Two ways a decision system can be wrong, and the surfaces that measure each. Grouping rejections by reason, applying a settings change through a server-clamped audited path, and reading the calibration report including which number to trust when two disagree.
Play Reject analytics and calibration. The video loads from YouTube only when you press play, so nothing is requested from them before that.
Prefer to watch there? Open this video on YouTube.
A system that decides things can be wrong in two directions. It can turn down trades it should have taken, and it can be overconfident about the ones it takes. This tab covers the first. Calibration covers the second. Most platforms show you neither.
Reject analytics groups every rejection by reason and counts them. If one reason dominates for weeks, that is worth understanding rather than overriding. It might be your settings, and it might be the market. A liquidity floor rejecting everything can mean your floors are too high, or it can mean you are pointing the agent at names that do not have the open interest to support the trade you want.
The analyze card surfaces near-misses: candidates that failed on one threshold by a small margin, alongside the settings change that would have let them through. Three things about that button. The candidates are derived on the server, not invented by a language model. It takes your click, it never applies anything on its own. And the server clamps whatever it applies to a sane range and writes it to an audit log. Loosening a floor because it keeps rejecting things is exactly how people talk themselves into bad trades, so the path is deliberately slow.
Calibration asks whether the probabilities were honest. Each forecast bucket is shown against what actually happened in your closed trades, with a confidence range around it, and a status chip reading OK, watch, or alert. The asymmetry here is important. Doing better than forecast is expected for premium selling and never raises an alarm. Doing worse than forecast, beyond the confidence range, is the dangerous direction, and two or more buckets on that side raise an alert and log an issue overnight.
Two reading rules. When the status is not OK, trust what your closed trades actually did over the forecast. And when the pooled comparison and the held-to-expiry comparison disagree, trust held-to-expiry, because that is the honest test of a probability-of-profit number.
That closes this series. Optionscanner is operated by Avenix Solutions LLC. It is not a broker-dealer and not an investment adviser, and it never holds customer funds. Options involve risk and are not suitable for every investor. Thanks for watching.