AlphaForge Changelog¶
A user-facing summary of notable changes in each public AlphaForge CLI release. Distribution binaries (macOS arm64) and checksums for every version are published on GitHub Releases. See Getting Started for install and update instructions.
About this page
This page is a user-facing summary. Internal refactors, tests, CI, and documentation chores that do not affect usage are omitted. Update to the latest version with alpha-forge self update.
v0.18.0 — 2026-07-19¶
Added the full FX carry (swap) backtesting suite and the fastest onboarding path via alpha-forge demo. Multi-strategy portfolios gained strategy-level dynamic allocation.
- FX carry accrual:
backtest run --carryreports daily carry approximated from short-term interest-rate differentials as reference values (carry_adjusted_*). Pairs formed from the 8 major currencies with built-in FRED rate-series mappings (USD / JPY / EUR / GBP / AUD / CAD / CHF / NZD) work with zero configuration, and importing broker swap-point CSVs withdata alt import-swaptakes precedence over the approximation. Results are persisted to the DB and shown on the alpha-visualizer v0.9.0 detail screen. - Fastest onboarding: The single
alpha-forge democommand runs everything from initialization to a backtest on the bundled DEMO data.system initnow auto-places sample data plus a demo strategy, missing-data errors are paste-ready commands, andsystem doctorsuggests concrete resolution commands (next_actions). - Dynamic allocation for multi-strategy portfolios: Added strategy-level
risk_parity/vol_targetallocation and--max-pillar-weight(per-pillar weight cap). WFT warmup pre-fill removes gaps at the start of combined curves. optimize walk-forward --save: Records WFT results so they automatically appear in alpha-visualizer's WFO tab.- Run provenance: Backtest results now record their execution source.
v0.17.0 — 2026-06-28¶
Extended the hedge triggers of the buy-hold-overlay strategy scaffold and polished the CLI UX.
- Hedge trigger extensions: Added the fast TRAILING_DD hedge trigger and an HMM × TRAILING_DD composite trigger to buy-hold-overlay. The default lookback is now configurable.
- REGIME_RULE optimization: Regime-condition firing thresholds can now be optimization targets.
combine --json: Outputs the combined portfolio equity_curve as JSON (for chart generation).- CLI UX: Eight polish items including scaffold exit help, chart-path existence validation, and did-you-mean suggestions, plus consistent error display, exit codes, and
--json/--helpbehavior.
v0.16.0 — 2026-06-24¶
The second wave of AI-agent integration: a machine-readable catalog and a --json output contract.
system describe: A new machine-readable catalog of every command and argument, letting agents self-discover the CLI surface.--jsoncontract:backtest run --jsonnow includesrun_id/result_idwith token-economized output. Extended--jsonto all read-only commands, unified the default WFT / MC verdicts, and uncaught exceptions now return a structured error envelope. Published a contract reference and versioning policy for the--jsonschemas.- Next-step guidance: Successful commands suggest the next action.
strategy templateslists the bundled templates. - Distributed skills: Bundled forge-* skills for Codex, and
system initnow deploys a generic AGENTS.md.
v0.15.0 — 2026-06-23¶
Hardened diagnostics and data fetching, and improved the first-run experience.
system doctor/system paths: New environment-diagnostics command, plus a listing of all data locations with backup instructions.- Crash resilience: Uncaught exceptions now fall back with a reporting path and an always-on crash log.
- Robust data fetching: Providers gained retries with exponential backoff, OANDA warns instead of silently truncating at its page-fetch limit, and ALTDATA sentiment sources gained retries / caching.
- Native Pine conversion: Ten staple indicators convert natively, with a
pine_supportedflag. - Maintenance: Introduced schema versions plus a migration framework for strategy / result JSON. Added
--dry-runtostrategy delete/pine delete, and a WARNING for the silent fallback whenFORGE_CONFIGis unset. - Docs / English:
FORGE_LANG=enfully localizes the top-level help. Added Troubleshooting, a Glossary, and a CLI-only end-to-end sample.
v0.14.0 — 2026-06-19¶
Substantially expanded Alpha158-derived indicators, extended metrics, and position sizing.
