-->>Handler: cached GetRiskScoresResponse (may lack methodologyVersion/eventMultiplier if written pre-deploy)
        Handler-->>Client: response (fields may be proto-default empty-string / 0.0)
    else cache miss
        Handler->>ACLED: fetchACLEDEvents()
        Handler->>Aux: fetchAuxiliarySources()
        ACLED-->>Handler: acledEvents[]
        Aux-->>Handler: AuxiliarySources
        Handler->>Handler: computeCIIScores() — sets methodologyVersion + eventMultiplier
        Handler->>Handler: computeStrategicRisks()
        Handler->>Cache: setCachedJson('risk:scores:sebuf:v1', result, 600)
        Handler->>Cache: setCachedJson('risk:scores:sebuf:stale:v1', result, 3600)
        Handler-->>Client: fresh GetRiskScoresResponse with methodologyVersion and eventMultiplier
    end

Comments Outside Diff (2)

server/worldmonitor/intelligence/v1/get-risk-scores.ts, line 621-622 (link)

Cache keys not bumped for new wire fields

Both RISK_CACHE_KEY ('risk:scores:sebuf:v1') and RISK_STALE_CACHE_KEY ('risk:scores:sebuf:stale:v1') are unchanged. A cached response written by the pre-deploy binary won't contain methodologyVersion or eventMultiplier; when the new binary reads it back, the proto defaults kick in — methodologyVersion = "" and eventMultiplier = 0.0. The stale cache path compounds this: RISK_STALE_TTL = 3600 means a pre-deploy stale entry survives for up to an hour after deploy, so any client or health check that asserts eventMultiplier > 0 or methodologyVersion === 'v1' will fail for that full window. Renaming to 'risk:scores:sebuf:v2' and 'risk:scores:sebuf:stale:v2' forces a cold start and closes the gap.
proto/worldmonitor/intelligence/v1/intelligence.proto, line 19-22 (link)

dynamic_score proto validation conflicts with the published methodology

The published methodology doc (section 2) states dynamicScore = composite − staticBaseline. For low-baseline countries (US, DE, GB, JP — all baselineRisk = 5) with no live events, composite ≈ round(5 × 0.4) = 2, so dynamicScore = 2 − 5 = −3. The gte = 0 constraint on dynamic_score would reject that value if validation is enforced. This inconsistency pre-dates the PR, but publishing the formula now makes the conflict visible to downstream consumers reading the methodology doc alongside the proto. Either the formula note should clarify the floor (max(0, composite - baseline)) or the validation should drop the lower bound.

_{Reviews (1): Last reviewed commit: "fix(intelligence): publish CII risk scor..." | Re-trigger Greptile}

greptile-apps · 2026-05-18T06:05:09Z

+ * Rationale: gives the most-affected country full weight, lets the next four
+ * contribute decreasing influence, and reaches zero weight at position 6 so
+ * the slice size of 5 is naturally bounded.


Contradictory claim about when the positional weight reaches zero

The comment says "reaches zero weight at position 6" but at position 6 (0-based), weight = 1 − 6 × 0.15 = 0.10, not zero. The weight only turns negative at position 7 (−0.05). The sibling STRATEGIC_RISK_TOP_N JSDoc in the same file correctly states "position 6 hits weight=0.10 (still contributing)", so the two comments directly contradict each other.

Suggested change

* Rationale: gives the most-affected country full weight, lets the next four

* contribute decreasing influence, and reaches zero weight at position 6 so

* the slice size of 5 is naturally bounded.

* Rationale: gives the most-affected country full weight, lets the next four

* contribute decreasing influence, and reaches weight 0.10 at position 6 —

* still non-trivial, which is why STRATEGIC_RISK_TOP_N caps the window at 5

* to keep the roll-up focused on the very top of the dashboard.

