Every artifact the machine produces, what it is, and where to drill in. If it isn't listed here, the machine doesn't make it.
A one-off study — its numbers below are exactly what the run reported.
| system doctor · at 2026-07-15T19:16:51.509679+00:00 · failed 3 | 7/15/2026 |
| system doctor · at 2026-07-15T13:17:29.702661+00:00 · failed 2 | 7/15/2026 |
| system doctor · at 2026-07-15T07:21:06.226699+00:00 · failed 0 | 7/15/2026 |
| system doctor · at 2026-07-15T01:17:07.031766+00:00 · failed 0 | 7/15/2026 |
| system doctor · at 2026-07-14T19:16:16.167089+00:00 · failed 0 | 7/14/2026 |
| system doctor · at 2026-07-14T13:16:03.607237+00:00 · failed 1 | 7/14/2026 |
| system doctor · at 2026-07-14T01:15:50.109241+00:00 · failed 5 | 7/14/2026 |
| system doctor · at 2026-07-13T19:15:17.890488+00:00 · failed 3 | 7/13/2026 |
| system doctor · at 2026-07-13T13:15:25.957701+00:00 · failed 1 | 7/13/2026 |
| system doctor · at 2026-07-13T07:15:21.750018+00:00 · failed 1 | 7/13/2026 |
| system doctor · at 2026-07-13T01:15:28.648917+00:00 · failed 1 | 7/13/2026 |
| system doctor · at 2026-07-12T19:15:16.844294+00:00 · failed 0 | 7/12/2026 |
| system doctor · at 2026-07-12T18:55:24.963661+00:00 · failed 0 | 7/12/2026 |
| system doctor · at 2026-07-12T13:15:22.443210+00:00 · failed 0 | 7/12/2026 |
| system doctor · at 2026-07-12T08:27:40.355843+00:00 · failed 0 | 7/12/2026 |
| system doctor · at 2026-07-12T07:15:39.157849+00:00 · failed 0 | 7/12/2026 |
| system doctor · at 2026-07-12T01:15:23.908217+00:00 · failed 0 | 7/12/2026 |
| system doctor · at 2026-07-11T19:15:23.270179+00:00 · failed 0 | 7/11/2026 |
| system doctor · at 2026-07-11T16:39:40.826681+00:00 · failed 0 | 7/11/2026 |
A one-off study — its numbers below are exactly what the run reported.
Ranks every backtested model by skill against the bookmaker's closing prices — the fairest single scoreboard we have.
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/14/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/12/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/12/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/11/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/11/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/10/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/10/2026 |
| strategy & model leaderboard · 24 backtests ranked by skill vs the market | 7/7/2026 |
| strategy & model leaderboard · 5 backtests ranked by skill vs the market | 7/3/2026 |
| strategy & model leaderboard · 5 backtests ranked by skill vs the market | 7/3/2026 |
A one-off study — its numbers below are exactly what the run reported.
| observatory: predictability, biases, signal decay · model run 18 | 7/14/2026 |
| observatory: predictability, biases, signal decay · model run 18 | 7/12/2026 |
| observatory: predictability, biases, signal decay · model run 18 | 7/11/2026 |
| observatory: predictability, biases, signal decay · model run 18 | 7/11/2026 |
| observatory: predictability, biases, signal decay · model run 18 | 7/10/2026 |
| observatory: predictability, biases, signal decay · model run 18 | 7/7/2026 |
| observatory: predictability, biases, signal decay · model run 18 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| gap-filter fit for live executor · n train 13,698 · base rate model right 0.4915 | 7/14/2026 |
| gap-filter fit for live executor · n train 13,698 · base rate model right 0.4915 | 7/12/2026 |
| gap-filter fit for live executor · n train 13,698 · base rate model right 0.4915 | 7/11/2026 |
| gap-filter fit for live executor · n train 13,698 · base rate model right 0.4915 | 7/11/2026 |
| gap-filter fit for live executor · n train 13,698 · base rate model right 0.4915 | 7/6/2026 |
| gap-filter fit for live executor · n train 14,202 · base rate model right 0.4868 | 7/6/2026 |
| gap-filter fit for live executor · n train 13,698 · base rate model right 0.4915 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| tennis lab: normal + ratio + crazy rules at real closing prices · matches replayed 14,400 | 7/11/2026 |
| tennis lab: normal + ratio + crazy rules at real closing prices · matches replayed 14,400 | 7/11/2026 |
A one-off study — its numbers below are exactly what the run reported.
