v1.3.6

1.3.6 (2026-04-30)

Bug Fixes

• security: real fix for B110 (logging in factor_proposal.py #746) (16624e0)
• security: real fix for B110 (logging in factor_runner.py #744) (88cf0fb)
• security: real fix for B110 (logging in quant_proposal.py #741) (7cf2a64)
• security: real fix for B110 (logging in quant_proposal.py #741) (ef985f8)
• security: real fix for B404/B603 (sys.executable in factor_runner.py #745) (819655a)
• security: revert broken read_pickle encoding arg in kaggle template (B301) (3574907)
• security: validate SQL identifiers in _add_column_if_not_exists (B608) (e10dfa2)
• security: whitelist-validate metric column in get_top_factors (B608) (e50519f)

v1.3.5

1.3.5 (2026-04-27)

Bug Fixes

• auto-fixer: add five new factor code fixes for groupby/apply errors (449c8fd)
• auto-fixer: add four new factor code fixes for common runtime errors (40484f6)
• auto-fixer: add groupby([level=N,'date']) SyntaxError fix (ca77c00)
• auto-fixer: disable _fix_min_periods for intraday data (77b0740)
• auto-fixer: fix chained groupby(level=N).groupby('date') pattern (7d5fe32)
• auto-fixer: fix df.loc[instrument] DateParseError on MultiIndex frames (b7860ea)
• auto-fixer: fix df['instrument'] KeyError on MultiIndex frames (aad6bd1)
• auto-fixer: fix two assignment-target bugs in instrument column fixers (421eedf)
• auto-fixer: preserve date dimension in groupby(['instrument','date']) fix (b58fdd8)
• auto-fixer: remove ddof from rolling() args, not only from std()/var() (b0fc328)
• auto-fixer: replace zero $volume with price-range proxy for FX data (3e6732f)
• auto-fixer: strip spurious .reset_index() after .transform() calls (8708aae)
• loop: prevent step_idx advance on unhandled exceptions + fix consecutive assistant messages (5ec4ad1)

v1.3.4

Bug Fixes

• Compress old experiment history in proposal prompt to reduce LLM context usage (keeps last 2 full, summarizes older to bullet points ~40-60k tokens vs ~121k)
• Fix factor result.h5 lookup path in _evaluate_factor_directly and _save_factor_values (was using Qlib template workspace, now correctly uses CoSTEER sub-workspace)
• Fix Series.to_frame().to_parquet() in factor value saving
• Correct bars-per-day documentation: 1440 bars/day (1-min data), not 96
• Update prompts to accept 2024-only debug dataset output as valid factor result

v1.3.3 — Fix strategy generator crashes on None backtest metrics

Bug Fixes

• Fix TypeError: bad operand type for abs(): 'NoneType' crash when backtest returns no IC value (ad2ad3ab)
• Fix TypeError: unsupported format string passed to NoneType in rejected strategy log output (2197f521)
• Disable slow wf_rolling and reduce mc_n_permutations to 50 for faster strategy generation runs

v1.3.2 — Security: bump python-dotenv (CVE symlink overwrite)

Security Fixes

• fix(deps): bump python-dotenv to >=1.2.2 — fixes CVE where set_key() follows symlinks and can overwrite arbitrary files via cross-device rename fallback (Dependabot alert #74)

v1.3.1 — Security: Resolve all Bandit alerts (B301, B614, B104)

Security Fixes

• fix(security): resolved 30 Bandit security alerts — B301 (unsafe pickle), B614 (torch.load without weights_only), B104 (binding 0.0.0.0)
• fix(deps): bump python-dotenv to >=1.2.2 (CVE: symlink following allows arbitrary file overwrite)

v1.3.0 — Rolling Walk-Forward Validation & Monte Carlo Permutation Test

1.3.0 (2026-04-21)

Features

• backtest: add rolling walk-forward validation and Monte Carlo trade permutation test (637a94c)

Bug Fixes

• security: resolve all 30 Bandit security alerts (B301, B614, B104) (ce5983d)

v1.2.2 — FTMO OOS Walk-Forward & Test Coverage

1.2.2 (2026-04-19)

Tests

• backtest: add comprehensive test suite for FTMO engine and walk-forward OOS validation
  — covers leverage caps, zero-signal edge cases, IS/OOS independence, bar-count integrity,
  and Monte Carlo permutation helper; slow tests excluded from default CI run

v1.2.1 — FTMO-Realistic Backtest Engine & Walk-Forward OOS Validation

Predix v1.2.1

Highlights

FTMO-Realistic Backtest Engine

• New backtest_signal_ftmo() — replaces backtest_signal everywhere
• Risk-based position sizing: 0.5% equity risk/trade, 10-pip stop, max 1:30 leverage
• FTMO daily loss limit (5%) and total loss limit (10%) enforced in simulation
• Realistic EUR/USD costs: 2.14 bps (1.5 spread + 0.5 slippage + 0.35 commission)
• Compounding: position size recalculated from current equity each trade

Walk-Forward Out-of-Sample Validation

• IS period: 2020–2023 | OOS period: 2024–2026 (independent FTMO simulation)
• Strategy acceptance requires oos_sharpe > 0 AND oos_monthly > 0
• OOS metrics stored in strategy JSON and CSV reports

Strategy Generator

• Full file logging to timestamped .log files
• OHLCV-only mode (OHLCV_ONLY=1) — strategies from raw 1-min price data
• Threshold tuner — auto-relaxes signal thresholds if trade count too low
• MIN_TRADES raised to 300 for daytrading mode

Infrastructure

• release-please-config.json — patch bumps for feat: commits

v1.2.0 — Kronos Foundation Model Integration

Predix v1.2.0

Highlights

• Kronos integration — Nixtla Kronos-mini OHLCV foundation model for zero-shot forecasting
• Batch GPU inference — 75× faster Kronos predictions via predict_batch
• Auto-generation — Kronos factor auto-generated before each fin_quant loop
• New CLI commands — rdagent kronos_predict, rdagent kronos_evaluate
• CI fix — replaced rdagent_logger with stdlib logging for CI compatibility