#	Question	My Pick	2×	Pts Available	Pts Earned
01	Predict the podium	Norris · Piastri · Leclerc	2×	90	48
02	Where does Verstappen finish in The Sprint?	4th–10th	—	10	10
03	How many teams score points in The Sprint?	6	—	15	—
04	Which driver takes pole position?	Lando Norris	—	15	15
05	Who qualifies highest (Hülkenberg vs Bortoleto)?	Gabriel Bortoleto	—	15	15
06	Fastest pitstop (team)	Racing Bulls	—	15	—
07	How many SC/VSC calls during the GP?	3	—	20	—
08	Fastest lap of the race	Oscar Piastri	—	20	—
09	How many classified finishers?	Fewer than 18	—	30	30
10	Most points, Sprint + GP (Audi vs Williams)?	Audi	—	10	10
TOTAL				240	128

Final: 128 pts · 50% accuracy (5/10 clean hits: Q2, Q4, Q5, Q9, Q10) — the season’s second-best round, behind only R10’s 163, and a return to its best accuracy.

Q1 podium: Norris · Antonelli · Russell. Norris was called at P1 and won, so the leg paid the exact-slot bonus — 12 doubled to 24, then doubled again by the multiplier for 48. Piastri (P2 call, finished P6) and Leclerc (P3 call, finished P5) both missed. Antonelli and Russell took the other two slots, and Antonelli was omitted deliberately for the second round running.

Notes

Pre-race decisions:

• 2× on the podium (Q1) for the fifth consecutive round, on ~29.3 combined EV against ~7.5 for the best binary — (the picks file called it “sixth straight”; the streak actually runs R08–R12, five rounds, since R07 used Q6).
• The R11 inconsistency failure was named and corrected explicitly. Every question was priced off three declared beliefs, with a written anti-correlation audit confirming no pick required another to fail.
• Antonelli omitted from the podium at a payout of 5 for the second straight round, on the stated grounds that this is “variance, not a broken model.”
• Q6 retained against two prior recommendations to drop it, reframed as taking a demonstrated winner rather than a value play — “about stopping the bleed, not finding an edge.”
• Q8 followed the R10 fastest-lap heuristic (“a car out of position with a free late stop, not the leader”) after R11 stated it and ignored it.
• Piastri named in advance as “the weakest link in the sheet”, carrying both the Q1 P2 leg and Q8.

Result:

