Why we grade our own calls, and how to read the record

Any tool can look right. Screenshots are cheap, winners are easy to remember, and losers are easy to forget. The only question that separates a real track record from a highlight reel is: what happens to the misses? This guide explains how our grading works, why flat days are marked neutral, and how to read any tool's record, including ours, with clear eyes.

The highlight reel problem

Trading tools are usually marketed with their best week. A screenshot of five green days, a cherry-picked month, a win rate with no methodology attached. None of that is evidence, because you are only ever shown the sample that survived. The days that would have changed your mind were quietly left out of frame.

This is why we publish no performance figures on our marketing pages at all. Not because the record is hidden, but because a number on a sales page is exactly where a number is easiest to dress up. The record lives where it can't be staged: on the dashboard, produced automatically, one row per market day.

How the grading works

Every morning the engine prints one call before the open: bullish or bearish, with confidence (the full pipeline is covered in how the engine works). After the close, the call is compared against what the session actually did, under three rules that fit on a napkin:

• Right about the direction of the close: a hit.
• Wrong about it: a miss, logged in the same row style as the hits, permanently.
• Session closed essentially flat, within about 0.15% of unchanged: neutral. Nobody honestly won a coin flip that landed on its edge, so no result is claimed.

Days written off in advance for scheduled high-impact news carry their NEWS VETO tag in the record too, so a veto can never be silently recycled as a dodged miss.

Why flat days matter more than they look

The neutral rule sounds like a small technicality. It is actually where most graded records quietly cheat. Index sessions close within a hair of flat surprisingly often, and a tool that counts every one of those as a win, since technically the close was two points in its favor, inflates its record with days where nothing it predicted actually happened. Marking them neutral costs us wins on paper. That is the point.

How to read any track record

These questions apply to every tool that claims a record, ours included. A record that cannot answer them is a story, not a record.

• Are the losses visible? Not a loss percentage, the actual losing days, in the same log as the winners. If you cannot point at a specific bad day, you are looking at marketing.
• How are flat and unreadable days handled? Wins, losses, or honestly set aside? This one rule tells you more about a tool's integrity than its win rate does.
• Were the calls timestamped before the fact? A call worth grading is issued before the open and never edited. Anything reconstructed after the close grades itself generously.
• Is the methodology fixed? Same rules every day, not a definition of "win" that migrates when the week goes badly.
• Is the sample honest? Ten days of anything means little. A record earns weight slowly, day by day, which is exactly why it has to be automatic.

The misses are the product

There is a second reason we grade every call, beyond honesty: the misses are fed back into the engine. When a call is wrong, the factor weights that produced it are adjusted before the next open. A record that hid its misses would not just be dishonest, it would starve the model of the only thing it learns from. The public grading and the self-improvement loop are the same mechanism.

And the standing reminder: a track record, however honest, is a history, not a forecast. Nothing here is a promise of future results, and no record replaces your own judgment and risk management.