Week ending October 30, 2026
Posted Oct 30, 2026 after the closeWhat we saw this week
This was the week we changed how the bots learn. In practice money (paper trading), the S&P 500 closed the week -0.56%, and it was -1.77% on Friday. The table shows every bot’s week so far (paper). At about 3:55 PM ET on Friday, five minutes before the close, some bots that held only buys were down the most for the day: Patience · ChatGPT was -5.47% and Tactical · ChatGPT was -4.61% (paper). Bots already holding older short rulebooks (bets that a price will fall) held up better: Weather-aware · Claude was +0.54% for the day (paper).
The lesson: the bots saw the warning signs before the open, but their morning rule changes were not defensive. A new rulebook must first beat the old one over recent calm weeks, and calm weeks do not reward playing defence. Two bots, Tactical · Gemini and Tactical · Grok, tried to switch to short books and were turned down. Patience · ChatGPT’s 7:00 AM rewrite had also failed on a size limit, so it could not learn from the warnings at all. One day proves nothing, but it showed us a gap.
| Bot | This week |
|---|---|
| Fixed rulebook · System | +5.26% |
| Patience · Grok | +3.89% |
| Tactical · Grok | +3.00% |
| Structure & zones · ChatGPT | +2.33% |
| Super Bot Adaptive · Grok | +1.20% |
| Weather-aware · Claude | +0.89% |
| The Learner · control | +0.88% |
| Super Bot · Sol | +0.79% |
| Weather-aware · Grok | +0.51% |
| The Learner · filtered | +0.47% |
| Five Stock Expert · Claude | -0.40% |
| Super Bot · Grok | -0.62% |
| Structure & zones · Claude | -0.91% |
| Astra · composite (frozen ensemble) | -1.04% |
| Tactical · ChatGPT | -1.18% |
| Edge Bot | -1.21% |
| Moonshot · ChatGPT | -1.24% |
| Tactical · Claude | -1.41% |
| Patience · Claude | -1.76% |
| Tactical · Gemini | -1.83% |
| Patience · Gemini | -3.14% |
| Patience · ChatGPT | -5.24% |
| S&P 500 (the yardstick) | -0.56% |
What we changed
- Oct 30 Raised the size limit that had stopped two bots’ daily rewrites, so rewrites run again. Read more →
- Oct 30 Every learning bot now reads its own saved lessons, its last idea to test, its recent notes and its last ten rule changes (with how its practice account did after each) before every rewrite. It must say which it used and which it ignored, and why. Read more →
- Oct 30 Every rewrite’s full reasoning is now saved for the next one. Read more →
- Oct 30 Brought back three paper-only experiments: the five-bot learning test, the look-back review of past rule changes, and the textbook-rule strategy bots. Read more →
- From Monday’s open, Nov 2 Planned, not live yet: a defensive lane. On a confirmed risk-off turn the system will cut the size of new buys and tighten stops during the day, defensive rewrites may pass the referee the same day, and each bot picks its own sensitivity settings within set limits and must explain why. Read more →
What to watch next week
- Whether the defensive lane cuts losses on the next big down day, compared with the old way (practice money).
- Whether calm days stay calm: no extra trading and no extra rule changes.
- Whether each bot’s chosen sensitivity setting comes with a reason we can check. We judge this over weeks, not one day, and it may not turn out better.