The desk's own marks understate both excursions

Found 2026-09-03, comparing the scorer's tape-derived MFE/MAE to positions Status open — informational, but it biases a specific diagnosis

What we saw

The desk records mfe_r and mae_r on the position row, computed from marks it samples during the ride. The tape is not sampled. They disagree, and the disagreement has a direction:

ride  symbol   measure   desk       tape       gap
137   GPRO     MFE       +0.88 R    +1.79 R    tape is 2.0x further
133   DAKT     MAE       -1.08 R    -1.57 R    tape is 45% deeper
127   LHAI     MFE       +2.05 R    +2.09 R    close
127   LHAI     MAE       -0.46 R    -0.52 R    tape is 13% deeper
116   FRVO     MAE        0.00 R    -0.22 R    desk saw no adverse move at all

Every gap runs the same way: the desk's marks are inside the tape's extremes, never outside them. That is what sampling does — a mark taken every few minutes cannot see a spike between marks, and extremes are exactly the thing that lives between samples.

(GPRO's MFE gap is partly a different effect: the desk stops marking near the close while the raw tape continues into extended hours. With the horizon fix applied its tape MFE drops to +0.84R, close to the desk's +0.88R. The DAKT, LHAI and FRVO gaps are inside regular hours and are pure sampling.)

Why it matters

MFE and MAE are not decoration. They drive:

capture = realised ÷ MFE — understating MFE makes capture look better than it was, so a trader who left a lot on the table looks like they took most of it.
the "cut winners early" diagnosis, which is entirely an MFE comparison.
stop-distance analysis, which is an MAE comparison. FRVO's desk MAE of exactly 0.00 would say the stop was never approached; the tape says the trade went 0.22R against before working.

So sampled marks systematically flatter the trader on adverse moves and hide how much was available on favourable ones. Both errors point toward "nothing to coach here".

What we did

Nothing yet. The scorer computes its own MFE/MAE from the tape and reports the desk's alongside for comparison, which is how this was noticed. Neither number overwrites the other.

Still open

Quantify it across all 139 closed rides rather than the four sampled here. The direction is consistent but the magnitude is not established.
Decide which number the digest shows. The tape figure is more accurate; the desk figure is what the trader saw at the time. For coaching, "what was actually available" is the right one — but the gap between them is itself worth showing when it is large.
Only 86 of 139 closed rides carry mfe_r/mae_r at all (62%), so for a third of the book there is nothing to compare against and the tape is the only source.