How accurate are our rankings? 2026

Every published weekly board, scored against what happened the Tuesday after — the engine's number, the baseline it starts from and the betting market on the same players. Nothing on this page is typed by hand; every row links to the board version it scores. How we score ourselves · The 2026 season report

Season to date · PPR positional boards · forecast weeks only

QB · MAE6.502 weeks · ρ 0.43 · top-12 17/24 · coverage 60%

RB · MAE6.792 weeks · ρ 0.49 · top-12 12/24 · coverage 63%

WR · MAE7.832 weeks · ρ 0.15 · top-12 11/24 · coverage 53%

TE · MAE5.762 weeks · ρ 0.27 · top-12 12/24 · coverage 58%

K · MAE5.082 weeks · ρ 0.27 · top-12 12/24 · coverage 50%

DST · MAE3.412 weeks · ρ 0.27 · top-12 11/24 · coverage 65%

Model · the pinned fit and the Tuesday canary

Pinned fitens-fit-2026w05-mktfit_id fit-2026w05-mkt · fitted Tue, Oct 13, 5:31 AM ET

Trained on2025 W5–2026 W5out-of-time: nothing after Week 5

Sources blended7baseline · season_eb · usage · opp_adj · market · nerds · analyst

Next candidateTue, Oct 20pinned without a comparison until the canary can decide (from Oct 27)

Canary · every Tuesday's candidate vs the pin, scored out of time on the last 4 scored weeks

MAE in fantasy points per positional board (PPR, the accuracy page's rows), candidate / pin · composite = the mean over QB/RB/WR/TE of candidate ÷ pin − 1 (all rows pooled shown as candidate / pin) · pinned only if the composite is within +2% and no position is worse by more than 4%
Tuesday	Candidate fit	Scored on	QB	RB	WR	TE	Composite	Decision
Tue, Oct 13	ens-fit-2026w05-mkt	W5	7.7 / 7.7	6.8 / 6.8	8.3 / 8.3	6.4 / 6.4	7.37 / 7.37	Pinned · no comparison yet 1 scored forecast week(s) of the 4 needed — the pin stays (have: W5) — pinned anyway (CANARY_INSUFFICIENT=pin: nothing to compare on yet, so the weekly refit stands as before item 8)
Tue, Oct 6	ens-fit-2026w04-mkt	—	— / —	— / —	— / —	— / —	— / —	Pinned · no comparison yet 0 scored forecast week(s) of the 4 needed — the pin stays — pinned anyway (CANARY_INSUFFICIENT=pin: nothing to compare on yet, so the weekly refit stands as before item 8)
Tue, Sep 29	ens-fit-2026w03-mkt-20260929	—	— / —	— / —	— / —	— / —	— / —	Pinned · no comparison yet 0 scored forecast week(s) of the 4 needed — the pin stays — pinned anyway (CANARY_INSUFFICIENT=pin: nothing to compare on yet, so the weekly refit stands as before item 8)

A rejected candidate is rung 3 of the outage ladder: the boards are built from the pinned fit under a dated banner that links here. Both scores and both version ids are stored on every row.

Week by week

Week by week · PPR · MAE per board (back-filled weeks labelled)
Week	QB	RB	WR	TE	K	DST	Flex	Overall
Week 2 back-filled	7.21	4.07	7.87	5.44	4.38	4.67	6.38	7.52
Week 3 back-filled	5.08	5.49	8.04	5.92	5.56	4.62	6.92	6.56
Week 4	5.29	6.88	7.50	5.18	5.44	2.71	7.66	6.20
Week 5	7.70	6.69	8.15	6.35	4.71	4.12	8.04	7.51

What the chat said

Medians shownNo chat numbers scored yetEvery median the chat shows is written to the forecast ledger and scored the Tuesday after.

Start/sit callsNo calls yetBrier score of P(A > B) against who actually outscored whom; skill = 1 − Brier / 0.25 (a coin flip scores 0).

How this is scoredOnly answers whose players both played are scored; locked players are excluded from calls. The gate keeps every number the chat says inside the engine's payload, so this strip measures the engine, not the model.

Definitions

tilesPPR positional boards, retrospective weeks excluded; MAE and coverage weighted by scored rows, Spearman averaged over weeks

back-filledPublished after the week's first kickoff (a seed run later from pre-week data). Shown, labelled, never counted.