8 projects
workspace-metabolism
MCP server and CLI for policy-driven file governance in AI workspaces: audit, reversible cleanup, rollback, and verification.
holdout-governance
Fail-closed evidence manifests for financial AI research.
pit-adjuster
Point-in-time fixed-basis back-adjustment for daily prices, with corporate-action factor chains, convention-drift detection, and snapshot equivalence checks.
falsification-ledger
Hash-chained, append-only pre-registration ledger for research claims, with falsification contracts, adjudication, and Wilson-CI hit-rate reports.
ashare-data-immunity
A-share daily-bar quality checks: OHLCV cleaning, board-aware price limits and suspensions, listing and continuity audits, and SHA-256 snapshot manifests.
lookahead-free
Verifiable look-ahead checks for value-independent data pipelines: decision availability, windows, PIT reads, joins, monotonicity, and explicit heuristic boundaries.
factor-qc
Fail-closed quality gate for quantitative backtests: DSR, PBO, Harvey-Liu haircut, and MinTRL in one NumPy engine with P0/P1/P2 severity grading.
lesson-book
Local-first, deterministic mistake ledger for trading: record costs and tags, then match similar situations before you act.
