Every read the desk produces is recorded at the time, with its claim stated precisely enough to be marked, and scored later against a rule fixed in advance. None of that is published.
That looks evasive. The reasoning is worth setting out.
A published record becomes a sales claim
The moment a hit rate appears on a marketing page, it is a performance claim, and performance claims carry obligations — about the methodology, the completeness of the sample, the period covered, and what a reader is entitled to infer. Most published track records in this industry fail those obligations badly.
The common failures are familiar: selected periods, claims vague enough to be scored generously, losers quietly dropped, the rule adjusted after the fact. A number produced that way is worse than no number, because it looks like evidence.
The sample is not there yet
Even done honestly, a record needs enough cases across enough different market conditions to mean anything. A hit rate over a few months of one regime is a description of that regime. Publishing it would invite an inference the data cannot support.
What the record is for internally
Calibration. Finding out where the desk is systematically overconfident and correcting it. That is the entire value, and it works whether or not anyone outside sees the numbers.
It only works if the record is append-only. A scored read that turns out badly gets annotated, never rewritten, because a record you can edit is one that will always agree with you.
What would change this
A long enough series across several regimes, a scoring rule published in advance, and every case included. If those hold, the numbers can be published. Until then, claiming them would be the exact thing we are trying not to do.