Calibration & misses

On a market cohort, how wrong is fair AAV — and are residuals data or model?

Market MAE 1.16 pp · n=349 · 2025-26

Next · Weekly MAE changelog after each dial.

Residuals for v1.38. Primary target is %cap (signing share when known): fair%cap − actual%cap. Positive = model above paycheck (underpay). Dollar MAE is secondary.

715 scored · generated Oct 1, 2026, 8:17 PM

Market cohort

n: 349
MAE |%cap|: 1.16 pp
Mean %cap surplus: -0.56 pp
$ MAE (ref): $1.05M
Labels: F 292 · O 37 · U 20

Curated 40+ GP

n: 485
MAE |%cap|: 1.19 pp
Mean %cap surplus: -0.25 pp
$ MAE (ref): $1.15M
Labels: F 424 · O 39 · U 22

ELC / bridge

n: 258
MAE |%cap|: 0.90 pp
Mean %cap surplus: +0.60 pp
$ MAE (ref): $0.93M
Labels: F 258 · O 0 · U 0

Recent vs legacy (market)

Cut: start_season ≥ 2023-24 (last 3 seasons). Rows without start_season omitted.

Recent deals

n: 273
MAE |%cap|: 1.07 pp
Mean %cap surplus: -0.55 pp
$ MAE (ref): $0.98M
Labels: F 240 · O 21 · U 12

Legacy deals

n: 76
MAE |%cap|: 1.48 pp
Mean %cap surplus: -0.61 pp
$ MAE (ref): $1.31M
Labels: F 52 · O 16 · U 8

Holdout slice (market)

every 5th market player by playerId. Use holdout MAE as a quick overfit check when dials move.

Holdout

n: 70
MAE |%cap|: 1.35 pp
Mean %cap surplus: -0.52 pp
$ MAE (ref): $1.18M
Labels: F 54 · O 9 · U 7

In-sample

n: 279
MAE |%cap|: 1.11 pp
Mean %cap surplus: -0.57 pp
$ MAE (ref): $1.02M
Labels: F 238 · O 28 · U 13

MAE by position (market, %cap pp)

C n=1101.18 pp$1.01M
LW n=561.34 pp$1.43M
RW n=521.11 pp$1.00M
D n=1081.09 pp$0.95M
G n=231.04 pp$0.92M

Misses constellation

Largest |%cap residual| · size = |pp| · color = trust

high · med · low

Largest |%cap residual|

Market cohort only · top 25

RESEARCH NOTE →

How to use this

Open a miss → confirm AAV / quality on the player page.
If AAV is right, note position pattern (D vs F vs G).
Change a dial → npm run contracts → re-check market + recent MAE (legacy can lag on old deals).