v0.5.0 — Evidence gates, 20/80 reserves and optimized sizing

This preview blocks unqualified models from proposing buys and adds protected 20/80 allocation for recorded USDC and SOL. Every real swap still requires your Phantom signature.

• Research selection, first watch start, quote preparation and outstanding model approvals require evidence tied to the selected model and parameters. Old, rejected and insufficient evidence cannot authorize model buys. Existing position exits remain available.
• Reserve 20% of recorded assets and each new trade receipt. Reserved balances are not reused on the next trade. Record existing SOL using its original received quantity and USDC cost from the original capital; no duplicate funding or fabricated trade fills.
• Research now searches bounded buy and partial profit-exit fractions (25/50/75/100%) alongside entry/target parameters. Holdout requires at least eight buy entries, positive net return and drawdown below 10%. Partial exits cannot inflate the entry count.
• Profit exits check proportional costs in USDC and reference THB. Real quotes and network costs may veto a proposed trade. Backtests remain research evidence, not proof of profitability.

After updating, open Live trading → สำรอง 20% · ส่วนเทรด 80%. Record the original SOL receipt and its USDC cost, or leave both blank to configure recorded capital while unknown-basis SOL remains fee funds. Select a passing new research candidate before model buys; a fresh holdout may require waiting for new candles. Restarting keeps trading paused.

Windows x64 MSI includes a signed updater artifact, not Windows Authenticode. No funded trade or Phantom signature was submitted during verification.

Verification: 63 Rust tests, Dioxus WASM check, 3 wallet mock tests, 2 public read-only provider tests, MSI payload comparison, updater signature/tamper/version checks and packaged UI inspection passed. Full MSI system installation and actual in-app replacement remain unverified. See VERIFICATION.txt for scope.

Source: 97b0841163cb2ec043c41e10c5b2f0e5ce153016.