Status
Current progress, what’s proven, and what must be decided before we can compute and validate actionable betting “edges” for NBA player props.
Latest status (Decision Packet Ready)
- Performance truth (schedule/box/PBP): stable at scale; idempotent reruns; multi-date backfill + resume proven.
- Market truth (props lines + prices + snapshots): queued pending provider selection.
- Decision packet: ready; can be answered in under 2 minutes (see Decision point).
- Market adapter scaffold: schema + adapter contract drafted; code scaffold staged so the provider adapter can be implemented immediately.
Readiness gates
- Idempotent ingest — passed.
- Multi-date backfill + resume — passed.
- Coverage reporting — passed.
- Decision packet readiness — passed (we can proceed as soon as Jeffrey answers).
- Market-layer snapshots — queued (provider selection).
What we need next (decision point)
Jeffrey’s choices determine scope, costs, and what “edge” strategies are realistic. Once we have them, we implement the provider adapter and begin market snapshot capture.
Past releases
Release 2025-12-18 (bundle v3) — expand
Status
Current progress, what’s proven, and what must be decided before we begin full backtests and edge evaluation.
Latest status (as of 2025-12-18 CT)
- Season-to-date truth capture: 387 distinct games across 55 dates captured with schedule + box + PBP present for every game in coverage checks.
- Operational behavior confirmed: long pulls can appear “quiet” (low CPU) because they are waiting on network responses and intentional
--sleeppacing between calls. Safe interrupt (Ctrl+C) +--resumeis supported. - Next focus: integrate a market provider and persist line/price snapshots with capture timestamps.
Readiness gates
- Idempotent ingest — passed (primary keys prevent duplicates).
- Multi-date backfill + resume — passed (week/month ranges completed; resume run produced 0 new games as expected).
- Coverage reporting — passed (coverage_report outputs show missingness lists empty).
- Market-layer snapshots — queued (requires provider choice).
What we need next (decision point)
We need Jeffrey’s input on budget/constraints and the target betting markets so we can pick an odds/props provider and lock a schema.
What we will do immediately after provider selection
- Create canonical tables for market snapshots (book, market, player/team, line, price, timestamp).
- Backfill historical markets for the same date range we already captured in the truth layer (so joins and backtests are possible).
- Start an EV/CLV loop: project → compare to line → record outcome → measure closing line value.
Past releases
Release 2025-12-16 (bundle v2) — expand
Status
Where we are, what’s working, and what must be true before we scale external data capture and begin real backtests.
Current phase
Phase 3: Truth-layer POC stabilization + report surface.
We have a working end-to-end sample pull (schedule + box + optional PBP) writing to SQLite. Current hardening focus is making re-runs and larger pulls reliable.
Readiness gates (what “done” looks like)
- Idempotent ingest: safe re-run without collisions.
- Multi-date backfill: run a week or month of dates without manual babysitting.
- Odds snapshot ingestion: capture sportsbook lines/odds with timestamps and market identifiers.
- Backtest harness: replay historical games + historical lines and score EV/CLV.
Next up
- Finalize PBP idempotency patch confirmation on a clean DB
- Decide odds/props provider and schema for market snapshots
- Implement external ingestion tool (resumable backfills)