This surface runs Warconomy's data-integrity checks deterministically at build time over the committed data and reports the result. It is a static audit — there is no runtime monitoring, no scheduled job, and no external service. The same invariants are enforced locally by `npm run audit:data`, so a published build always shows every check passing. Checks cover source references, observation/graph/route/distribution/snapshot integrity, structured-data visibility, and the no-real-time / partial-coverage disclosure rules.
- 41 / 41 checks passing · 0 failing.
- Static build-time checks — not runtime monitoring, not scheduled.
- Machine-readable at /data-audit/data.json.
Summary
Source references 3/3
Every quantitative claim links to a resolvable source.
- every observation has a sourceId
- every observation sourceId resolves
- every fact has a resolvable sourceId
Observation integrity 3/3
Observations carry a data mode, as-of date, and confidence.
- dataMode is live or sample
- every observation has an asOf date
- every observation has a confidence
Graph integrity 3/3
The citation graph has no dangling edges and a consistent summary.
- no dangling edges
- nodeCount matches nodes
- edgeCount matches edges
Route / path integrity 1/1
Subject pages resolve for live indicators.
- live observations resolve to a page
Dataset distribution integrity 2/2
CSV/JSONL distributions are enumerated with format and path.
- at least 25 distributions
- each distribution has a format and path
Version snapshot integrity 3/3
Frozen payloads exist for materialized versions and the registry is consistent.
- each frozen version has a committed payload
- each frozen version has a manifest fingerprint
- current version is materialized
JSON-LD / FAQ visibility 1/1
FAQ JSON-LD only where a visible FAQ is rendered.
- every FAQ path has visible questions
Promotion log integrity 5/5
Promoted entries are sourced and link real observations; dry-run/blocked entries carry no value; the log routes and exports are cataloged.
- every promoted entry has a source id
- no dry-run/blocked entry carries a promoted value or observation
- promoted entries link to real observation ids
- every blocked entry states a blocker reason
- the promotion-log page + machine exports are cataloged
Source-readable refresh harness 5/5
Every source-gated candidate has a documented gate and route checklist, the manual workflows define their required fields, and no blocked candidate is marked live.
- every refresh candidate has a source of record, blocker, and verification steps
- every candidate has a route/update checklist + fields to promote
- every source workflow defines accepted formats, required fields, and rules
- the manual-transcription + promotion checklists are populated
- every candidate linked to a data need resolves to a real one
Coverage-depth integrity 5/5
Live values are sourced, every coverage-depth candidate is triaged, and no gated item was promoted without a source.
- every live observation resolves to a source (provenance)
- every coverage-depth data need has a triage status
- every attempted/blocked candidate documents its gate
- the SIPRI 2025 deepening is applied with a source
- no blocked grain/commodity sample was silently promoted to live
Answer-engine surfaces 5/5
The answer-hub, query-matrix, FAQ-caveat, structured-data, crawler, and methodology surfaces are all registered (Trains 126–140).
- answer hubs + query matrix pages registered
- query + FAQ CSV/JSONL exports registered
- FAQ-by-caveat + structured-data surfaces registered
- full crawler brief + crawl priorities registered
- standing answer-engine methodology pages registered
Static route budget 3/3
Static page counts stay under the hard build-cost budgets (warnings allowed).
- no route family is over its hard page budget
- total static pages under hard budget
- per-version expansion under hard budget
No real-time / complete-coverage overclaim 2/2
Public status copy negates real-time and complete-coverage claims.
- site status says not real-time
- site status discloses partial coverage
Related
See the data review queue, data maintenance, review actions, source quality, the QA report, the data dictionary, and the platform inventory. Machine-readable: data-audit/data.json.