| revalidate roster: consensus price + train/test holdout · n live 30 · n scored 5 · n skipped 25 | 7/9/2026 |
| revalidate roster: consensus price + train/test holdout · n live 33 · n scored 7 · n skipped 26 | 7/9/2026 |
| revalidate roster: consensus price + train/test holdout · n live 33 · n scored 10 · n skipped 23 | 7/9/2026 |
A one-off study — its numbers below are exactly what the run reported.
| subset check: did the mid-dog model's picks have OOS skill? · min edge 0.08 | 7/9/2026 |
A one-off study — its numbers below are exactly what the run reported.
| forward paper scoreboard · | 7/9/2026 |
| forward paper scoreboard · | 7/9/2026 |
| forward paper scoreboard · | 7/9/2026 |
A one-off study — its numbers below are exactly what the run reported.
| edge parlay: does a real edge compound up across legs? · price agg max · single roi 4.63 | 7/9/2026 |
| edge parlay: does a real edge compound up across legs? · price agg avg · single roi 3.02 | 7/9/2026 |
A one-off study — its numbers below are exactly what the run reported.
| edge doubles: does a real edge compound like the vig does? · | 7/9/2026 |
| edge doubles: does a real edge compound like the vig does? · expected double roi if independent -1.6 | 7/9/2026 |
A one-off study — its numbers below are exactly what the run reported.
| matrix holdout: full rule grid, train/test, all leagues · n survivors 6 · predictions 219,577 · cells scanned 167 | 7/9/2026 |
A one-off study — its numbers below are exactly what the run reported.
| epl mid-dog v2: filter stack vs v1, holdout + generalization · | 7/7/2026 |
A one-off study — its numbers below are exactly what the run reported.
| proven-logic retro: derivative classes + simple parlays at real prices · hot ref threshold 3.6 | 7/7/2026 |
| proven-logic retro: derivative classes + simple parlays at real prices · hot ref threshold 3.6 | 7/7/2026 |
A one-off study — its numbers below are exactly what the run reported.
| season watch: weekly Bayesian re-evaluation, all league-seasons · league seasons 146 | 7/7/2026 |
The tournament: every prediction engine × every betting rule × every staking policy, replayed season by season with each season restarting from $1000. Finds which COMBINATIONS survive, not just which models predict.
| matrix backtest: engine x rule x staking, per-season $1000 replays · 23,640 engine × rule × staking combos, each season from $1000 | 7/7/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 55,349 engine × rule × staking combos, each season from $1000 | 7/7/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 10,797 engine × rule × staking combos, each season from $1000 | 7/6/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 10,122 engine × rule × staking combos, each season from $1000 | 7/6/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 7,392 engine × rule × staking combos, each season from $1000 | 7/5/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 3,110 engine × rule × staking combos, each season from $1000 | 7/4/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 933 engine × rule × staking combos, each season from $1000 | 7/4/2026 |
| matrix backtest: engine x rule x staking, per-season $1000 replays · 39 engine × rule × staking combos, each season from $1000 | 7/4/2026 |
A one-off study — its numbers below are exactly what the run reported.
| day mode: $100 -> $1000 in one day, structure tournament · target turn $100 into $1000 same day · days replayed 2,070 · avg slate size 14 | 7/7/2026 |
| day mode: $100 -> $1000 in one day, structure tournament · target turn $100 into $1000 same day · days replayed 2,070 · avg slate size 14 | 7/7/2026 |
A one-off study — its numbers below are exactly what the run reported.
