# Prospective PSX shadow-paper study

Research only. No broker connection, order placement, or live-capital authorization.

- Study: `prospective_v1`
- Frozen strategy: `psx_xgb_5d_topn_v1`
- Fingerprint: `a8cf67171c3e6d2e478b923d071bc5c281ccbf7afe7cd4f7129ff7029378d733`
- Start date: 2026-08-19
- Sessions processed: 0 / 60 minimum / 120 preferred
- Audit dates recorded: 0
- Top-1 generated / entered / completed: 0 / 0 / 0
- No-signal sessions: 0
- Data-incomplete sessions: 0
- Operational cancellations: 0
- Open positions: 0
- Next gate: `30_SESSION_OPERATIONAL_REVIEW`
- Current classification: `COLLECTING_PROSPECTIVE_EVIDENCE`

Performance conclusions are prohibited before the pre-registered review gates. At 60 sessions, fewer than 25 completed Top-1 trades is `INSUFFICIENT_SAMPLE`. At 120 sessions, fewer than 40 is `INCONCLUSIVE_LOW_SAMPLE`.

## Frozen operating contract

`paper-eod` validates same-session equities, benchmark, structural OHLC, liquid-universe size, official point-in-time membership, frozen features, model artifact, and strategy fingerprint. Each date receives one immutable result: generated candidates, `NO_ELIGIBLE_SIGNAL`, or `SKIPPED_DATA_INCOMPLETE`. A skipped date cannot be backfilled later.

Original signal fields stay in append-only `signals.jsonl`. Reviews, entries, exits, cancellations, and corporate-action discoveries are separate append-only events in `operational_log.jsonl`.

Human review permits only objective operational checks: tradability, suspension, market closure, corporate action, symbol/source error, or membership error. Directional opinion, news sentiment, chart judgment, or model disagreement is prohibited.

`paper-update` requires `PREMARKET_REVIEWED`, records next observed PSX-session OPEN, then exits at CLOSE of fifth held session under frozen T+5 semantics. Primary net return uses 0.25% round-trip cost; 0%, 0.10%, 0.50%, and 1.00% remain separate sensitivities. Entry-open stress cases remain simulated, never actual slippage.

Top-1 and Top-3 use separate PKR 1,000,000 five-sleeve portfolios. Top-3 splits each sleeve equally across actual eligible names. Fractional-share accounting is primary; whole-share output is sensitivity only.

Random controls are deterministically assigned from each day's exact eligible candidate set before outcomes exist. Drift is observational against frozen Task 6 reference features; no automatic retraining or filter is permitted.

## Pre-registered gates

- 30 sessions: `OPERATIONAL_REVIEW_ONLY`.
- 60 sessions and 25 completed Top-1 trades: PASS requires positive 0.25%-cost mean return, PF above 1.20, positive portfolio return, drawdown better than -25%, no material flaw/leakage, no majority-P&L symbol, and acceptable completeness.
- 120 sessions: classify exactly `PROSPECTIVE_CONFIRMED`, `PROSPECTIVE_FRAGILE`, `PROSPECTIVE_FAILED`, or `INCONCLUSIVE_LOW_SAMPLE`.