Day Trader Skill for Claude: What the Time Machine Odds Mean
By Matt Cullerton, founder of Mavric and Apso
day-trader-skill is a free, MIT-licensed set of eleven Claude skills for day trading, published on GitHub on October 20, 2026. Its "Time Machine" skill replays the rest of a trading day 10,000 times from a stock's recent 5-minute bars. A TikTok from @buildwithai.club calls the resulting 29.4 percent "the historical odds." That figure is a simulation, and the repository's own README reports a different historical result for the same trade: 13 of 33.
What is day-trader-skill?
The repository is Jakeschincariol/day-trader-skill. Its single commit is titled "The Trader skill v2: 11 day-trading skills," and the GitHub API lists it as MIT licensed and not a fork. On 2026-10-26 it had 1,845 stars and 390 forks. Many identical copies exist under other accounts, and the video never names the author or shows a URL.
The eleven skills, with the names the video uses:
- Chart Scanner (
trader-scanner) - News Monitor (
trader-news) - Entry Timer (
trader-entry) - Exit Planner (
trader-exit) - Time Machine (
trader-timemachine) - Risk Manager (
trader-risk) - Budget Manager (
trader-budget) - Order Executor (
trader-broker) - Price Alerts (
trader-alerts) - Earnings Analyst (
trader-analyst) - Pelosi Tracker (
trader-pelosi)
The last two read SEC filings and House Clerk disclosure data rather than price charts.
How does the Time Machine skill work?
The code in trader-timemachine/timemachine.py defines simulate(... replays=10000, seed=19, window=2 ...). Its docstring says it runs "Monte Carlo replays of bars k+1..last." For each future bar, it picks a past bar from the same time of day (within two slots) at random from the stock's previous sessions, and scales it to the current price. This is resampling with replacement, the bootstrap method from Efron (1979), applied to 5-minute candles.
The fixed seed means anyone can reproduce the README's numbers.
Are the Time Machine results historical odds?
No. The video says "you get to see the historical odds before putting your money on the line." Its own graphic labels the numbers "SIMULATED ODDS," and a tag on screen reads "EXAMPLE RUN · FROM THE README."
The README's worked example is a Tesla trade on Monday, October 19, 2026. It reports two results for the same trading rule:
- Simulated from 10,000 replays drawn from the prior 59 sessions: target first 29.4 percent, stop first 25.5 percent, still open at the close 45.1 percent.
- Historical, re-running the rule on those 59 real sessions: it fired 33 times, with 13 hitting the target (39.4 percent), 15 the stop (45.5 percent) and 5 neither (15.2 percent).
The README's verdict is "close to a coin flip." The video shows only the simulated numbers, which differ from the historical ones by 10 points on target, 20 on stop and 30 on "neither."
What do 10,000 replays actually tell you?
They reduce the simulator's own sampling noise and nothing else. At 29.4 percent, 10,000 replays give a 95 percent margin of about 0.89 percentage points. A thousand replays would give about 2.82. Every replay draws from the same 59 past sessions, so running more replays does not add information about the market.
The historical check is much less precise. With 33 trades, the 95 percent range for 13 of 33 runs from 24.7 to 56.3 percent. That sample cannot tell a good setup from a bad one, while the simulated figure looks far sharper than the data supports.
Why do the simulated and historical results differ?
The simulator draws each bar independently, which removes the tendency of big moves to follow big moves. That clustering is a documented property of asset returns, catalogued by Cont in Quantitative Finance (2001). Independent draws are known to be a poor fit for dependent data, which is why block bootstrap methods exist, including Künsch (1989) and the stationary bootstrap of Politis and Romano (1994).
Without clustering, simulated paths rarely run far enough in one direction to hit either the target or the stop, so they pile up in the "still open" bucket. The simulation put 45.1 percent there, while the real sessions put 15.2 percent there.
Does the scanner really check thousands of charts?
The video says one skill "scans thousands of stock charts," with a "THOUSANDS OF CHARTS" card on screen. The README's documented run scanned 231 charts and found 42 that had closed above their 15-minute opening range. The scanner reads Nasdaq Trader symbol lists, so it can cover more, but 231 is the only documented run.
The video never shows the tool running. There is no terminal, no Claude Code session and no repository page, only motion graphics.
What risks does the repository disclose?
The README's fine print is more careful than the video. It says "The Time Machine is a simulation, not a prediction. It is built from the stock's own last 60 days of 5-minute bars. The future can do things those 60 days never did: an earnings report, a halt, a headline." It also notes:
- "Most people who day trade lose money."
- The Yahoo Finance chart endpoint is "not an official API: it can lag, change or rate-limit you."
- Only 5-minute bars are used, and runs below 20 full sessions are refused.
- A bar that touches both stop and target counts as a stop, and a gap through the stop is scored as -1R, while "a real fill can be worse."
Neither source says how many rule settings were tried before the documented one (a 15-minute opening range, a stop at the range midpoint, a 1x risk target). Bailey, Borwein, López de Prado and Zhu argue that backtests that "almost never declare the number of trials" are likely to report false positives. The video's only caveat is "Of course, it's a simulation, not a guarantee."
Bottom line
day-trader-skill is a real, free set of Claude skills, and its Time Machine is a reproducible bootstrap simulation. Its 29.4 percent is a simulated figure, and the repository's own replay of real sessions put the same trade at 13 of 33, with too few trades to judge either way. Read the README's fine print before trading on any of it.
Sources
- Jakeschincariol/day-trader-skill - the repo and its eleven skills.
- day-trader-skill README - the Tesla example, both result sets and the fine print.
- timemachine.py - the simulation code and defaults.
- Time Machine SKILL.md - the method description.
- GitHub API record - creation date, license and counts.
- Efron (1979), Bootstrap Methods - the bootstrap.
- Politis and White (2004), block-length selection - block bootstraps for dependent data.
- Bailey et al., The Probability of Backtest Overfitting - undeclared trial counts.
Published October 23, 2026. Based on a TikTok by @buildwithai.club. Full video notes (markdown)