…ghten drift test, split tooltip key for i18n, bound event_multiplier Review-requested follow-ups for #3725 (PR #3780): - Bump CII risk-scores Redis cache prefix `risk:scores:sebuf:v1` → `risk:scores:sebuf:stale:v2` across every reader (get-risk-scores writer, get-country-risk, chat-analyst-context, brief-story-context, _shared/ cache-keys, api/bootstrap, api/health, api/mcp, regional-snapshot/*, seed-cross-source-signals, seed-forecasts, plus the two tests that hardcode these keys). The proto change in #3725 added REQUIRED methodologyVersion + eventMultiplier; pre-existing v1 payloads no longer satisfy the shape. - Split the methodology link into its own i18n key `components.cii.methodologyLink` so the transparency disclosure ships in all 21 locales immediately without holding the feature for translation. Revert `components.cii.infoTooltip` in `src/locales/en.json` to the pre-#3725 wording. Add the new key to all 20 non-en locale files with the English value as stop-gap; translators can localize asynchronously. CIIPanel.ts concatenates both keys for render. Embed a sibling `_methodologyLink_translatorNote` in en.json with a translation-pipeline TODO. - Tighten the methodology-doc drift guard (`tests/cii-scoring.test.mts`) — previously checked only that a row exists for each code; now parses the numeric columns and asserts every baseline + multiplier in the doc matches `BASELINE_RISK[code]` and `EVENT_MULTIPLIER[code]` exactly. Catches silent drift between code and doc. Required exporting `BASELINE_RISK` / `EVENT_MULTIPLIER` from `get-risk-scores.ts`. - Add `lte = 10` upper bound on `CiiScore.event_multiplier` in `proto/.../intelligence.proto` (de-facto editorial max is ~3.0; 10 gives ~3.3x headroom while still catching obviously-wrong values). Refreshed openapi schemas via `make generate`. - File issue #3789 to track the deferred drift reconciliation (7 countries: AF, EG, IQ, JP, KR, LB, QA). Link it from the methodology doc. Verification: - `make generate` succeeds with the new proto bound (openapi schemas carry maximum: 10). - `npm run typecheck` and `npm run typecheck:api` clean. - `npm run test:data` 9108 passing (was 9107; +1 from the stricter drift test). - `npm run lint` no NEW warnings on touched files (pre-existing complexity warnings on `computeCIIScores`, `buildOilStocksCover`, etc. are unchanged).

…try constants (#3725, closes #3722) The /api/intelligence/risk-scores endpoint emits a per-country Composite Instability Index whose editorial weights (baselineRisk, eventMultiplier, the positional-decay step, scale-floor, and scale-factor) were undisclosed magic numbers — see #3725 (and strict subset #3722). API consumers had no way to detect drift or audit the formula. This change: - Adds CiiScore.methodology_version (string) and CiiScore.event_multiplier (double) to the proto so every score carries its provenance on the wire (regenerated via `make generate`). - Extracts the strategic-risk roll-up coefficients (positional decay, scale floor, scale factor, top-N) into a named module server/worldmonitor/intelligence/v1/_risk-config.ts with a change-protocol header. - Publishes the methodology in docs/methodology/cii-risk-scores.md including the per-country baseline + multiplier table, the U/C/S/I component meanings, and the strategic-risk band derivation. - Documents the drift between server BASELINE_RISK / EVENT_MULTIPLIER and the frontend CURATED_COUNTRIES table (12 cells across 7 codes) rather than silently reconciling either side — server values remain authoritative on the wire. - Adds an `Includes per-country baseline + event multiplier — see /docs/methodology/cii-risk-scores for the published table` sentence to the cii infoTooltip locale (en.json only — translation team can catch up). - Extends tests/cii-scoring.test.mts (24 -> 35 tests) with disclosure-field assertions, strategic-risk roll-up coverage (floor/cap, all-zero case, empty input), and a doc-drift guard that fails loudly when CURATED_COUNTRIES or TIER1_COUNTRIES gains a code not listed in the methodology table. Verification: `npm run typecheck`, `npm run typecheck:api`, `npm run test:data` (9107 passing). Lint clean on changed files (pre-existing complexity warning on computeCIIScores is unchanged).

