{
 "ok": true,
 "product": "Strategy Arena Lab",
 "channel": "beta",
 "current_version": "0.1.0-243",
 "minimum_version": "0.1.0-85",
 "recommended_version": "0.1.0-243",
 "package_version": "0.1.0-243",
 "release_build": "20261011-243-n1",
 "build_id": "20261011-243-n1",
 "update_required": false,
 "download_url": "https://strategyarena.io/download",
 "changelog_url": "https://strategyarena.io/desktop-demo",
 "docs_url": "https://strategyarena.io/tradingview-pine-script-to-cuda",
 "release_notes": [
  "Version 243: a fairer judge (the SPA test now compares with buy-and-hold over the test period itself, trials are counted as effectively independent, a rule fixed in advance can be judged), any return series can be judged (prove_return_series: your own bot, a portfolio, an external strategy), and many more rules can be written: high/low and offsets, arithmetic between indicators, daily filters on a 4h script, day/hour filters, risk-based sizing (risk 1% per trade), break-even, ATR trailing with activation, giveback exits. Research on past and simulated data, not investment advice.",
  "Version 242: a new rule family can again earn survives (test-period openings are counted per rule family, not for the whole Lab), named parameters work everywhere (stop and target multiples, ATR length, timeout, size), so the GPU sweep searches stops too. Short margin calls on gap bars now match the real TradingView (45/45 trades). Research proof v8 and Monte Carlo on the account equity. Portfolio on a test period never opened, weekly challenge rehearsal on past weeks. Windows executable 63 MB instead of 96 MB. Research on past and simulated data, not investment advice.",
  "Version 241: scientific reliability. Statistics now come from the engine's own equity curve (position size, fees, open positions, partial exits), stricter verdicts (a missing control or an already opened test period can never give survives), proofs written atomically and checked on read. The judge uses TradingView rules by default, so the exported Pine is the judged strategy; short margin calls confirmed in the real TradingView (361/361, 360/360, 364/364 trades). Pine export adds Bollinger and MACD (263/263 trades identical in TradingView). AI loop: what an AI missed reaches us, solved only when its scenario passes, proven recipes. Simpler screens. Research on past and simulated data, not investment advice.",
  "Version 240: weekly challenge (submit a frozen strategy before Monday 00:00 UTC, judged on the following week that nobody knew: strategyarena.io/defi-de-la-semaine), \"My idea in plain words\" without AI (type a rule in French or English, the Lab turns it into a script and judges it), \"Does it hold elsewhere?\" (same frozen settings replayed on the other markets), add a market in one click (public Binance and Crypto.com candles, no key, no account: ETH, SOL, BNB, XRP, ADA), my portfolio (simulation: mix 2 to 6 strategies, weights frozen on training, every weight tried counted), one window at a time on launch, live counter, mirage share card, 3 first steps. Research on past and simulated data, not investment advice. Version 239: the judge in 60 seconds (first launch: a famous strategy tested, thousands of variants searched, then winner or mirage on the sealed period), in-app update with SHA-256 check, TradingView judge now confirmed in the real TradingView on every convention including partial short margin calls. Version 238: public verified registry (strategyarena.io/verified): publish a badged strategy after a full server-side re-check (Binance candles, our engine, trade by trade vs the Pine Logs), with a Post on X button and share card. Night mode for Operator: research runs overnight on several markets, frozen selection rule, pooled proof on unseen data, morning report (free preview: 1 market x 20,000 variants). Lighter MCP answers with loadable tool packs. Version 237, the voice of the AIs: compact MCP answers (lab state 0.3 KB instead of 104 KB, tool list 59 KB instead of 103 KB), named parameters such as SMA(fast), parenthesised conditions (A or B) and C, consistent defaults (fees 10 bps everywhere), one clear line per dataset. Rigorous judge: no statistical claim under 10 trades, TradingView receipt says what was confirmed in the real TradingView and what is still unverified, one bootstrap library for proofs and the duel, Pine import fixed (strategy.close no longer read as a short). Multi-asset proof: one script judged on 2 datasets for free (3 to 8 with Operator) so slow strategies get enough trades; CSV import; pair backtests (pair: A/B).",
