Skip to content

CHECKLIST.md

Latest commit

 

History

History
36 lines (27 loc) · 2.27 KB

CHECKLIST.md

File metadata and controls

36 lines (27 loc) · 2.27 KB

Checklist

U-001

Current Ingestion State

  • Discovery current through 2026-04-08.
  • FL-001 current through 2026-04-08.
  • FL-002 holders ingestion working again in capped mature-only slices.
  • FL-002 coverage is still low relative to the mature universe and needs repeated catch-up passes.
  • RD-001 recent-coin batch path works with conservative free-tier defaults.
  • RD-001 still has too many partial and error statuses for research-wide use.

Operational Loop

  • Keep running make u001-holders in capped slices until FL-002 coverage is materially higher.
  • Re-check the 30 historical FL-002 error rows after more holders passes and separate stale residue from active auth issues.
  • Keep running ./scripts/run_batch.sh --max-coins N for guarded RD-001 catch-up on recent coins.
  • Decide when to start capped Helius backfill for older RD-001 windows instead of only recent Shyft coverage.
  • Start capped Helius backfill for old RD-001 partial rows with a dedicated wrapper.
  • Add an explicit opt-in lane for free-tier-guarded historical RD-001 partial rows instead of mixing them back into the normal retry queue.

RD-001 Hardening

  • Conservative free-tier defaults are in place for batch workers, parse workers, RPC batch size, and signature caps.
  • Oversized recent coins are skipped before they can blow up a free-tier batch.
  • Reduce Shyft transport instability and watch whether Server disconnected errors fall after the HTTP client change.
  • Add a targeted RD-001 recovery path for historical error rows instead of relying only on the generic batch queue.
  • Add a targeted RD-001 recovery path for historical partial rows instead of relying only on the generic batch queue.
  • Review noisy conformance warnings such as Multiple trade events found and treat duplicate or foreign-pool extras as harmless redundancy while preserving warnings for same-pool ambiguity.

Research Readiness Gate

  • Raise FL-002 coverage enough that holder-derived features are not based on a tiny subset.
  • Raise RD-001 coverage enough that trade-flow features are not biased toward a narrow recent sample.
  • Only start major U-001 feature/signal expansion after FL-002 and RD-001 coverage are good enough to trust.