The Data Desk · Season Scorecard

The 2026 Scorecard

The accuracy ledger, read whole: every call our models and the tracked commentators logged across the 2026 season — rounds 1–27, preseason calls included — graded against what actually happened. Corrections stay visible, populations are stated, and what could never be graded is counted rather than buried.

As at 28 Oct 2026 · Season 2026 · Vol. 1

The price model, all season

49.9%

read the direction of 2,319 price moves (up, flat or down) · average miss $27.4K · 2,546 withdrawn — better than guessing, though not by a margin worth leaning on.

Best score model

18.6 pts

THESCIP Score Model (XGBoost) average miss over 4,895 graded projections, of which 2,510 were players who did not play — graded as misses by ruling, no error recorded.

The engines

Our own predictors, one line per published record. A line below the 20-call bar shows its counts and no rate. Trade lines cover only calls whose full 3-round grading window was played and closed; grades the window rule excludes are counted in the final column, not hidden.

Predictor	Graded	Hit rate	Avg miss	Withdrawn	Never gradeable	Pending	DNP misses
THESCIP Score Model (XGBoost)round-score projections	4,895	25.7%	18.6	429	914	0	2,510
THESCIP Price Modelprice-change calls	2,319	49.9%	$27.4K	2,546	564	0	0
THESCIP Scoring Engineround-score projections	2,050	12.2%	19.2	438	3,846	0	1,574
THESCIP Score Model (ridge)round-score projections	550	16.9%	28.3	4,962	43	0	281
THESCIP Trade Enginetrade-in calls	218	56.4%	15.6	14	57	28	33
THESCIP Two-Trade Plannertrade-out calls	101	44.6%	19.3	22	131	3	0
THESCIP Trade Enginetrade-out calls	53	79.2%	24.4	0	100	17	0
THESCIP Two-Trade Plannertrade-in calls	34	73.5%	19.1	10	20	1	3
THESCIP H2H Captain Verdictcaptain picks	8	below the 20-call bar		0	3	0	1
THESCIP Replay Enginecaptain picks	3	below the 20-call bar		0	0	0	0
THESCIP Replay Enginetrade-in calls	2	below the 20-call bar		0	0	0	0
THESCIP Replay Enginetrade-out calls	2	below the 20-call bar		0	0	0	0

Population: season 2026 ledger rows, active graded per line · hit-rate denominator = graded (a neutral verdict counts as a non-hit) · price average miss in dollars, all others in SC points · captain error measured from the recorded score against the 100-point bar.

The commentators

Commentator grades are withheld while we re-measure them. An audit of how calls are read from podcasts and videos found that many stored calls were not genuine calls, so the grades built on them are being rebuilt and checked. The commentators and their calls are still tracked; the grades return once that work is done.

Provenance

Every figure on this page is recomputed from the prediction ledger on each read (last grade landed 28 Oct 2026), scoped to season 2026 only.

The read-time rules

Trade figures cover only calls whose full 3-round grading window was played and closed. Grades recorded on shorter windows are excluded and counted, per line.
Captain error is measured from the recorded score against the 100-point bar, never from the stored error column.
A call logged after its round locked out is never gradeable: it is counted, kept on the record, and graded as nothing.

Known edges

0 graded round-score actuals were later restated by SuperCoach; the ledger keeps the grade as it stood when the round settled.
Withdrawn price calls — players who could not move by rule — stay counted beside the price record permanently; put back, the record would read 31.2% over 3,959 calls at $25.4K average miss.
Selection and vice-captain calls run in shadow mode and are ungraded by design; they appear nowhere on this page.

The live, round-by-round view is the Accuracy Ledger. · For entertainment purposes only · Legal & Disclaimers