  "Version 239: the judge in 60 seconds (first launch: a famous strategy tested, thousands of variants searched, then winner or mirage on the sealed period), in-app update with SHA-256 check, TradingView judge now confirmed in the real TradingView on every convention including partial short margin calls. Version 238: public verified registry (strategyarena.io/verified): publish a badged strategy after a full server-side re-check (Binance candles, our engine, trade by trade vs the Pine Logs), with a Post on X button and share card. Night mode for Operator: research runs overnight on several markets, frozen selection rule, pooled proof on unseen data, morning report (free preview: 1 market x 20,000 variants). Lighter MCP answers with loadable tool packs. Version 237, the voice of the AIs: compact MCP answers (lab state 0.3 KB instead of 104 KB, tool list 59 KB instead of 103 KB), named parameters such as SMA(fast), parenthesised conditions (A or B) and C, consistent defaults (fees 10 bps everywhere), one clear line per dataset. Rigorous judge: no statistical claim under 10 trades, TradingView receipt says what was confirmed in the real TradingView and what is still unverified, one bootstrap library for proofs and the duel, Pine import fixed (strategy.close no longer read as a short). Multi-asset proof: one script judged on 2 datasets for free (3 to 8 with Operator) so slow strategies get enough trades; CSV import; pair backtests (pair: A/B).",
  "Version 238: public verified registry (strategyarena.io/verified): publish a badged strategy after a full server-side re-check (Binance candles, our engine, trade by trade vs the Pine Logs), with a Post on X button and share card. Night mode for Operator: research runs overnight on several markets, frozen selection rule, pooled proof on unseen data, morning report (free preview: 1 market x 20,000 variants). Lighter MCP answers with loadable tool packs. Version 237, the voice of the AIs: compact MCP answers (lab state 0.3 KB instead of 104 KB, tool list 59 KB instead of 103 KB), named parameters such as SMA(fast), parenthesised conditions (A or B) and C, consistent defaults (fees 10 bps everywhere), one clear line per dataset. Rigorous judge: no statistical claim under 10 trades, TradingView receipt says what was confirmed in the real TradingView and what is still unverified, one bootstrap library for proofs and the duel, Pine import fixed (strategy.close no longer read as a short). Multi-asset proof: one script judged on 2 datasets for free (3 to 8 with Operator) so slow strategies get enough trades; CSV import; pair backtests (pair: A/B).",
  "Version 237, the voice of the AIs: compact MCP answers (lab state 0.3 KB instead of 104 KB, tool list 59 KB instead of 103 KB), named parameters such as SMA(fast), parenthesised conditions (A or B) and C, consistent defaults (fees 10 bps everywhere), one clear line per dataset. Rigorous judge: no statistical claim under 10 trades, TradingView receipt says what was confirmed in the real TradingView and what is still unverified, one bootstrap library for proofs and the duel, Pine import fixed (strategy.close no longer read as a short). Multi-asset proof: one script judged on 2 datasets for free (3 to 8 with Operator) so slow strategies get enough trades; CSV import; pair backtests (pair: A/B).",
  "Version 236 (build n6): Linux 236 (CPU) published; Back to Jarvis button and Esc in the Colisee; free-account call to action for the launch offer. Colisee rebuilt for humans, cards ordered by rank and scaled to large screens: one card per AI with rank, plain-language action, result vs Buy & Hold and the judge badge; Jarvis opens full screen; fix for the sealed-period challenge (dataset list paging). Version 236 (build n2): the GPU now runs TradingView-exact semantics and shorts (200,000 variants in about 3.5 s, every kept row and a random control sample replayed on the CPU reference, 0 difference measured). \"Verified by Strategy Arena\" badge: when the judge finds every TradingView trade, the Lab issues a signed certificate and a share card with its own verdict (winner or mirage). Colisee: several AIs connected at once each get their own lane and score; Jarvis stays live when another window covers the Lab, with an animated replay and a built-in video recorder. AI strategy bank, ensembles and a sealed one-year duel: fixed rules vs an ML meta-model vs Buy & Hold, forward test from 9 October 2026.",
