Results, as the bot recorded them
20 September – 26 October 2026 · BTC/USDC · 4h candles · 10 000 USDC simulated capital · paper book with a testnet executor.
The ledger
Every number on this page is read from the bot's own runtime files (statistics.json,api_costs.json, trade_history.db, metrics/cycles.jsonl) plus one public Binance klines request for the benchmark. Nothing is typed from memory:python scripts/results_report.py prints it, anddocs/RESULTS.mdis its output.
| Closed trades | 12 |
|---|---|
| Winners / losers | 5 / 5 |
| Win rate | 41.7% |
| Return | +0.14% (+$13.95) |
| Profit factor | 1.41 |
| Sharpe / Sortino | 0.19 / 0.36 |
| Max drawdown | −0.25% |
| Best / worst trade | +4.06% / −2.60% |
| Inference cost | $6.02 (8.1M in / 3.9M out tokens) |
| Net after inference | +$7.93 |
| Analyses | 50 cycles · median 266 s · $0.0117 each |
Versus holding the coin
Over the same 36 days the underlying went from 81 245.48 to80 036.26 — a −1.49% buy-and-hold (only closed 4h bars). The bot finished +0.14%.
Closed trades
| Closed | Side | P&L | Exit |
|---|---|---|---|
| 24 Sep | LONG | −2.60% | closed |
| 24 Sep | LONG | −2.06% | stop loss |
| 28 Sep | LONG | −0.28% | closed |
| 02 Oct | LONG | −0.99% | closed |
| 05 Oct | LONG | −1.12% | stop loss |
| 10 Oct | SHORT | +3.45% | take profit |
| 12 Oct | SHORT | +2.04% | take profit |
| 14 Oct | SHORT | +0.07% | stop loss |
| 15 Oct | SHORT | — | excluded (fill divergence) |
| 17 Oct | SHORT | — | excluded (fill divergence) |
| 26 Oct | SHORT | +1.93% | take profit |
Caveats
- Twelve trades cannot separate skill from noise. The sample is too small to read as a track record.
- Fees are configured but not booked. The paper book records
fee = 0.0(or no fee at all) on its rows, so the +$13.95 is gross of trading costs; at the configured 0.075% per side it would mostly disappear. See where the numbers come from. - Two trades are excluded on purpose — both sandbox exits printed below the real market low in the same minute, so they stay in execution history but not in statistics or in the brain's learning.
- This is a paper book. The executor can reach an exchange testnet, not live funds.
Full methodology, per-trade table and the reproduction command:docs/RESULTS.md. Watch the next cycle live on the dashboard.