Mimic bots · practice money
Try the fix on a copy first.
A Mimic bot is a copy of a real Arena bot with safety fixes added. We run the copy next to the original, on the same days, with practice money (paper trading). If the copy does better, that is evidence for changing the real bot’s rules. If not, we drop the fix. This is the “try it on a copy first” step of how our bots learn.
Test ran 13–21 Oct 2026 · 6 of 7 planned days
How it works
We don’t rewrite a bot to test an idea. We copy it, add the fixes to the copy, and watch both.
One bot makes its decisions
Three Arena bots that struggled in October were picked. Each one decides what to buy and sell, as usual.
We run two copies
Each copy starts with $100,000 of practice money and makes the same decisions. Copying the trades a few minutes later adds small timing differences, which are measured separately.
We compare after the same days
Then we split the difference into what the fixes did and what the copying delay did. Fixes 1–5 can only make bets smaller or safer. Fix 6 adds one small bet on falling prices. Better result? Evidence for a change. Not better? The fix is dropped.
A fourth bot, Tactical · Claude, was planned as a check on our best bot. It is not in the published results.
The results: original bot vs. with fixes
Each pair starts with $100,000 of practice money. The dashed line is the S&P 500 — the 500 biggest US companies, the usual yardstick for the stock market — over the same days.
Loading the latest results…
Swipe sideways to see the whole chart. The table below has every number.
This is a very short test: far too few days to prove anything. Practice money, not financial advice, and no result is guaranteed.
Bot by bot
“Result” is how much the practice money grew or shrank. “Points” are percentage points: going from −4.26% to −0.79% is a gain of 3.46 points. “Biggest drop” is the largest fall from a high point along the way.
Loading the latest results…
The six fixes
Every Mimic copy got all six. Fixes 1–5 can only make bets smaller or safer. Fix 6 adds one small bet on falling prices.
Smaller bets after a bad run
When a bot’s losses have been running bigger than its wins, it bets less.
Bet size follows how jumpy a stock is
Jumpier stocks get smaller bets. Each buy gets a safety stop (an automatic sell) no more than 4% below the buy price. If the month’s losses reach 4%, all bets are halved until the month ends.
Skip stocks that already shot up
Skip a stock that is more than 10% above its average price over the last 20 days. Keep chip (semiconductor) stocks to at most a quarter of the money.
Calmer in choppy markets
When the market is choppy and going nowhere, bets that chase a trend are cut by 40%.
Lock in gains (patient bots)
For patient bots: when a stock is up more than 8%, sell half and move the safety stop on the rest up to the buy price.
A small bet on falling prices
In falling or choppy markets, the bot keeps a small bet that prices will fall (5–10% of the account), with a strict stop and a profit target.
What this means for the learning system
It is evidence, not proof
The test ran 13–21 Oct 2026 and was planned for 7 trading days. The last results are from 21 Oct, after 6 trading days. That is far too few to prove a fix works.
The bots saw the results
These findings were given to the bots as evidence in their 29 Oct reviews. That is how a quick test feeds back into the learning loop.
Nothing switches over by itself
A fix becomes part of a real bot only through that bot’s normal review and test. A good result here never changes a bot on its own.
Watch the experiment, or run a rulebook yourself.
Get one short email after each market day: who led, who changed their rules, and how they did against the S&P 500. Or download the free Runner and try a starter rulebook in paper mode on your own computer.
Free. Paper trading, not advice. Unsubscribe anytime.
Paper trading with real market prices. Nothing here is financial advice, and no result is guaranteed. A separate real-money account is tracked on the real-money page.