Methodology changelog
Every material change to data capture, ranking rules, truth-state logic, and coverage scope. Dated, ordered newest first, immutable.
Full technical methodology is documented at /methodology. This page records when and why that methodology has changed. For the underlying snapshot artifacts cited by each entry, browse the public archive.
- Surface
Snapshot permalinks surfaced to search
Individual snapshot artifact pages at /archive/snapshots/[id] are now declared in the sitemap and carry Dataset JSON-LD with CC-BY-4.0 licensing. The archive has always been public; this change makes it discoverable.
- Surface
Research and press landing page published
New /press page consolidates citation format, story angles, methodology links, and press contact into a single journalist-ready URL.
- Data model
Fee-model correction boundary
Cost model updated to correct under-estimation in several card-rail scenarios. Snapshots captured before this date may use a different estimation method and are not directly comparable with post-correction snapshots. Archive pages that span the boundary surface an explicit interpretation note.
- Trust
Truth-state taxonomy formalized
Per-row states collapsed to three canonical values: rankable, unrankable, no_snapshot. Rows lacking required fields or verification coverage are explicitly excluded from the ranked view instead of being silently down-weighted.
- Capture
Coverage pointer scoping fix
Only canonical ingest runs (no explicit exchange allowlist) own the latestSnapshotId pointer in SnapshotCoverage. Non-canonical runs write snapshot rows but no longer advance the pointer, preventing partial runs from replacing canonical coverage.
- Coverage
Phase 1 batch coverage
16 batch adapters added as deterministic static adapters using computeDeterministicCost. All produce rankable evidence for US, UK, DE, SE, CA, AU, SG across BTC and ETH on card and bank rails.
Estimates, not quotes.