Sharpen v1.1.2

Added

• FinRL-X audit study (studies/finrl-x/): the Adaptive Rotation audit with the per-list
  results of all 37 growth lists and the pre-registered verdict, plus the keel-v1 final report.

Changed

• scripts/research/finrl_x_hindsight_runs.py works on Linux and macOS and no longer needs
  results/finrl_x/hindsight/pool.json: without it the growth lists are regenerated from the
  published pool and seed and checked against the published fingerprint.

Sharpen v1.1.1

Fixed

• Crucible v18.0 withdraws v17.0's cohort book. The cohort analytic floor and Monte Carlo null
  use the raw combiner again. The v17.0 book, which fell back to the base book on bars where a
  member could not be sized, made the null test reject every no-edge panel built from a real pool
  (60 of 60 at alpha 0.05, against 1 of 60 for the raw combiner). No recorded verdict was affected.

Full changelog: v1.1.0...v1.1.1

Sharpen v1.1.0

Added

• SharpOps promotion ladder (docs/sharpops_promotion_standard.md, sharpen/sharpops/): one
  pre-registered primary test per rung (paper, challenge, live), artifact tripwires, a hashed
  registry of every gates file, and scripts/sharpops_promotion_check.py.
• Crucible community ledger: export a closed search with its power, so a null result says
  whether the search was finished or too weak to tell (scripts/crucible_export_closed_search.py).
• Scorecard capturability reports the traded book at every rebalance phase and with a one-bar
  execution lag; PROMISING also requires the median phase to clear the frictionless floor.
• Opt-in point-in-time corrections for the Taiwan small-cap universe and panel (off by default).
• sharpen.__version__, this changelog, and a test that keeps the version consistent.

Changed

• Crucible v15.0 → v17.0. v15.0: every pre-registered hypothesis is tested and only decisions
  are charged; the pre-registered direction is the tested direction; release-calendar look-ahead
  fixes. v16.0: sequential LORD++ levels, the power guard at the live level, the lockbox decides by
  a sequential test on the forward Sharpe difference. v17.0: cohort books admit a member only where
  it can be sized; one trial ledger per substrate; engine and data-store fixes.
• Protocol v2 is renamed SharpOps (docs/sharpops.md).
• scripts/run_full_pipeline.py requires --stage, and the stage decides which phases run.
• Bar-level profit-factor floors are diagnostics, no longer promotion gates.

Fixed

• run_walk_forward.py read metric keys the backtest never returned and did not seed the envs.
• The deploy tooling's paramiko dependency is declared (the deploy extra).

Full changelog: v1.0.0...v1.1.0

Sharpen v1.0.0

First public release of Sharpen: an agentic-AI, full-featured quant strategy builder — from a concept to a validated, trained and deployment-ready system.

Highlights

• Agent-driven workflow. Built to be driven by an AI coding agent such as Claude Code or Codex, in any language. The rules it follows ship in CLAUDE.md: pre-registration, gates, staged training, and the invariants that must never break.
• Seven-tier validation funnel (sharpen/signals/). Covers causality tripwires, IC and decay, capturability after costs, stability, purged path resampling, deflated Sharpe with FDR and the Harvey-Liu-Zhu hurdle, and orthogonality to a factor book.
• Backtesting and falsification toolkit. Walk-forward and recent out-of-sample stages, realistic cost models, circular-shift timing nulls, edge-sign-flip and matched-exposure tests, planted-signal power, matched-null search, gate-reachability checks, multiplicity accounting, deflated Sharpe with PBO, and stress and noise robustness.
• Crucible alpha mining (sharpen/crucible/, crucible-v14.0). Continuous automated alpha mining over free-data connectors (FRED, CFTC COT, SEC EDGAR, GDELT, Stooq, TWSE, TAIFEX). An LLM hypothesis proposer never sees any verdict. Gates are hash-frozen and every candidate is forward-incubated in a lockbox.
• Frontier deep RL. SAC with CrossQ-style critics, distributional SAC, IQN, BDQ and PPO. Environments cover multi-asset allocation, execution scheduling and market making. Includes distributed Optuna HPO across GPU fleets and the six-stage SharpOps pipeline with a config validator that rejects leaky or fused runs.
• Execution. Six broker adapters, a paper-portfolio executor with a parity harness, and a Docker stack with Prometheus, Grafana, Telegram alerts, drift detection and a kill file.

Get started

pip install -e ".[dev]"
python scripts/research/eval_signals.py --batch demo --panel "synthetic:1400,60" --out results/demo

No data and no API keys needed. The headline research numbers reproduce from a fresh clone in about a minute. See docs/REPRODUCE.md.

Also in this release

• NEGATIVE_RESULTS.md: 76 strategies and probes across eight families, each with the number that decided it.
• docs/LEAKS_FOUND.md: three bugs that had manufactured results, how each was caught, and the test that now guards it.
• Test suite: 3261 passed, 25 skipped, 28 deselected, 1 xfailed on the release tree.

Apache-2.0. Not investment advice; see DISCLAIMER.md.