…ghten drift test, split tooltip key for i18n, bound event_multiplier Review-requested follow-ups for #3725 (PR #3780): - Bump CII risk-scores Redis cache prefix `risk:scores:sebuf:v1` → `risk:scores:sebuf:stale:v2` across every reader (get-risk-scores writer, get-country-risk, chat-analyst-context, brief-story-context, _shared/ cache-keys, api/bootstrap, api/health, api/mcp, regional-snapshot/*, seed-cross-source-signals, seed-forecasts, plus the two tests that hardcode these keys). The proto change in #3725 added REQUIRED methodologyVersion + eventMultiplier; pre-existing v1 payloads no longer satisfy the shape. - Split the methodology link into its own i18n key `components.cii.methodologyLink` so the transparency disclosure ships in all 21 locales immediately without holding the feature for translation. Revert `components.cii.infoTooltip` in `src/locales/en.json` to the pre-#3725 wording. Add the new key to all 20 non-en locale files with the English value as stop-gap; translators can localize asynchronously. CIIPanel.ts concatenates both keys for render. Embed a sibling `_methodologyLink_translatorNote` in en.json with a translation-pipeline TODO. - Tighten the methodology-doc drift guard (`tests/cii-scoring.test.mts`) — previously checked only that a row exists for each code; now parses the numeric columns and asserts every baseline + multiplier in the doc matches `BASELINE_RISK[code]` and `EVENT_MULTIPLIER[code]` exactly. Catches silent drift between code and doc. Required exporting `BASELINE_RISK` / `EVENT_MULTIPLIER` from `get-risk-scores.ts`. - Add `lte = 10` upper bound on `CiiScore.event_multiplier` in `proto/.../intelligence.proto` (de-facto editorial max is ~3.0; 10 gives ~3.3x headroom while still catching obviously-wrong values). Refreshed openapi schemas via `make generate`. - File issue #3789 to track the deferred drift reconciliation (7 countries: AF, EG, IQ, JP, KR, LB, QA). Link it from the methodology doc. Verification: - `make generate` succeeds with the new proto bound (openapi schemas carry maximum: 10). - `npm run typecheck` and `npm run typecheck:api` clean. - `npm run test:data` 9108 passing (was 9107; +1 from the stricter drift test). - `npm run lint` no NEW warnings on touched files (pre-existing complexity warnings on `computeCIIScores`, `buildOilStocksCover`, etc. are unchanged).

…bump

…ator pipeline picks it up (#3796) PR #3780 (CII methodology link tooltip) shipped a new i18n key `components.cii.methodologyLink` in en.json and pre-populated the same English string into all 20 non-EN locale files as a stop-gap. Discovered during PR #3793 review: scripts/translate-locales.mjs:175 detects missing keys via `!(k in flat)`. Pre-populating the key makes every non-EN locale appear "already complete", so the translator script will never re-translate them — the English copy would be stranded in 20 locales forever. Fix: remove `components.cii.methodologyLink` from the 20 non-EN locale files (leaving en.json untouched). On the next backfill, the translator pipeline will see the key as missing and translate it properly. Same class of bug surfaced in #3793; this is the retroactive cleanup for the already-merged #3780 keys. Refs: #3780, #3793