| season mode: futures bets at season start, every league-season · roi 1.3644 · bets 793 · league seasons 144 | 7/7/2026 |
| season mode: futures bets at season start, every league-season · roi 1.3117 · bets 795 · league seasons 144 | 7/7/2026 |
A one-off study — its numbers below are exactly what the run reported.
| bayes replay: in-play policies on synthesized live markets · model run 18 · in play vig 0.06 · matches replayed 3,800 | 7/7/2026 |
| bayes replay: in-play policies on synthesized live markets · model run 18 · in play vig 0.06 · matches replayed 3,454 | 7/7/2026 |
| bayes replay: in-play policies on synthesized live markets · model run 18 · in play vig 0.06 · matches replayed 3,454 | 7/7/2026 |
A one-off study — its numbers below are exactly what the run reported.
| dutch-cover backtest: covered vs straight, same picks · model run 18 · picks considered 5,068 | 7/6/2026 |
Side-vs-side questions: does the team better at X actually get more Y? Tested for every factor × every countable outcome, conditioned on how big the X advantage was.
| relational: does the better-X side get more Y? (football) · 220 side-vs-side questions tested | 7/6/2026 |
| relational: does the better-X side get more Y? (football) · 212 side-vs-side questions tested | 7/6/2026 |
| relational: does the better-X side get more Y? (football) · 168 side-vs-side questions tested | 7/5/2026 |
| relational: does the better-X side get more Y? (football) · 148 side-vs-side questions tested | 7/4/2026 |
Dissects every disagreement between our models and the market: who was right in which situations, whether two engines agreeing means anything, and whether a classifier can predict when OUR side of the gap wins.
| gap intelligence: model #20 vs market (consensus with #17) · 21,280 model-vs-market gaps dissected · survival AUC 0.6778 | 7/6/2026 |
| gap intelligence: model #19 vs market (consensus with #17) · 21,280 model-vs-market gaps dissected · survival AUC 0.6787 | 7/6/2026 |
| gap intelligence: model #18 vs market (consensus with #17) · 21,280 model-vs-market gaps dissected · survival AUC 0.6664 | 7/5/2026 |
| gap intelligence: model #16 vs market (consensus with #17) · 21,280 model-vs-market gaps dissected · survival AUC 0.6895 | 7/4/2026 |
A one-off study — its numbers below are exactly what the run reported.
| scorer model v1: P(player scores) · AUC 0.762 (real signal) | 7/6/2026 |
| scorer model v1: P(player scores) · AUC 0.737 (real signal) | 7/4/2026 |
| scorer model v1: P(player scores) · AUC 0.737 (real signal) | 7/4/2026 |
A one-off study — its numbers below are exactly what the run reported.
| book profiling: who is generous, who is asleep · days 30 · n quotes 4,744 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| ref-line scan · n refs known 400 · tercile threshold 3.8 | 7/6/2026 |
| ref-line scan · n refs known 345 · tercile threshold 3.61 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| regime alerts 2026-07-06 · n games 42 · n alerts 1 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| goal hazard curves: minute profile of scoring · n games 5,521 · n goals 17,000 · share after 75 0.243 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| WC 2026 full-tournament retro at real prices · games 92 · games with 1x2 price 77 · games with soft prices 24 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| soft-market retro from 2021-07-01: cards/corners gates at synthetic prices · from 2021-07-01 | 7/6/2026 |
A one-off study — its numbers below are exactly what the run reported.
| outcome matters: motivation asymmetry + stage behavior · n with closing odds 5,149 · n late season matches 10,808 | 7/5/2026 |
Asks which pre-match factors drive one specific outcome, using a walk-forward model and permutation importance.