- Indicator expansion: Added Alpha158-derived indicators such as CMO / PERCENTRANK / STDDEV / AROON / PV_CORR / ROLLING_QUANTILE / UP_RATIO / WVMA, plus LINEAR_REG pred / resid (with native Pine conversion). Bundled an EXPR preset collection of nine KBAR formulas.
- Extended metrics: Added seven families of extended metrics.
- Position sizing: Added
kelly/vol_target. backtest timeframes: New one-shot sweep across timeframes.- explore: Added the
acknowledgecommand to mark escalations as read. - License grace period: Extended the offline grace period to 14 days; expiry now degrades to Trial instead of disabling the CLI.
- Security / fixes: Resolved 15 dependency vulnerabilities (pillow / urllib3 / protobuf and more). Fixed
risk_basedPine output size to match the engine, and invalid optimizer param_ranges are now detected before execution (exit 2).
v0.13.0 — 2026-06-08¶
Substantially strengthened unattended use from AI agents / CI: eliminated confirmation-prompt hangs, broadened --json coverage, and unified exit codes. Many bug fixes as well.
- AI agent / CI integration: Added
system configto observe the effective configuration. Extended--jsonoutput to theanalyze/live/journal/ideagroups. Unified the exit-code convention (0=success /1=not found or expected failure /2=argument error) across all commands so unattended loops run safely. - Non-interactive execution:
FORGE_NONINTERACTIVE/CI/ non-TTY detection eliminates confirmation-prompt hangs. Destructive operations now stop with exit code2unless--yesis given. - Cleanup commands: Added
optimize clean/pine list・delete・clean/data tv-mcp cache-clean. self updatenow supports Windows. The combine Pine webhook payload now includestarget_qty(alpha-strike integration).- Fixes:
backtest --regimenow shows per-regime performance as a label aggregate (--jsonstays pure JSON). Fixed monte-carlo on saved results, auto journal recording, empty-overwrite protection indata fetch, and more. - ⚠️ Breaking changes: Renamed
strategy delete's--forceto--yes/-y, and--strategy-idto--strategyforexplore log/live events/live convert-check(no compatibility aliases). Many commands that returnedexit 0on error now exit non-zero. Review your scripts / CI branching.
v0.12.0 — 2026-06-04¶
Built out the Windows distribution pipeline and improved the first-run setup experience.
system initimprovements: Added aDIRECTORYargument so creation and initialization complete in one command. Added a confirmation prompt against accidental expansion into unintended locations (e.g. directly under home).- Windows build pipeline: Automatic builds on tag push with clcache acceleration, and clang-cl to avoid MSVC heap exhaustion.
- Auth: Guides users to the Whop purchase page URL when no license is present.
- Unified the CLI name to
alpha-forge(resolved rename gaps in docs and messages).
v0.11.0 — 2026-06-02¶
Added FRED macro-data integration and macro-regime support.
- FRED macro-data integration: Added a macro-data provider (strict vintage look-ahead merge), a macro regime classifier, and
EXTERNAL_SERIESML features. - Macro event blackout: Halts entries around FOMC / NFP / CPI.
backtest run --regime-filtergates entries by macro regime. FORGE_ACCEPT_EULA: Accept the first-run EULA non-interactively (for CI).- Added
--jsontostrategy list/backtest list. Made the walk-forward OOS aggregation method configurable. - Fixes: Warn instead of silently ignoring
--leveragein buy-hold-overlay, bundlev6_signatures.yamlinto the distribution binary, and hardened the authentication flow.
v0.10.0 — 2026-05-30¶
Strengthened live (paper-trading) integration and portfolio-level live aggregation.
- alpha-forge live / replay: Treats order-reconciliation events (order_reconciled) as the authoritative source and excludes phantom (unfilled) fills, so live performance is computed accurately.