…ng false positives (#3791) (#3793) * fix(disease-outbreaks): disclose alert-level classifier + fix substring false positives (#3791) The disease outbreaks panel labels each item with an ALERT / WARNING / WATCH badge derived from `detectAlertLevel` in `scripts/_disease-outbreaks-helpers.mjs`. Previously the function used naked `text.includes(...)` substring checks with no methodology disclosure to users — same defect class as #3725 (CII), at the smaller scale appropriate for a 6-keyword classifier. Changes - Extract `ALERT_KEYWORDS` and `WARNING_KEYWORDS` as frozen exports + bump to word-boundary regex matching. Prior substring matching could fire "epidemic" inside "antiepidemic" and "spread" inside "widespread", silently over-promoting items. Both regressions are now covered by tests. - Add a methodology header comment + `ALERT_LEVEL_METHODOLOGY_VERSION = 'v1'` so callers can pin a version and the change protocol is in-code. - Publish `docs/methodology/disease-alert-level.md` with the full keyword table, inputs, known limitations, and a change protocol (mirrors the pattern established by docs/methodology/cii-risk-scores.md in #3780). - Add `components.diseaseOutbreaks.methodologyNote` to en.json + 20 sibling locales (English stop-gap) so the disclosure ships in every locale on day one — same approach as #3780 review round 2 (i18next does not fall back on stale values for present keys). - Render the new key in `DiseaseOutbreaksPanel`'s infoTooltip after the existing copy. - Tests (+9 cases in `tests/disease-outbreaks-seed.test.mjs`): cover each keyword class, substring-false-positive regression guards, case-insensitive matching, title+desc concatenation, null/undefined inputs, alert-beats- warning precedence, frozen-keyword invariant, and version-string format. Closes #3791 * fix(disease-outbreaks): address review #3793 — restore translator pipeline visibility, ship version field, rename to avoid ALERT_KEYWORDS collision Restores the translator pipeline's visibility into the methodologyNote key: the prior approach pre-populated EN text in all 20 non-EN locales, which made scripts/translate-locales.mjs treat every locale as "already complete" and permanently skip translation. Now only en.json carries the key; the next translator run will fill in the missing locales naturally. Plumbs ALERT_LEVEL_METHODOLOGY_VERSION onto the published payload as `alertLevelMethodologyVersion`. The constant existed but nothing consumed it, so the change protocol's "bump the version" step had no observable effect. The version field now flows through diseasePublishTransform and lands on the canonical Redis key, giving clients a real signal when the editorial keyword set changes. Renames ALERT_KEYWORDS → DISEASE_ALERT_KEYWORDS and WARNING_KEYWORDS → DISEASE_WARNING_KEYWORDS to avoid name collision with the unrelated geopolitical ALERT_KEYWORDS export in src/config/feeds.ts (war/invasion/ nuclear). Exports the precompiled DISEASE_ALERT_RE and DISEASE_WARNING_RE regexes so future callers don't reach for text.includes(kw) and silently re-introduce the #3791 substring bug. Drops the brittle line-range pointer in the methodology doc in favor of the symbol name detectAlertLevel — symbols don't rot. Tests: 28/28 pass (was 26 — added version-field plumbing test and exported-regex word-boundary test). * fix(disease-outbreaks): #3793 round 3 — add alertLevelMethodologyVersion to proto contract (P2 review finding) The seeded JSON payload already carries alertLevelMethodologyVersion (round 2), but it was not declared on the proto contract — so generated clients and OpenAPI consumers couldn't see it. This closes the gap: - proto: add `string alert_level_methodology_version = 3` on ListDiseaseOutbreaksResponse with a comment pointing at the methodology doc. - generate: regenerated client/server TS + HealthService + worldmonitor OpenAPI specs via `make generate` (no hand-edits). - handler: server/worldmonitor/health/v1/list-disease-outbreaks.ts now reads the field off the cached payload and falls back to 'v1' so payloads cached before the field was published still satisfy the proto contract during the deploy → next-seed-publish window (transitional read tolerance). - src/services/disease-outbreaks.ts emptyOutbreaks constant gets the same 'v1' fallback so the offline/empty path satisfies the new required field. - tests/list-disease-outbreaks-handler.test.mjs (new): asserts the proto + generated bindings declare the field, plus behavioural coverage that the handler echoes the field when present and falls back to 'v1' when absent.

