Methodology
How the answer gets made
Collection, research, evidence checks, and editing are separate steps. The public site receives accepted evidence and concise publication records, not raw feeds or an evidence API.
Edition published
Editorial principles
The short version
- Facts trigger publication; new commentary alone waits for the next factual run.
- Nonzero traffic is not treated as proof of routine unrestricted passage.
- Primary claims stay attributed. Disagreement is shown instead of averaged away.
- A failed source keeps its last good value and cannot be replaced with a synthetic number.
- Every quantitative module states the upstream data-through date.
Internal records
What passes through the pipeline
The public receives concise output, not a raw evidence API.
- Candidate
- A deduplicated lead from a registered source or open research.
- Evidence item
- Accepted metadata, a short excerpt or paraphrase, provenance, and claim links. Full articles are not stored.
- Claim
- A classified factual statement with explicit supporting evidence IDs.
- Research run
- What was checked, accepted, rejected, disputed, and judged material.
- Live brief
- The concise editor output produced only from accepted evidence.
Source health
Quietly honest about the clocks
Warnings live here instead of becoming bright global alarms.
| Source | State | Data through | Consecutive failures | Note |
|---|---|---|---|---|
| IMF PortWatch | Healthy | 0 | Latest committed deterministic snapshots loaded successfully. | |
| JMIC / UKMTO | Research Fallback | 0 | Latest accepted advisory came through editorial research; deterministic product-page collection is not yet represented in this backfill. | |
| Associated Press | Research Only | 0 | Reporting is discovered during the research pass rather than scraped as a dataset. | |
| Reuters | Research Only | — | 0 | No deterministic adapter is promised for wire reporting. |