Stability Index Changelog: v3 archive

Impact notes

• Audit-triggered repairs and bounded backfills share daily prices, native-domain evidence, DEWS archives and persisted input provenance. Event exclusion or restoration recomputes PSI from the post-audit eligible universe in the same atomic transaction.
• Historical seven-day trend compares only paired admitted supply identities, independently of the current market-cap denominator. Observed zero remains valid, but an unavailable or unusable prior denominator holds the accepted day instead of inventing flat growth.
• Missing required v3 DEWS archives remain unavailable and preserve accepted days; an observed all-CALM archive still contributes measured zero stress. Archive date and row count remain part of replay provenance.
• Every historical entrypoint loads the same full twenty-one-day supply lookback, combining the seven-day trend interval and fourteen-day as-of margin, so narrow replay windows do not truncate the eligible cohort.
• Contributor attribution preserves the full-precision scoring factor without rewriting legacy factors. Displayed thirty-day extrema and averages use source-anchored UTC calendar coverage rather than row count; neither changes the scoring formula.
• Historical repairs are score-moving. Fixed-clock replay, attribution of changed days and owner review are release prerequisites; this entry does not claim a completed replay or rewrite of stored history.
• Activation is provisionally 2026-10-11 00:00 UTC (1791676800); re-date effectiveAt at release

Impact notes

• Absent, empty, wholly invalid, nonfinite or negative current totals are omitted; explicit observed zero remains admissible
• Unavailable previous-week totals are omitted from both trend sides; a nonpositive or nonfinite paired denominator holds the prior sample with trend-inputs-unavailable instead of inventing zero trend
• Stored snapshots and the API disclose supplyUnavailableIds and trendUnavailableIds alongside existing unpriced-open-event diagnostics
• A successful sample and prune returns psi-sample-published with ok plus quality for omissions; failed dependencies still hold publication
• Six-hour USD/native price-evidence and 3600-second accepted-DEWS budgets, pure formula and historical rows are unchanged

Impact notes

• Each active native event retains its latest validated CoinGecko value, upstream observation clock, and source in D1 cache
• Live sampling selects the latest native observation at or before evaluation, strictly younger than six hours; absent or stale evidence still discloses open-depeg-no-price
• USD prices and FX cannot synthesize native deviation; producer fetch count and historical replay inputs are unchanged

Impact notes

• An all-null component remains null; an observed zero remains numeric zero
• Partially observed components average only their observations and disclose their sample count
• All-day averaging and mixed-version provenance are unchanged; historical rows are not rewritten
• Live open-depeg inputs reject nominal par and legacy protocol-redeem cache rows on protocol-par routes as observed prices; absent replay-safe evidence is disclosed as open-depeg-no-price

Impact notes

• Native events use timestamped native start or recovery evidence retained in D1, strictly as-of and younger than six hours; USD prices and changing FX cannot manufacture native deviation
• No continuous native quote history is retained: unprovable windows disclose open-depeg-no-price instead of borrowing an untimed event peak or rewriting native references
• Daily scores still average every sample across version transitions, with explicit mixed-version breakdown and per-component sample counts
• Release A readers accept null daily components; the numeric snapshot writer remains unchanged until the separately gated nullable-producer activation

Impact notes

• A replayed UTC day can no longer take market cap or historical price from a supply snapshot observed after that day
• A day with no snapshot within 14 days before it leaves the coin out of that day's universe instead of borrowing a later observation
• Live 30-minute sampling and daily aggregation are unchanged

Impact notes

• The live 7-day trend term pairs current and previous-week market caps only for assets with observed previous-week supply
• Assets without observed previous-week supply no longer contribute zero to the denominator while contributing current market cap to the numerator
• Published current market-cap totals and the historical replay path are unchanged

Impact notes

• The live and historical PSI universes include active core stablecoins, active cash equivalents, and PSI-only shadow assets
• Tracked variants remain readable on their detail and browse surfaces but no longer contribute market cap, trend, depeg severity, breadth, or DEWS stress breadth to PSI
• Stable-value investment products remain tracked for research while staying outside the ecosystem monetary aggregate

Impact notes

• Retained stale DEWS rows for assets absent from the current published generation no longer fail PSI freshness
• Stress breadth is computed from a single coherent DEWS generation rather than a mix of current rows and older retained latest rows
• The no-pointer fallback remains bounded by the recent scan window and the existing staleness gate

Impact notes

• The 30-minute cron skips PSI sample publication when latest DEWS rows are absent, missing computed_at, or older than two compute-dews intervals
• DEWS stress breadth is now computed from warning-band rows only after dependency freshness is proven across the latest DEWS row set
• Public PSI remains anchored to the last valid stored sample instead of publishing a healthier fresh score from stale or absent early-warning inputs

Impact notes

• Pre-launch tracked assets no longer enter the PSI-eligible registry before launch
• Frozen tracked archives remain available on readable public detail surfaces but are excluded from live PSI computation
• PSI-only shadow assets remain included for historical continuity and admin replay repairs

Impact notes

• The 30-minute cron now returns degraded and skips the sample when `depeg_events` cannot be queried
• Only already-open depegs missing a current stablecoins price may use the replay-safe `price_cache` fallback; table-level depeg input loss is no longer treated as an empty event set
• Public PSI remains anchored to the last valid stored sample instead of silently publishing from partial core inputs

Impact notes

• Severity and breadth now fall back to the last replay-safe positive `price_cache` value for already-open depegs when the current stablecoins snapshot price is missing
• Replay fallback is capped to recent cache entries (6-hour TTL) rather than unbounded historical prices
• Prevents transient contributor/sample omissions during price-validation churn without changing the PSI formula, caps, or bands

Impact notes

• Formula changed to: 100 - severity - breadth - stressBreadth + trend
• New stressBreadth cap of 5 points
• Cron now reads latest DEWS bands (ALERT/WARNING/DANGER) to derive stress breadth