Pool owner
Model audit
Every official pick from The Model and KATI, the evidence available when it was made, and how completed results are changing future decisions.
Audit range
Defaults to this week and the previous two. Progression always starts at Week 1.
Coverage
Week-by-week progression
Points are the pool score: one for the correct result plus one for the exact score. Expected points are the model's own E[X] = P(outcome) + P(exact score) at the seal; the z-score is how many SDs the realized total sits from that expectation.
| Week | Model | Picks | Completed | Week points | Cumulative | Expected pts | Cum expected | vs realized z | Priced @ seal | Briefing missing | Blend (mkt/free/dc) | Outcome accuracy |
|---|
Probability scorecard
Proper scores (log-loss, lower is better) per week and cumulatively. Paired comparisons use a seeded pick-week block bootstrap on common fixtures only.
| Week | Model close | KATI pure | Market close | Pinnacle close |
|---|
Cumulative + paired 95% CI
| Comparison | n | weeks | mean ΔLL | 95% CI | Leader |
|---|
KATI independence
Agreement is measured, not forced. A different score with the same winner is still shown separately from a genuinely different H/D/A call. The radar ledger scores only fixtures where her upset radar fired.
| Week | Compared | Same score | Different outcome | KATI right (diff) | Model right (diff) | Radar fired | Dog won | Dog unbeaten | Scoreline overridden | Radar no market |
|---|
KATI tip vs official pick
The Wednesday tip (the numbers-only stream members can copy) scored as a shadow entrant against the official Thursday pick. The tip is never in pool standings.
| Week | Tip picks | Official picks | Same outcome | Same score | Tip points | Official points | Tip right (diff) | Official right (diff) |
|---|
KATI followers
Members whose card came from the tip (copied it, took it as the deadline default, or matched it on 70%+ of fixtures) vs everyone else, mean pool points per week.
| Week | Followers | Others | Follower mean pts | Other mean pts | Difference |
|---|
Model trajectory
The adaptive-ensemble posterior over time (its effective market/DC/Elo weights and the settled/priced fixtures behind it), plus the biggest week-over-week rating movers.
| Week | As of | Eff weights (mkt/dc/elo) | Settled (priced) | HFA | mu | Top movers (Δ Elo) |
|---|
Calibration
The Model's stated pick-outcome probability vs how often that outcome actually happened. Well-calibrated buckets track the diagonal.
| P(outcome) bucket | Picks | Hit rate |
|---|