| sweep: match.red_card_shown (bool) · AUC 0.609 (real signal) | 7/5/2026 |
| sweep: match.red_card_shown (bool) · AUC 0.609 (real signal) | 7/5/2026 |
| sweep: match.late_goal (bool) · AUC 0.384 (no real pattern — honest result) | 7/5/2026 |
| sweep: match.late_goal (bool) · AUC 0.384 (no real pattern — honest result) | 7/5/2026 |
| sweep: match.second_half_goals >= 1 · AUC 0.523 (no real pattern — honest result) | 7/5/2026 |
| sweep: match.second_half_goals >= 1 · AUC 0.523 (no real pattern — honest result) | 7/5/2026 |
| sweep: match.comeback_happened (bool) · AUC 0.494 (no real pattern — honest result) | 7/5/2026 |
| sweep: match.comeback_happened (bool) · AUC 0.494 (no real pattern — honest result) | 7/5/2026 |
| sweep: match.total_fouls >= 23 · AUC 0.600 (real signal) | 7/5/2026 |
| sweep: match.total_fouls >= 23 · AUC 0.600 (real signal) | 7/5/2026 |
| sweep: match.total_cards >= 3 · AUC 0.561 (weak signal) | 7/4/2026 |
| sweep: match.total_cards >= 3 · AUC 0.561 (weak signal) |
A one-off study — its numbers below are exactly what the run reported.
| market misses: when is the market wrong, and what changed? · sport football · n matches 26,088 · upset rate p lt 20 0.0501 | 7/4/2026 |
| market misses: when is the market wrong, and what changed? · sport football · n matches 26,088 · upset rate p lt 20 0.0501 | 7/4/2026 |
Compares opening and closing odds over 20+ years: when a price shortens (steam), is that information already fully absorbed by kickoff, or does following the move still pay?
| line movement (clean open/close discipline) · 48,132 selections · does steam beat the close? | 7/4/2026 |
| line movement study: is steam fully priced? (5 leagues, 20yr) · 78,279 selections · does steam beat the close? | 7/3/2026 |
A one-off study — its numbers below are exactly what the run reported.
| portfolio policy replay (predictions of backtest #16) · n bets 5,959 · hit rate 0.3536 · source run 16 | 7/4/2026 |
| portfolio policy replay (predictions of backtest #11) · n bets 2,333 · hit rate 0.3918 · source run 11 | 7/4/2026 |
A one-off study — its numbers below are exactly what the run reported.
| lead-lag: which trends move first (football) · sport football · n metrics 31 · n team contests 104,910 | 7/4/2026 |
Theory-free sweep: correlates every pre-match factor with every outcome, corrects for the thousands of tests, and keeps only effects stable in every era.
| interaction miner v3: AH line x H2H x stakes x everything · 65044 pairs tested · 31815 survive every era | 7/4/2026 |
| interaction miner v2: HT momentum x market x events · 58128 pairs tested · 28782 survive every era | 7/3/2026 |
| discovery v3: half-time momentum + event targets · 1092 pairs tested · 550 survive every era | 7/3/2026 |
| discovery v3: interaction mining · 41400 pairs tested · 21958 survive every era | 7/3/2026 |
| discovery v2: full feature factory · 590 pairs tested · 327 survive every era | 7/3/2026 |
| discovery sweep (football) · 240 pairs tested · 124 survive every era | 7/3/2026 |
Measures the same abstract effect (fatigue, momentum, rest) independently in each sport. Patterns that replicate across unrelated sports are laws, not quirks.
| universal v3: FOUR sports · 5 concepts tested across sports | 7/3/2026 |
| universal v2: three sports · 5 concepts tested across sports | 7/3/2026 |
| universal: cross-sport pattern replication · 5 concepts tested across sports | 7/3/2026 |
Tries thousands of betting rules on old seasons, then judges the winners ONLY on years they never saw. Survivors are candidate strategies; everything else was luck.
| sweep of market-aware engine (v2) · 176 rules tested · 13 survived the unseen years | 7/3/2026 |
Monte Carlo runs — every one stores its full recipe and is bit-for-bit reproducible.
History replays at real odds — leak-free, graded against the closing line.
