Bot report: September 2026
Our paper bot's first (half) week: 14 closed trades, +0.63R, -0.84% on the virtual account. The numbers, what the learner changed and how it compares with the 6-year backtest.
Our self-learning futures bot started running in paper trading mode (virtual money) on the evening of 26 September. So this “monthly” report really covers five days. There is very little data, and we would rather say so up front than dress it up.
In brief
- Period: 26-30 September 2026 (the bot started on 26 September), all paper trading.
- Virtual account: $100 → $99.16 (-0.84%).
- Closed trades: 14; 6 wins, 8 losses (43% win rate), +0.63R in total.
- Learner: capital was split into a 2-hour and a 4-hour book, and the AlphaTrend strategy was added to both.
- Caveat: this few trades says nothing conclusive.
The numbers
| Metric (paper trading) | September (26-30) |
|---|---|
| Start / month-end virtual balance | $100.00 / $99.16 |
| Return | -0.84% |
| Closed trades | 14 (6 wins / 8 losses) |
| Win rate | 43% |
| Total R | +0.63R |
| Best / worst trade | +2.91R / -1.11R |
Nine of the trades were squeeze_expansion shorts: two hit their 3R target (+2.91R and +2.63R), seven were stopped out, for roughly -2R net. rsi2_reversal (+1.42R) and donchian_breakout (+1.27R) finished positive. The trades added up to +0.63R while the balance ended slightly down; the gap most likely comes from the core basket, which holds half the capital, and from costs.
On signals, the published list only keeps the latest 100, and the oldest is from 30 September 16:00. Within that window, 1 of the 19 September signals was opened, 15 were skipped because that strategy was switched off, and the rest were blocked by sizing or ban-list rules; none were missed. A full-month signal count is not available.
What the learner changed
The first automatic 100-trade relearn has not happened yet (about 97 trades to go). The changes this month were configuration-level:
- 27 September: capital was split in two: $50 for the 2-hour book and $50 for the 4-hour book. The core basket was limited to the top 100 coins by volume.
- 29 September: AlphaTrend was added to both books after passing a 20-seed test. Risk per trade in the 2-hour book went from 0.25% to 0.5%.
Right now 8 strategy/side combinations are enabled in the 2-hour book and 7 in the 4-hour book (details on the dashboard).
Compared with the backtest
The 6-year backtest (2020-2026) of the live 2-hour setup averaged +2.47R per 100-trade block, about 0.025R per trade. September’s 14 trades came in at ~0.045R per trade, above that average, but 14 trades is not even a seventh of one block. The same backtest had a maximum drawdown of -64%, so a good long-run average does not rule out deep losing stretches.
Our take
Five days and 14 trades are far too few to judge a bot. What we can say so far: the rules behave as designed live, losses are cut around 1R, and a few large winners balance the picture. A meaningful read needs the first few hundred trades.
This post is not investment advice. All results are paper trading (virtual money) or backtest results; past and backtest results do not guarantee future results. Leveraged trading can lose all of your money.
Comments
Become a free member to comment and like.
No comments yet. Be the first.