• The R11 fix worked, and it is measurable. R11 forfeited 33 points to backing Norris on the podium and then pricing pole and “new winner” off contradictory beliefs. This sheet priced Q1 and Q4 off the same belief — and those two questions returned 63 of our 128. The single question R11 got wrong for structural reasons is the single question that carried this round. The anti-correlation audit is now the most valuable process in this workflow; keep it.
• Belief 3 was the cleanest read on the sheet: 2-for-2. “Verstappen is a home-race threat in an unstable car” drove Q2 (4th–10th) and his omission from the podium. He qualified P7, finished the sprint P6 — inside the called band — and then retired on lap zero of the Grand Prix. Declining to back a three-time Zandvoort winner at his home race was the least comfortable call on the sheet and the most completely vindicated.
• Belief 1 went 2-for-4, and split exactly along the line the file predicted. Both Norris legs hit — Q1 P1 (48) and Q4 pole (15). Both Piastri legs missed — the Q1 P2 leg and Q8, 38 points of availability returning zero. The file named him “the weakest link in the sheet” before the race and he was precisely that. Correctly identifying a weak link is not the same as pricing it out; the McLaren belief was sound and the driver chosen to express half of it was not.
• Belief 2 went 1-for-3, and the hit was right for the wrong mechanism. Q9 (<18 classified, 30) was the best non-multiplied call of the round — 17 classified, five retirements. But the named drivers of that thesis were “rain, one practice session, Lawson debuting in the Red Bull and Tsunoda returning with no 2026 running”, and not one of them produced a retirement: Lawson finished P7 and Tsunoda P11. The attrition came from Verstappen, Bearman, Stroll, Ocon and Bottas. The 30 points are real; the evidence they provide for the model is weaker than it looks.
• Q3’s mechanism was inverted, which is the useful thing to record. The sheet took 6 teams scoring in the sprint on the argument that “the top four teams rarely place all eight cars in the top eight” and that Lawson and Tsunoda in unfamiliar cars would open a gap. The opposite happened: the top four teams took seven of the eight sprint scoring positions, with only Gasly’s Alpine breaking through, for an answer of 5 — the market favourite the file reasoned against. A short race with no pit stops locks in a top-heavy grid rather than scrambling it. Note this before R16 Singapore, the last sprint of the season.
• ⚠ Q6 (fastest pitstop) is now 0-for-4, and this round it was taken with the market and still missed. R10 and R11 both recommended dropping it; the R12 sheet overrode that with better reasoning — take a demonstrated recent winner at the highest payout — and got the same result. Four rounds, ~80 points of availability, zero returned.
• ⚠ And we cannot post-mortem Q6 or Q7, because our results files do not record their outcomes. content/results/R12-Netherlands.md has no fastest-pitstop time and no SC/VSC count. We are paying ~35 points of availability per round on two questions whose answers we never write down, which is why four rounds of Q6 data have produced no model. Either add both fields to the results template or stop answering the questions — continuing to guess at outcomes we decline to record is the worst of the three options.
• ⚠ Q8 (fastest lap) is 0-for-4, and the answer has been the same driver twice running. Leclerc took it at Budapest (1:22.000) and at Zandvoort (1:14.230, lap 60) — and he was on our sheet as a Q1 podium leg on both occasions. The heuristic was correctly stated and correctly applied this time: Piastri finished P6 from a P4 grid, genuinely out of position with a stop in hand. It still lost to the incumbent. The heuristic is not wrong, it is under-powered — it identifies a category of driver, and Leclerc has been the answer inside that category two rounds running. Take the incumbent unless there is a specific reason not to.
• The Antonelli omission is now three rounds and three podiums old — but the EV line is not the thing to change. He won at R10, took P3 at R11 and P2 here, omitted every time at a payout of 5. The arithmetic still defends it: even at P(podium) 0.85 with a fair exact-slot chance, a payout of 5 cannot clear the 10.8 the model gave Piastri. The real pattern is elsewhere — Q1 has paid on exactly one leg in each of the last two rounds, and both times it was the leader leg, while the “value” legs the model prefers (Piastri P2, Leclerc P3 here; Leclerc P1, Hamilton P2 at R11) have returned zero from four attempts. That is the leg to re-price, not Antonelli.
• Q5 vindicated the “full season over one session” rule that R11 broke. The sheet took Bortoleto at 15 while openly acknowledging Hülkenberg had out-qualified and out-finished him at Budapest — the exact shape of evidence that produced R11’s Pérez error, resolved the other way because a season of scoring outweighed one session. Bortoleto qualified P9 to Hülkenberg’s P13, into Q3. Worth noting the race then went the other way: Hülkenberg finished P8 for Audi’s points and Bortoleto P13. The question asked about qualifying, and the discipline was correct for the question actually asked.
• Q10 paid for the second straight round on the same reasoning. Taking the market favourite because the evidence supports it — the inverse of the R11 Q4 error — returned 10 as Audi scored 4 championship points (Hülkenberg P8) to Williams’ zero, with both Williams cars outside the points in the sprint and the race.
• The 2× was right on EV and cost 6 points in hindsight. Q1 graded 29.3 against Q9’s 4.7 — a factor of six — and Q1 duly returned the round’s largest score. Placing it on Q9 instead would have produced 134 rather than 128. A 6-point miss on a call that was right by 6× is the best kind of error to make; no change to the policy.
• Season predictions total: 1,072 pts (R01: 97 · R02: 99 · R03: 87 · R04: 90 · R05: 28 · R06: 88 · R07: 56 · R08: 78 · R09: 93 · R10: 163 · R11: 65 · R12: 128). 128 is the second-best round of twelve and the largest single-round improvement of the season (+63).
• For R13 Monza: the three-belief structure and the anti-correlation audit are the parts of this sheet to keep verbatim. The open items are Q1’s non-leader legs (0-for-4), fastest lap (take Leclerc unless argued otherwise), and the two questions we answer but never record. Monza is a standard weekend — no sprint — and the Mercedes engine question that Zandvoort failed to test gets its real answer there, which should drive the beliefs rather than sit beside them.