How MarketBotsLab works

The short version

• We scan a universe of liquid US stocks for classic chart patterns that often come before a move higher.
• Each setup gets a 0–100 score from the same rulebook, so setups can be compared fairly.
• We separately track stock picks made by finance YouTubers and buying by company insiders.
• Every signal we publish is kept, and its result is measured afterwards — winners and losers.

How stock ideas (breakouts) are found

A “breakout” is when a stock pushes above a price level it has repeatedly failed to clear. The scanner looks for 23 well-known pattern types — rectangles, cup-and-handles, flags, wedges, triangles, head-and-shoulders and others — using daily price and volume data.

Each pattern is followed through stages (forming → near the trigger → triggered → following through or failed), so you can see whether a setup is still early or already playing out.

What the 0–100 score means

The score combines pattern quality, volume confirmation, trend context and market conditions on a single 100-point scale. Rough bands:

• 85+ Elite — the rules are strongly satisfied.
• 70–84 Strong — most rules satisfied.
• 55–69 Watchlist — worth watching; wait for confirmation.
• Below 55 — ignore.

A score describes how well a setup matches the rules. It is not a probability of profit. Whether higher scores actually lead to better results is measured on the scorecard (see diagnostics).

How we test without hindsight

Backtests use walk-forward testing: rules and weights are only ever evaluated on data that came after the period used to set them. Market regime (for example trending vs. choppy markets) is recorded at the time of each signal, so results can be split by the conditions that existed then — not by what we know now.

The public track record

Every published signal is logged when it fires and never edited afterwards. We then measure its forward return after 5, 10, 20 and 60 trading days, plus the worst drop along the way (“maximum adverse excursion”). The scorecard groups these by pattern, score band and market regime; the transparency report lists them month by month.

Synthetic test rows used during development are excluded from all public figures, and repeated detections of the same setup are counted once.

How influencer picks are tracked

We read the transcripts of finance YouTube videos and extract explicit stock calls — buy, sell, watchlist and so on — together with the quote they came from, so every pick can be traced back to the moment in the video. Calls that are ambiguous (for example a ticker that could mean two companies) are held back for review rather than guessed.

Each pick’s performance is measured from the recommendation date and compared with the S&P 500 (SPY) and Nasdaq-100 (QQQ) over the same period. Influencers are ranked on this benchmark-adjusted return, not on how often they are right.

Insider buying

Company officers and directors must report their trades to the SEC (Form 4). We highlight clusters — several insiders buying the same stock with their own money within a short window — which historically carries more information than a single purchase.

Options data

Options pages use delayed (about 15 minutes) CBOE quotes: real open interest, spreads and implied volatility. Greeks are calculated by us with the Black-Scholes model so the maths can be checked. Income strategies (covered calls, cash-secured puts, the wheel) are ranked on premium relative to risk.

Diagnostics: calibration and health checks

Some pages have a collapsed Methodology / diagnostics panel. It holds the internal checks we run on our own numbers. We show them because hiding them would be worse:

• Calibration asks: when the score is higher, does the signal actually work more often? We measure it with the Brier score and compare it with a naive guess that always predicts the average win rate (the “base rate”). Uncalibrated means the score does not yet beat that naive guess, so you should not rank setups on score alone.
• Out-of-sample skill repeats the check using only data the model had not seen, which is the honest version of the test.
• Negative-expectancy patterns are pattern types that have lost money on average over a reliable sample. They are flagged so we can review or retire them.
• Review backlog is the number of extracted influencer calls waiting for a human check.

Limitations

• Past results do not guarantee future results. Samples for some patterns are small.
• Data is delayed and can contain errors from upstream providers.
• Returns shown ignore commissions, taxes, slippage and position sizing.
• MarketBotsLab is educational research, not a registered investment adviser.

Questions or corrections? Contact us.