Race retro — Brazilian Grand Prix (2026 R20)
Winner: norris
Prospective paired scores (confirmation pending)
2026 R20
| market | pl_ll | sim_ll | pl_brier | sim_brier | route_advantage_ll |
| winner | 1.32896 | 0.95789 | 0.63640 | 0.48817 | -0.37106 |
| podium | 0.30517 | 0.36676 | 0.10554 | 0.12154 | -0.06159 |
| top5 | 0.24376 | 0.27967 | 0.08033 | 0.09131 | -0.03591 |
| top10 | 0.17486 | 0.27107 | 0.04228 | 0.06574 | -0.09620 |
| points | 0.17486 | 0.27107 | 0.04228 | 0.06574 | -0.09620 |
Team gap shadow (descriptive; no adoption)
winner: missing pre-race archive; no backfill.
podium: missing pre-race archive; no backfill.
top5: missing pre-race archive; no backfill.
top10: missing pre-race archive; no backfill.
Checkpoint scores (signals archive)
| checkpoint | snapshot | mode | winner p (model) | winner p (market) | podium Brier | rank corr |
| post-practice | 2026-11-07T17:50 | PRE-QUALI | 9.3% | 18.9% | 0.1129 | 0.82 |
| post-quali | 2026-11-08T13:47 | POST-QUALI | 26.5% | 28.6% | 0.1205 | 0.86 |
Model vs outcome
Probability routing: market_v1.
Prediction snapshot: POST-QUALI, generated 2026-11-08T13:47:20.608967+00:00
Winner NLL: 1.329 — model gave norris p_win 26.5% (gate-of-record avg 1.202 — worse than average)
Top-10 log loss: 0.278
Biggest misses
bearman: p_points 45%, finished P17
piastri: p_points 55%, finished P6
arvid_lindblad: p_points 38%, finished P11
Per-driver table
| driver | p_win | p_points | finish |
| norris | 26.5% | 87.3% | P1 |
| max_verstappen | 5.3% | 64.1% | P2 |
| hadjar | 3.7% | 70.3% | P3 |
| russell | 14.1% | 85.8% | P4 |
| antonelli | 21.9% | 85.7% | P5 |
| piastri | 1.6% | 55.3% | P6 |
| lawson | 2.2% | 74.4% | P7 |
| leclerc | 13.8% | 82.9% | P8 |
| hamilton | 6.0% | 73.8% | P9 |
| hulkenberg | 1.1% | 63.6% | P10 |
| arvid_lindblad | 0.8% | 37.7% | P11 |
| alonso | 0.2% | 25.1% | P12 |
| colapinto | 0.3% | 21.7% | P13 |
| gasly | 0.4% | 23.3% | P14 |
| albon | 0.3% | 28.7% | P15 |
| bortoleto | 0.3% | 12.6% | P16 |
| bearman | 0.6% | 45.1% | P17 |
| perez | 0.0% | 4.1% | P18 |
| sainz | 0.2% | 17.1% | P19 |
| rafael_camara | 0.4% | 29.1% | P20 |
| stroll | 0.1% | 8.1% | DNF |
| bottas | 0.0% | 4.1% | DNF |
Market signals at last scan
31 divergence(s) were live (watchlist rows marked):
maxverstappen top5 NO edge +40.1% (model 77% vs ask 35¢) (watchlist)_
max_verstappen podium NO edge +32.8% (model 93% vs ask 58¢)
piastri top5 NO edge +27.7% (model 87% vs ask 57¢) (watchlist)
leclerc podium YES edge +23.6% (model 45% vs ask 19¢)
piastri points NO edge +22.7% (model 45% vs ask 20¢)
max_verstappen points NO edge +20.8% (model 36% vs ask 14¢)
antonelli top5 NO edge +17.3% (model 26% vs ask 8¢) (watchlist)
hamilton top5 NO edge +17.3% (model 65% vs ask 46¢) (watchlist)
lawson top5 YES edge +17.1% (model 37% vs ask 18¢) (watchlist)
russell podium YES edge +16.9% (model 60% vs ask 41¢)
antonelli podium NO edge +16.3% (model 44% vs ask 26¢)
antonelli winner NO edge +15.1% (model 78% vs ask 61¢)
… 19 more
Recorded paper plays
10 recorded signal(s), 5 settled, realized P&L $-18.11
In-week adjustments — directional review
Practice form (fp_form): Spearman(form, finish) = +0.45 over 14 drivers (Practice 2+Practice 1) — directionally RIGHT (positive = faster practice pace finished further up)
Rain: forecast 76% at last prediction vs realized 4% of race weather samples wet
Trackfit: active this round — review the fit-weight contribution against the result (no automated check yet)
Follow-ups
Anything above marked WRONG or a big miss: file it in STATE.md open items before the next race week.
Manual Kalshi plays are not captured here — record them in the ledger or the retro stays blind to them.