  "Version 236: the TradingView judge. Export the exact Pine (TradingView-exact semantics: TradingView indicators, intrabar stops and targets, commission in cash, open last position), run it in TradingView, paste the Pine Logs or the List of Trades, and the Lab compares trade by trade. Checked in the real TradingView on 8 October 2026: the 6 strategies of the AI leaderboard give the same 234 trades out of 234, quantities included.",
  "Version 236: your AI at the controls. An essential MCP profile (start_here, compact answers, optional SHA, teaching errors) lets small local models complete the whole path. \"Connect my AI\" sets up Claude Desktop, Cursor, LM Studio or opencode with a stable port and token.",
  "Version 236: \"Winner or mirage?\" judges a script on data unseen during the search, against Buy & Hold on the same windows, with PBO, CPCV, deflated Sharpe and SPA in plain words. Background GPU jobs run large sweeps; every kept row is replayed on the CPU reference.",
  "Version 236: three doors on the home screen and a first backtest in one click; human buttons for proof, GPU jobs, night mode and the TradingView comparator; full French interface.",
  "The default execution (legacy) gives the same results as before, bit for bit; TradingView-exact semantics is opt-in. The judge's rule is unchanged. Windows EXE/MSI remain unsigned. Past results, not a forecast.",
  "Version 195: Pine export checks exits, parameter ranges and titles; VWAP exit inputs are preserved. SMA+RSI and dual DMI retain all entry clauses. OBV uses the documented Pine v5 variable. Native generated comparisons stop before compute on stale or unsupported Pine. 58 catalog families passed local checks; 12 are explicitly blocked. TradingView compilation and transaction parity remain unverified.",
  "Version 194: next to each verdict the end screen gives the average result per trade with its 95% range, and the result when take-profit and stop are filled at their level during the session (as TradingView does), instead of at the close. The judge still decides at the close; these two lines are information only.",
  "Version 193: the level of proof now replays the whole rule. The sealed entries are moved to other days and the engine replays take-profit, stop, timeout and protection on the real price path, as advised by an independent review. Calibrated on 352 runs with random entries: false alarms 0.4 to 0.6% at the 1% threshold. No verdict changes meaning.",
  "Version 192: next to each verdict the end screen now gives the level of proof. The same trading calendar is placed at every other position of the sealed period and the app counts how often it did as well or better. When a run stops for lack of trades, the app says how many trades were expected and what the chance of reaching ten was.",
  "Buttons that lead to the website (plans, audit, account connection) now open the system browser; before this version they did nothing inside the app. The Account page shows the plans first.",
  "Daily histories now go back as far as the source allows (20 to 38 years for gold, silver, oil, Nasdaq 100, S&P 500, DAX and CAC 40) and Bitcoin 4h covers 6,000 bars, about two years and nine months.",
  "Measured on 2026-09-29 with this version: 352 runs on 8 markets in 4h and daily bars, 0 error, 8 strategies passed the sealed filter, each one reproduced when run alone. None of the 8 reaches a strong level of proof: with 10 to 41 trades the filter cannot tell skill from luck. Past results, not a forecast.",
  "Version 190: every recipe of the assembler can be judged; a run that stops says what to change in plain words.",
  "Windows189 adds long real tapes for Nasdaq 100, S&P 500, DAX and CAC 40 (4h over about three years, daily over ten years) and a market scan: one strategy judged on each market, 2 markets on Free, all 8 on Builder.",
  "Windows188 makes real market history the default tape (Bitcoin 4h over two years) and explains in one sentence what the judge measured. The previous default was a short synthetic practice tape that rarely reached a verdict.",
  "Windows187 adds the French interface, the new console skin, one-click account connection and a plan offer next to the verdict.",
  "Windows185 starts Stripe Checkout in-app after a verified plan review and two explicit consents; payment stays confirmed only by the server webhook.",
  "The judge's rule is unchanged: ten closed trades on the sealed part are required for a verdict. No new performance claim is made.",
  "Windows and Linux packages are rebuilt for 195. Windows EXE/MSI remain unsigned. Linux uses the CPU engine; GPU execution on Linux remains unverified. No TradingView compilation or new performance claim."
