v3.9.1.21

NoahAI v3.9.1.21 Web UI PAPER PnL, Cost & Venue Attribution Patch

v3.9.1.21 is a source candidate. The public Windows release remains v3.9.1.20 until a new immutable installer, blockmap, latest.yml, upgrade/restart cycle, and six-venue PAPER checks pass.

PAPER display, cost, logs, and recovery controls

• Stopping PAPER validation is authoritative; delayed closes and startup synchronization cannot silently reactivate it.
• Open Binance and unified PAPER positions are persisted atomically and restored with their exact strategy key/version and exit contract. Position IDs make replayed closes idempotent.
• New Binance PAPER closes record gross PnL, estimated round-trip fees, estimated slippage, and net PnL separately. A v3.9.1.19 zero-cost row is reconstructed only when entry price, quantity, gross PnL, and net PnL prove the legacy signature; the UI labels it as a default-contract estimate and never rewrites the append-only source row.
• Upbit, Bithumb, Bybit, Bitget, and OKX active PAPER positions update current price and estimated-cost-adjusted unrealized PnL before close, matching the Binance display contract.
• Shared analysis and decision recording use the owning venue as the log source. Legacy rows whose trade_runtime::<venue> owner contradicts the selected tab are hidden rather than presented under Binance.
• Profitability recovery applies independently to all six crypto venues using each venue's latest completed trades. Net PnL costs are not deducted twice, and the message states that the profile is re-evaluated after the next completed trade.
• Multiple PAPER-observation strategies share execution slots through deterministic round-robin; this is not a separate simulated account per strategy.
• Upbit public ticker requests use shared caching, minimum intervals, and bounded 429 backoff. Learning events use an append-only journal with bounded checkpoints. Binance TP/SL audits include both regular and Algo orders.

v3.9.1.21 release boundary

• Source tests and the Web build do not prove the Windows artifact or live credentials.
• Do not rename or overwrite v3.9.1.20 public assets. Promote v3.9.1.21 only after the release ledger in docs/V39121_PAPER_PNL_COST_VENUE_ATTRIBUTION_TEST_PLAN.md is complete.

Retained v3.9.1.20 and v3.9.1.19 behavior

v3.9.1.19 separates strategy storage ownership from the executing engine, preserves signed .noahstrategy JSON through browser download, and labels what each PAPER passport actually validates. It retains the v3.9.1.18 OKX scoring and multi-venue evidence contracts.

It also defines NoahAI's category-creation vision as building the world's first strategy validation-passport ecosystem: not the invention of backtesting, but one lifecycle from source meaning and constrained rules through fail-closed interpretation, venue/version evidence, and buyer-side re-validation. This vision never turns a passport into a profit guarantee and keeps current, source-candidate, free-beta, and later paid capabilities separate.

Strategy passport integrity

• A UNIFIED strategy closed by the native Binance PAPER engine accrues evidence to the exact UNIFIED key/version while retaining Binance as the execution venue.
• Legacy v3.9.1.18 rows are recovered only when a key/version has one unambiguous private-store owner.
• Browser downloads use Python-serialized UTF-8 JSON. Legacy browser-number normalization is accepted only when both the original IR integrity hash and package content hash verify.
• Evidence explicitly distinguishes compiled custom entry logic from NoahAI base entry + user-declared risk/exit. The latter is never presented as full source-strategy execution.
• Compiler-authored sources without executable entry nodes remain blocked until clarified or explicitly replaced by a user-declared overlay.

v3.9.1.19 release record

• v3.9.1.19 subsequently completed its Windows publication gate and its strategy-passport behavior is retained by the v3.9.1.20 public release and this v3.9.1.21 candidate.

Retained v3.9.1.18 behavior

OKX selection repair

• An OKX perpetual ticker without CCXT quoteVolume now derives USDT turnover from raw base-currency volCcy24h × last; contract-count vol24h/CCXT baseVolume is never mistaken for asset quantity.
• Repeated fallback_unscored recovery uses bounded exponential backoff from 60 seconds to 15 minutes. An identical failure snapshot is persisted at most once per 15-minute audit interval.
• This behavior was introduced in v3.9.1.18 and is retained in the v3.9.1.20 public baseline and v3.9.1.21 candidate.

Strategy validation evidence

• Stored versions expose Validation Lab sample size, net PnL, return, MDD, win rate, profit factor, out-of-sample result, walk-forward pass rate, overfit flags, and the minimum quality filter.
• PAPER evidence is grouped by venue and settlement currency. Valid results, legacy/unverified rows, wins, losses, breakeven, net PnL, and fees remain separate for KRW and USDT.
• The editor displays the supported KRW spot and USDT futures direction/order contracts before a new version is saved.
• A versioned Korean natural-language and Pine corpus verifies supported semantic preservation and fail-closed handling for unsupported Pine features and unitless TP/SL.

Strategy semantic integrity

• Confirm strategies preserve an explicit LONG or SHORT direction and cannot approve an opposite NoahAI base candidate.
• Dual-direction sources must compile separate LONG and SHORT branches; unresolved branches remain blocked for clarification.
• Supported Pine aliases are resolved, while unresolved entry variables cannot degrade into an unconditional direction.
• TP/SL require explicit units. TP, SL, position size, leverage and signal threshold use one stored/PAPER/order boundary and out-of-range values are rejected rather than silently clamped.
• Final edited rules are revalidated immediately before a new inactive version is saved.
• The structured eight-question mentor runs in the client and only loads educational drafts; it does not auto-save, approve, validate, PAPER-test, or apply a strategy.

Venue-specific historical validation

• Upbit and Bithumb replay KRW spot data; Binance, Bybit, OKX and Bitget replay USDT futures data.
• Bithumb uses its supported public 30-minute interval and the other crypto venues use 15-minute candles.
• Validation records venue, market type, interval and quote currency.

Strategy Studio now distinguishes source evidence, document/IR completeness, executable readiness and PAPER eligibility. Incomplete strategies cannot be approved. Confirm strategies explicitly inherit NoahAI dynamic exits unless the user declares strategy-owned TP/SL; independent strategies require their own unit-safe TP/SL and may define separate LONG/SHORT branches.

PAPER strategy isolation

• A PAPER validation candidate that does not match the current symbol, regime or entry condition delegates back to the base NoahAI signal instead of imposing a global HOLD.
• A strategy explicitly applied by the user remains fail-closed when its conditions do not match.
• New PAPER validation requires executable direction and declarative entry rules for independent strategies. Independent strategies own their TP/SL; confirm strategies may explicitly inherit NoahAI dynamic exits.
• Existing incomplete strategy files are preserved and are not silently rewritten or deleted.

AI report detail integrity

• Today, weekly, monthly and recent-hour details use the same NoahAI closed-trade ledger and both lower/upper period boundaries as their summary cards. Future-dated rows are excluded.
• Detail rows show venue, symbol, direction, realized PnL, PnL percentage, fees, entry price and exit price.
• Detail uses 100-row pages while a full-period count/PnL/fee checksum detects summary/detail mismatches instead of presenting them as settled results.

Current public artifact boundary

The v3.9.1.20 installer, blockmap, latest.yml, and release manifest remain the immutable public release. The v3.9.1.21 source candidate must produce new artifacts and pass docs/V39121_PAPER_PNL_COST_VENUE_ATTRIBUTION_TEST_PLAN.md; do not overwrite v3.9.1.20 assets with candidate files.