Every Friday, the AI grades its own week.
Every Friday at 16:30 IST the weekly evaluator measures the paper-trading week (Monday–Friday): trade count, win rate, P&L, drawdown, losing streaks — and issues a verdict: READY (confidence ≥ 70), MARGINAL (50–69) or NOT_READY (below 50). The AI never switches itself to live trading; the verdict is a recommendation only, and a human decides. Every verdict so far is published below — including the failures, verbatim, unedited. If the week was bad, the reason says so in the system's own words.
Loading measured audit data… Source: live read, GET /api/public/weekly-report (app.vedkuber.in)
Methodology — how the verdict is computed
The evaluator scores the week out of 100, in five transparent components (full scoring logic is in the open repository, backend/app/services/weekly_evaluator.py):
- Trade volume — ≥10 trades: +10; 3–9 trades: +5; fewer than 3: −10 (too little data to judge).
- Win rate — ≥65%: +30; ≥55%: +20; ≥45%: +5; below 45%: −15.
- Weekly P&L — above ₹2,000: +25; above ₹500: +15; positive: +5; small loss: −10; large loss: −20.
- Losing streak — ≤2 consecutive losses: +15; ≤4: +5; more: −15 (blowup risk).
- Max drawdown — ≤5% of capital: +15; ≤10%: +5; ≤20%: −5; worse: −15.
Verdict bands: READY ≥ 70 · MARGINAL 50–69 · NOT_READY < 50. NOT_READY is the honest outcome — most weeks score it, and the reason is published verbatim below with nothing softened. All figures are backward-looking paper-trading results; past performance is not predictive. Machine-readable: /api/public/weekly-report (JSON, public).
The 4-layer report suite: Daily Public Report · Monthly Transparency Report · Rating System · Track Record