 ],
 "published_at": "2026-10-08",
 "signature_status": "unsigned",
 "signature_required_for_broad_distribution": false,
 "unsigned_distribution_accepted": true,
 "distribution_policy": "unsigned_public_beta_accepted",
 "signature_note": "This public beta is unsigned by product decision. Verify SHA256 before running; Windows SmartScreen may warn until a future signed build earns reputation.",
 "signing_gate": {
  "schema": "strategyarena.desktop_signing_gate.v1",
  "status": "unsigned",
  "broad_distribution_blocked": false,
  "unsigned_distribution_accepted": true,
  "smartscreen_risk": "authenticode_deferred_unsigned_public_beta_accepted",
  "required_before_directory_submission": [
   "Unsigned beta warning remains visible on /download",
   "EXE and MSI SHA256 hashes are published and match local artifacts",
   "directory verification gate passes with unsigned_distribution_accepted=true",
   "directory verification gate passes"
  ],
  "readiness_commands": [
   "npm run signature:preflight",
   "npm run signature:ready",
   "npm run signature:plan",
   "powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts/verify-windows-signature.ps1 -RequireSigned",
   "npm run release:verify",
   "npm run directory:verify"
  ],
  "supported_certificate_inputs": [
   "SA_DESKTOP_SIGN_CERT_PATH",
   "SA_DESKTOP_SIGN_CERT_SUBJECT"
  ]
 },
 "installers": {
  "exe": {
   "url": "https://strategyarena.io/static/downloads/strategy-arena-desktop-0.1.0-243-x64-setup.exe?v=933ce7135aca0498",
   "sha256": "933ce7135aca0498e05b1e3cfea2647b9dd63247210b678e8f0bc58b8444d28c",
   "size_bytes": 15342936,
   "platform": "windows",
   "package_version": "0.1.0-243"
  },
  "msi": {
   "url": "https://strategyarena.io/static/downloads/strategy-arena-desktop-0.1.0-243-x64.msi?v=6b4286fb3983085a",
   "sha256": "6b4286fb3983085afd3a067f15e1f177252e25832da0395e02779c6d8af33069",
   "size_bytes": 20529152,
   "platform": "windows",
   "package_version": "0.1.0-243"
  },
  "appimage": {
   "url": "https://strategyarena.io/static/downloads/strategy-arena-desktop-0.1.0-243-amd64.AppImage?v=2e7e9fe8d9919d9a",
   "sha256": "2e7e9fe8d9919d9a30da9787e7f52986194da57062a41afb2251a921ec17274b",
   "size_bytes": 94820856,
   "platform": "linux",
   "note": "Linux stays on version 195 (CPU engine) for now: the 236 Linux packages depend on CUDA. Same interface and recipes as the Windows build. CPU verdict engine; GPU search is not verified on Linux. Start-up checked on Ubuntu; the full interface was not tested on Linux.",
   "package_version": "0.1.0-243"
  },
  "deb": {
   "url": "https://strategyarena.io/static/downloads/strategy-arena-desktop-0.1.0-243-amd64.deb?v=f0cf63b93900a5a5",
   "sha256": "f0cf63b93900a5a580e99b9b6af9d04a6ed0dc038f8dace19a864937c17cd8f3",
   "size_bytes": 21728088,
   "platform": "linux",
   "note": "Linux stays on version 195 (CPU engine) for now: the 236 Linux packages depend on CUDA. Same interface and recipes as the Windows build. CPU verdict engine; GPU search is not verified on Linux. Start-up checked on Ubuntu; the full interface was not tested on Linux.",
   "package_version": "0.1.0-243"
  }
 },
 "server": {
  "status": "static",
  "message": "Desktop release manifest static fallback online."
 },
 "platform_release_scope": "Windows x64 and Linux amd64 on 243. Linux build is CPU-only (no CUDA); Windows uses the GPU when available."
}