- Combine Portfolio live aggregation: Persists position-based live performance of always-in-market overlay strategies to SQLite, laying the groundwork for visualization in alpha-visualizer.
- Fix: Corrected
signal_idtimestamp formatting in Combine Portfolio Pine output.
v0.9.0 — 2026-05-29¶
Full support for a Buy & Hold benchmark with a hedge overlay, plus multi-strategy portfolios.
- Buy & Hold + hedge overlay: Added
always_in_marketand benchmark configuration, enabling strategies that stay invested while dynamically adjusting hedge allocation by regime. New metrics for alpha, hedge efficiency, and average exposure. - Combine Portfolio backtest: Combined backtests across multiple strategies (custom weights, walk-forward validation, dividend reinvestment).
- Combine Portfolio Pine v6 output: A hybrid-strategy mode that exports an integrated portfolio as a TradingView Pine script, with verification against the TradingView Strategy Tester.
- Exploration diagnostics: Over-trading gates, validation-window shortening adaptive to data length, and per-strategy-type blocker tallies.
v0.8.0 — 2026-05-21¶
Hardened the health diagnostics and quality gates of autonomous exploration.
- Exploration health: A selection guide with two-stage thresholds, insufficient-bars diagnostics, staged drawdown-floor relief, and a stall-detection trigger that halts the loop.
- Strategy scaffolding gates: Detection of mean-reversion × SuperTrend structural contradictions, asset-volatility-aware stop-loss/take-profit defaults, indicator-overload warnings (warn at 4+, abort at 5+), and an indicator-concentration guard.
v0.7.0 — 2026-05-21¶
Major progress on cost realism and TradingView consistency.
- Per-broker cost presets: Spread, maker/taker fees, and IBKR-style per-share fees reflected in backtests.
- TradingView cross-validation: Reconciles generated Pine against the TradingView Strategy Tester to verify result consistency, with tolerance profiles for SL/TP strategies.
- Pine quality: A post-generation validator based on the Pine v6 signature database, plus an option to bake the date filter into Pine output.
- Exploration metrics: Leverage-adjusted CAGR, Calmar recommendations, and walk-forward dispersion metrics.
v0.6.0 — 2026-05-18¶
Expanded the first-run experience and supported markets.
- Setup:
alpha-forge system initnow scaffolds goal-configuration samples and asset-class templates. Added--strategy-idtostrategy create. - Crypto support: A dedicated Binance data provider, fetching 4h/1h over 5+ years.
- alpha-strike integration:
pine generate --with-webhookautomatically adds order-routing inputs/alerts to Pine. - Benchmark:
--no-benchmarkdisables benchmark comparison.
v0.5.1 – v0.5.3 — 2026-05-15 to 16 (hotfixes)¶
- v0.5.3: Added a Trial-plan fallback so users without Whop authentication can still run the CLI. Fixed data-fetch guidance shown on errors.
- v0.5.2 / v0.5.1: Fixed resource-path resolution in the distributed (Nuitka) binary, and a Trial-UX improvement that lets
system init/docsrun without authentication.
v0.5.0 — 2026-05-15¶
Unified the product-family naming.
- Breaking change: Renamed the CLI command from
forgetoalpha-forgefor product-family symmetry. Distribution binaries are likewise namedalpha-forge-*. - Improvements: Added a mechanism that turns known exceptions into action-oriented messages. UX improvements to optimize/backtest output.
v0.4.0 — 2026-05-14¶
Dramatically improved startup speed.
- Performance: Lazy-loading of subcommands cut warm startup from ~2.6s to ~0.08s. Distribution-binary cold start dropped from ~97s to ~7s.
- self-update: Added a post-replace smoke test and a spinner display.
- Breaking change: Trimmed the bundled strategy templates from 34 to a curated 7.
v0.3.5 — 2026-05-13¶
- Fix: Resolved a failure where
alpha-forge system initcould not expand resources in the distributed binary.
For distribution assets and checksums of older versions, see GitHub Releases.