Evidence / Data kit / Variable-latency bridges

Variable-latency bridgesPhase Preliminary · since 2026-05-04

Reproducibility scope for the variable-latency bridge state classification (BS1 / BS2) served by Invarians on Chainlink CCIP lanes and Circle CCTP routes. The full reproduction kit publishes when the calibration reaches confidence HIGH (~2026-06-03).

Sources currently calibrated

Surface
10 Circle CCTP routes (BS1 / BS2 active). 10 Chainlink CCIP lanes (raw observability, classification deferred).
Capture
Rust collectors on dedicated VPS services (invarians-cctp, invarians-ccip), persisted to Supabase tables ans_cctp_route_signals and ans_ccip_lane_signals.
CCTP signal per sample
circle_api_latency_ms, Circle Iris attestation API health probe (proxy for end-to-end message latency until sustained throughput allows direct calibration on attestation_latency_p90_s).
CCIP signal per sample
last_sequence_advance_seconds, commit_latency_p90_s, rmn_cursed (binary safety override).
Sampling cadence
600 s polling per route / lane.
CCTP calibration window
P97/14d preliminary committed 2026-05-04 (confidence LOW). Lifecycle progresses to P97/25d intermediate (~2026-05-29, MEDIUM) then P97/30d production (~2026-06-03, HIGH).
CCIP calibration
Statistical classification deferred. Distribution dégénérée actuellement (8/10 lanes saturate at cap on last_sequence_advance_seconds). Phase 2B observation Q3, Phase 2C calibration Q4 2026 per CCIP-V2/calibration_ccip.md.

Reproduction kit availability

The full kit (SQL transaction, guard rails, expected thresholds, CSV samples) publishes when CCTP reaches confidence HIGH on 2026-06-03. The current preliminary calibration is documented openly in the calibration log so reproducers can verify the methodology and the thresholds committed.

MD
calibration_log.md
Public calibration log. Entry #036 (CCTP P97/14d preliminary, 2026-05-04). Entry #037 (CCIP deferred, distribution analysis).
MD
methodology.md §13
Variable-latency bridge calibration methodology: scope, method, guard rails, lifecycle (preliminary 14d → intermediate 25d → production 30d), limitations.
MD
limitations_and_plans.md
Honest limitations of the current preliminary calibration and the road to production grade.

Calibration timeline

  1. 2026-05-04 ✅ Preliminary committed — CCTP P97/14d on circle_api_latency_ms, confidence LOW. 10 routes, BS1 / BS2 served by the panel.
  2. ~2026-05-29 — Intermediate recalibration P97/25d, confidence MEDIUM.
  3. ~2026-06-03 — Production recalibration P97/30d, confidence HIGH. Full reproduction kit publishes here.
  4. ~2026-06-04+ — CCTP routes integrated to the live stress-events detector. CCIP RMN cursed override branched (binary protocol-level signal, no statistical calibration required).
  5. Q3 2026 — CCIP per-lane audit, identify lanes eligible for statistical calibration based on observed throughput.
  6. Q4 2026 — CCIP Phase 2C calibration on eligible lanes, BS classification activated lane-by-lane.
Public reproducibility for the variable-latency scope follows the same standard as the L1 backtests already published: every threshold in bridge_thresholds with bridge_type='cctp' is generated by a deterministic SQL transaction with falsifiable guard rails, identical to the native scope methodology. The current preliminary status reflects that 14 days of samples is below the 25-day support threshold required for production claims; it does not affect the determinism of the calibration itself.