…ift callout (#3816) * docs(mintlify): publish CII + disease-alert methodology pages with drift callout Promotes the two methodology docs from PRs #3780 (CII risk scores) and #3793 (disease alert level) from plain markdown into the Mintlify documentation site so they render in the public docs nav under "Intelligence & Analysis". - Convert `docs/methodology/cii-risk-scores.md` → `.mdx` with Mintlify frontmatter and a prominent `<Warning>` callout at the top linking issue #3789, so API consumers know the server-side `BASELINE_RISK` / `EVENT_MULTIPLIER` tables are authoritative over the frontend `CURATED_COUNTRIES` for the 7 drifted countries (AF, EG, IQ, JP, KR, LB, QA) until reconciliation lands. - Convert `docs/methodology/disease-alert-level.md` → `.mdx` with Mintlify frontmatter. No drift callout — this one has no known drift. - Wire both pages into `docs/docs.json` after the existing methodology entries in the "Intelligence & Analysis" group. - Add a one-line cross-link from `docs/algorithms.mdx` (existing CII section) to the new methodology page. - Update `tests/cii-scoring.test.mts` to read `cii-risk-scores.mdx` (the four `resolve(..., 'cii-risk-scores.md')` paths); test logic and the drift/version assertions are unchanged. No production code is touched. The proto and server-side comments that still reference the `.md` path are left as-is per scope; the public URL slug (`/methodology/cii-risk-scores`) is unchanged and is already what `src/locales/en.json` and other consumers link to. * docs(comments): update .md → .mdx references in proto, source, and test comments Follow-on to the .md→.mdx rename in this PR: chase down all in-source doc-path references so they point at the actual file that ships in Mintlify. Touches: - proto headers (regenerated OpenAPI .yaml/.json picked up the rename) - _risk-config.ts, get-risk-scores.ts header comments - _disease-outbreaks-helpers.mjs, seed-disease-outbreaks.mjs comments - cii-scoring.test.mts inline comments + assertion messages No behavioural change — only doc-path strings inside comments and error messages.

koala73 mentioned this pull request May 18, 2026

data quality: CII risk scores apply undisclosed country-specific event multipliers up to 3×, making scores non-reproducible and politically opaque #3722

Closed

mintlify Bot deployed to staging - docs May 18, 2026 06:01 View deployment

vercel Bot deployed to Preview May 18, 2026 06:03 View deployment

greptile-apps Bot reviewed May 18, 2026

View reviewed changes

koala73 mentioned this pull request May 18, 2026

Reconcile CII baseline/multiplier drift between server and frontend (7 countries) #3789

Closed

mintlify Bot deployed to staging - docs May 18, 2026 06:51 View deployment

vercel Bot deployed to Preview May 18, 2026 06:53 View deployment

koala73 added 3 commits May 18, 2026 09:18

fix(tests): update freshness key references to v2 after cache prefix …

66a5518

…bump

koala73 force-pushed the fix/3725-cii-risk-methodology-disclosure branch from f564e98 to 66a5518 Compare May 18, 2026 08:23

mintlify Bot deployed to staging - docs May 18, 2026 08:24 View deployment

vercel Bot deployed to Preview May 18, 2026 08:27 View deployment

koala73 merged commit e633dde into main May 18, 2026
13 checks passed

koala73 deleted the fix/3725-cii-risk-methodology-disclosure branch May 18, 2026 08:28

koala73 mentioned this pull request May 18, 2026

docs(mintlify): publish CII + disease-alert methodology pages with drift callout #3816

Merged

7 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(intelligence): publish CII risk scoring methodology + de-dup country constants (#3725)#3780

fix(intelligence): publish CII risk scoring methodology + de-dup country constants (#3725)#3780
koala73 merged 3 commits into
mainfrom
fix/3725-cii-risk-methodology-disclosure

koala73 commented May 18, 2026 •

edited

Loading

Uh oh!

mintlify Bot commented May 18, 2026 •

edited

Loading

Uh oh!

vercel Bot commented May 18, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented May 18, 2026 •

edited

Loading

Comments Outside Diff (2)

Uh oh!

greptile-apps Bot May 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

koala73 commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Test plan

Review round 2 — f564e98

Verification (round 2)

Uh oh!

mintlify Bot commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vercel Bot commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greptile-apps Bot commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 3/5

Important Files Changed

Sequence Diagram

Comments Outside Diff (2)

Uh oh!

greptile-apps Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

koala73 commented May 18, 2026 •

edited

Loading

Review round 2 — `f564e98`

mintlify Bot commented May 18, 2026 •

edited

Loading

vercel Bot commented May 18, 2026 •

edited

Loading

greptile-apps Bot commented May 18, 2026 •

edited

Loading