GridIQ · 2026 season · after round R13

How far off the GridIQ projection is

Every figure below can be checked: the projection we recorded against the points the game awarded. Each round is stamped with when its projection was recorded — before the deadline or after the race. The whole dataset is available as CSV and JSON.

driver error · frozen before lock10,7points / driver / roundrounds: 4 · projections: 88

driver error · everything recorded12,3points / driver / roundrounds: 13 · projections: 287, reconstructed ones included

driver top-3 called3 of 7walk-forward, rounds R5–R11 — the same run the guide quotes

frozen before lock projection recorded before the deadline, time shownreconstructed model re-run afterwards on data available before the round only (walk-forward)after deadline recorded after the deadline — excluded from the headline figure

By round

Mean absolute error for drivers and constructors. The headline figure above counts only rounds marked “frozen before lock”.

Round	Projection recorded	Error · drivers	Error · teams	Top-3	Status
R1 Australian Grand Prix	08/07 · after the race	15,2	28,0	0 / 3	reconstructed
R2 Chinese Grand Prix	08/07 · after the race	16,6	26,1	2 / 3	reconstructed
R3 Japanese Grand Prix	08/07 · after the race	8,5	15,0	2 / 3	reconstructed
R4 Miami Grand Prix	08/07 · after the race	15,9	23,0	1 / 3	reconstructed
R5 Canadian Grand Prix	08/07 · after the race	12,2	12,5	2 / 3	reconstructed
R6 Monaco Grand Prix	08/07 · after the race	15,4	16,6	1 / 3	reconstructed
R7 Barcelona Grand Prix	08/07 · after the race	11,9	15,4	1 / 3	reconstructed
R8 Austrian Grand Prix	08/07 · after the race	8,8	13,3	2 / 3	reconstructed
R9 British Grand Prix	03/07, 18:07 · 17 h before lock	11,9	17,1	1 / 3	frozen before lock
R10 Belgian Grand Prix	06/07, 00:00 · 13 d before lock	10,0	13,2	1 / 3	frozen before lock
R11 Hungarian Grand Prix	19/07, 18:00 · 6 d before lock	9,1	16,5	0 / 3	frozen before lock
R12 Dutch Grand Prix	27/07, 00:00 · 26 d before lock	11,6	15,8	1 / 3	frozen before lock
R13 Italian Grand Prix	07/09 · after the deadline	12,7	20,7	1 / 3	after deadline

By driver

Error is the mean absolute miss per round. Bias carries the sign: “+” means we underrated the driver, “−” overrated. The frozen columns count only rounds whose projection was frozen before lock.

Asset	Error · all	Bias	Error · frozen	Bias	Rounds
BOR G. Bortoleto	5,1	−0,9	4,1	+0,4	13 · 4
OCO E. Ocon	6,7	+1,4	9,8	−4,9	13 · 4
COL F. Colapinto	6,8	+3,5	8,2	+0,3	13 · 4
LIN A. Lindblad	7,0	+0,8	5,2	0,0	13 · 4
SAI C. Sainz	7,9	+1,6	1,5	+0,8	13 · 4
GAS P. Gasly	8,5	+1,2	5,6	−1,1	13 · 4
ALB A. Albon	9,2	+0,8	7,1	−0,1	13 · 4
BOT V. Bottas	10,9	−6,2	11,1	−4,3	13 · 4
BEA O. Bearman	11,0	−5,4	8,0	−2,8	13 · 4
HUL N. Hulkenberg	11,2	−2,3	6,3	+2,0	13 · 4
LAW L. Lawson	11,2	+4,2	7,2	+0,3	13 · 4
PER S. Perez	11,3	−2,4	13,1	−7,1	13 · 4
HAD I. Hadjar	12,9	+1,3	12,5	+2,4	13 · 4
PIA O. Piastri	13,1	+2,8	7,8	−7,1	13 · 4
HAM L. Hamilton	13,5	+0,2	12,6	−12,6	13 · 4
ALO F. Alonso	14,1	+0,4	10,9	+10,9	13 · 4
RUS G. Russell	14,2	−2,4	13,6	−7,8	13 · 4
STR L. Stroll	16,5	−5,2	15,5	+2,5	13 · 4
TSU Y. Tsunoda	17,9	+17,9	—	—	1
LEC C. Leclerc	18,0	+1,5	19,6	+19,6	13 · 4
ANT K. Antonelli	19,5	+6,5	9,5	−4,1	13 · 4
NOR L. Norris	20,1	+4,6	19,6	+19,6	13 · 4
VER M. Verstappen	21,4	+4,0	25,9	−5,8	13 · 4

By constructor

The same metric for teams. Team error runs higher by construction: two cars’ points add up.

Asset	Error · all	Bias	Error · frozen	Bias	Rounds
WIL	10,6	+2,5	7,6	+1,8	13 · 4
ALP	12,1	+6,1	13,0	−0,4	13 · 4
AUD	13,2	+0,8	8,6	+8,6	13 · 4
HAA	15,6	−4,4	15,8	−7,0	13 · 4
RBS	17,2	+10,0	20,5	+9,0	13 · 4
CAD	18,7	−11,7	18,9	−13,5	13 · 4
RBR	18,7	+5,2	28,7	−2,8	13 · 4
FER	20,5	+3,6	11,5	+6,8	13 · 4
AST	22,4	−4,4	21,0	+17,0	13 · 4
MER	23,1	+12,1	12,2	−0,3	13 · 4
MCL	25,1	+8,6	14,4	+11,3	13 · 4

What already shows: the favourites are where we miss most. That is the content of this page, not a reason to hide it.

The whole dataset

One row per round × asset: round, code, type, name, x_pts, actual, error, forecast_recorded_at, lock_at, provenanceLicensed CC BY 4.0 · schema.org Dataset markup · updated after every round

accuracy-2026.csv .json

How it is computed: actual minus projected for every asset, the absolute value averaged per round and per asset. The headline counts only rounds whose projection sat in the database before the game deadline. The P10–P90 interval width is not published until there is a sample the model never saw during calibration. What walk-forward means, and why the headline is an error rather than a hit rate →