How we score a call

What gets graded

One kind of thing only: the direction a call names for that day — bullish, bearish or neutral. A call to go long counts as bullish; a call to go short counts as bearish. Nothing else on the site is covered by this rule.

The call is the AI council's. Precisely: the market_pulse.regime line of the council's output file, maplegamma_analysis.json, together with the direction of up to five trade ideas published in the same file that morning. That is the file the scorer reads, and it is subscriber data.

It is not the number on the public dashboard. The 0–10 conviction score there, and the direction word beside it, come from a different file built by a separate model — nothing scores them, and this rule does not cover them. If you are holding us to this rule, the call being graded is the council's, not that gauge.

Against what, and over how long

The benchmark is the S&P 500 index (^GSPC). The window is the same day: the index close on the day of the call, measured against the previous trading day's close. There is no multi-day horizon and no second chance — a call is settled by that one close.

What counts as a hit

The day's move sets the outcome:

• Up more than 0.1% — the day was bullish.
• Down more than 0.1% — the day was bearish.
• Anything in between — the day was neutral. That is the scratch band.

The call is correct only if it named the same one of those three. A bullish call on a day that finished inside the scratch band is wrong, not excused — the band is part of the test, not a let-off.

What the score means

The score is simply the share of scored calls that were correct. Every scored call counts once, whatever confidence was attached to it — a high-conviction call and a marginal one are worth exactly the same to the score.

What is not scored

A call is scored only where the benchmark has a close to settle it. Days the market did not trade — weekends and holidays — settle nothing, and neither does a day whose benchmark close is unavailable. Under this rule those days are left out of the score entirely, never settled against a substitute measure.

A day the council produced no call is left out for the same reason: there is nothing to grade. Nothing is filled in for it — a missing call is not recorded as a neutral one, and it counts as neither a hit nor a miss.