Each bets its own $1000 paper bankroll daily. Scoreboard →
Match winner settles on 90 minutes; “to advance” includes extra time and penalties — different questions, different markets. Player props settle from Kalshi's own results.
| 7/9/2026 |
| 7/4/2026 |
| sweep: match.btts (bool) · AUC 0.511 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.btts (bool) · AUC 0.511 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_corners >= 10 · AUC 0.515 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_corners >= 10 · AUC 0.515 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_goals >= 2 · AUC 0.534 (weak signal) | 7/4/2026 |
| sweep: match.total_goals >= 2 · AUC 0.534 (weak signal) | 7/4/2026 |
| sweep: match.result D · AUC 0.527 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.rainfall_mm >= 0 · n 9,880 | 7/4/2026 |
| sweep: match.result D · AUC 0.527 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.rainfall_mm >= 0 · n 9,880 | 7/4/2026 |
| sweep: match.rained (bool) · AUC 0.499 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.rained (bool) · AUC 0.499 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.humidity_pct >= 73 · AUC 0.658 (real signal) | 7/4/2026 |
| sweep: match.humidity_pct >= 73 · AUC 0.658 (real signal) | 7/4/2026 |
| sweep: match.temp_c >= 10.7 · AUC 0.783 (real signal) | 7/4/2026 |
| sweep: match.temp_c >= 10.7 · AUC 0.785 (real signal) | 7/4/2026 |
| goals w/ granular · AUC 0.534 (weak signal) | 7/4/2026 |
| corners w/ granular · AUC 0.524 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.red_card_shown (bool) · AUC 0.646 (real signal) | 7/4/2026 |
| sweep: match.late_goal (bool) · AUC 0.501 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.second_half_goals >= 1 · AUC 0.518 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.comeback_happened (bool) · AUC 0.505 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_fouls >= 23 · AUC 0.600 (real signal) | 7/4/2026 |
| sweep: match.total_cards >= 3 · AUC 0.563 (weak signal) | 7/4/2026 |
| sweep: match.btts (bool) · AUC 0.512 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_corners >= 10 · AUC 0.518 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_goals >= 2 · AUC 0.532 (weak signal) | 7/4/2026 |
| sweep: match.result D · AUC 0.530 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.red_card_shown (bool) · AUC 0.385 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.late_goal (bool) · AUC 0.484 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.second_half_goals >= 1 · AUC 0.518 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.comeback_happened (bool) · AUC 0.509 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_fouls >= 23 · AUC 0.600 (real signal) | 7/4/2026 |
| sweep: match.total_cards >= 3 · AUC 0.565 (weak signal) | 7/4/2026 |
| sweep: match.btts (bool) · AUC 0.518 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_corners >= 10 · AUC 0.524 (no real pattern — honest result) | 7/4/2026 |
| sweep: match.total_goals >= 2 · AUC 0.543 (weak signal) | 7/4/2026 |
| sweep: match.result D · AUC 0.531 (weak signal) | 7/4/2026 |
| sweep smoke: late_goal (bool) · AUC 0.484 (no real pattern — honest result) | 7/4/2026 |
| what drives 4+ cards? (refs + closeness) · AUC 0.547 (weak signal) | 7/3/2026 |
| home wins v3: can we beat the market's own feature? · AUC 0.688 (real signal) | 7/3/2026 |
| 3+ goals v2 · AUC 0.524 (no real pattern — honest result) | 7/3/2026 |
| draws v2 (mismatch theory) · AUC 0.526 (no real pattern — honest result) | 7/3/2026 |
| 10+ corners v2 · AUC 0.505 (no real pattern — honest result) | 7/3/2026 |
| home wins v2 (feature factory) · AUC 0.677 (real signal) | 7/3/2026 |
| what drives 3+ goals? · AUC 0.505 (no real pattern — honest result) | 7/3/2026 |
| what drives home wins? · AUC 0.673 (real signal) | 7/3/2026 |
| what drives 10+ corners? · AUC 0.505 (no real pattern — honest result) | 7/3/2026 |