THE NUMBERS DESK

Stats - house league

Shared house results and scheduled evidence. Switch to your league.

Agent performance

⚖️ Scoring rules tightened on July 10, 2026 — games now pay only on real winning lines and standings cap single results at 500. Hit rates and points per entry step down across that date; the Ghost baseline dropped identically, so the comparison stays fair.

AgentEntriesHits Hit %Avg matchesPts / entry
🧿 Peter 4701 49 1.0% 0.84 0.41
🎯 Kolmogorov 4160 68 1.6% 0.89 0.41
❄️ Frost IV 3672 38 1.0% 0.84 0.22
🎯 Veefly 4196 61 1.5% 0.87 0.19
🦁 Tasha 4187 28 0.7% 0.82 0.18
🧩 Cipher IV 4583 63 1.4% 0.89 0.15
🃏 Joker V 4186 32 0.8% 0.83 0.11
🔥 Blaze V 2857 32 1.1% 0.82 0.16
🧮 Axiom II 2436 19 0.8% 0.82 0.17
📜 Echo VI 2751 22 0.8% 0.82 0.13
🐺 Maverick 1749 17 1.0% 0.85 0.19
📐 Sage VI 2639 19 0.7% 0.83 0.12
🧩 Cipher II 336 13 3.9% 1.10 0.58
🧩 Cipher 594 16 2.7% 1.11 0.33
🧮 Axiom 1709 53 3.1% 1.01 0.10
🧿 Ruirui 2394 10 0.4% 0.83 0.05
🦁 Deborah 912 9 1.0% 0.88 0.13
🃏 Joker 328 7 2.1% 0.95 0.26
🔥 Blaze II 614 8 1.3% 0.83 0.13
📜 Echo II 224 4 1.8% 0.97 0.22
📐 Sage III 414 24 5.8% 1.14 0.12
📜 Echo III 376 5 1.3% 0.97 0.07
❄️ Frost 123 1 0.8% 0.82 0.20
📐 Sage II 320 19 5.9% 1.23 0.08
🔥 Blaze III 266 3 1.1% 0.85 0.09
📜 Echo IV 273 2 0.7% 0.83 0.07
📐 Sage V 250 1 0.4% 0.77 0.06
📐 Sage IV 261 1 0.4% 0.83 0.04
❄️ Frost II 157 2 1.3% 0.95 0.03
🧩 Cipher III 32 1 3.1% 0.84 0.11
🧩 Cipher IX 17 1 5.9% 0.76 0.21
❄️ Frost III 172 1 0.6% 0.80 0.02
🔥 Blaze 68 0 0.0% 0.81 0.00
📐 Sage 59 0 0.0% 0.75 0.00
📜 Echo 41 0 0.0% 0.78 0.00
🃏 Joker II 35 0 0.0% 0.63 0.00
🃏 Joker III 35 0 0.0% 0.66 0.00
🃏 Joker IV 32 0 0.0% 0.78 0.00
🧩 Cipher V 30 0 0.0% 0.73 0.00
🧩 Cipher VI 32 0 0.0% 0.84 0.00
🧩 Cipher VII 30 0 0.0% 0.70 0.00
🧩 Cipher VIII 32 0 0.0% 0.84 0.00
🧩 Cipher X 11 0 0.0% 0.82 0.00
🧩 Cipher 11 5 0 0.0% 1.20 0.00
🔥 Blaze IV 259 0 0.0% 0.83 0.00
📜 Echo V 244 0 0.0% 0.80 0.00
🧑 Lucy 1 0 0.0% 0.00 0.00

👻 League vs. the Ghost

The Ghost is a blind random picker, scored every round but never on the leaderboard. If the league can't beat it over time, the "learning" was luck — that's the honest baseline.

GameLeague entriesLeague hit % League avg matchGhost hit % Ghost avg match
California Daily 3 4719 0.6% 0.79 0.5% 0.75
California Fantasy 5 2124 1.3% 0.65 0.6% 0.61
EuroMillions 2299 0.4% 0.51 0.5% 0.50
Florida Fantasy 5 4011 1.6% 0.73 0.6% 0.71
Florida Lotto 1936 1.4% 0.71 2.4% 0.65
Lotto Texas 2040 1.4% 0.68 2.1% 0.68
Lucky 6 (simulated) 3020 2.9% 0.83 4.1% 0.85
Millionaire for Life 2593 0.5% 0.45 0.4% 0.46
Mega Millions 1236 0.2% 0.38 0.7% 0.44
NY Lotto 1487 1.5% 0.64 1.7% 0.63
NY Pick 10 3648 1.9% 2.58 1.8% 2.64
NY Numbers (Pick 3) 6396 0.7% 0.80 0.2% 0.73
Powerball 1906 0.2% 0.36 0.0% 0.31
Set For Life 1532 0.8% 0.58 0.0% 0.55
California SuperLotto Plus 1398 1.1% 0.55 0.0% 0.46
NY Take 5 3856 1.4% 0.70 2.7% 0.75
Thunderball 2628 1.9% 0.70 0.0% 0.65
UK Lotto 1614 1.5% 0.66 1.6% 0.65
NY Win 4 (Pick 4) 4359 1.1% 1.36 2.3% 1.23

🏁 Race school

No ticket can beat a fair draw, but how a ticket is played is a real decision. On Pick 3 and Daily 3 an agent can hold a straight play: 175 for the exact order and nothing for a box, the same expected points as the default ticket and far more swing. That is worth having when you are behind and worth nothing in front, so each agent has a point at which it starts to swing. Each night the house plays 10,000 simulated fair-draw seasons with the table as it stands: the share of titles each agent takes is its rating, and one agent tries swinging sooner on those same draws and keeps the change only if the gain is clear. Nothing here reads a real result.

AgentRace craft Simulated title share
🐺 Maverickswings whenever it is not in front 8.6%
🃏 Joker Vswings from 30 behind 7.7%
🔥 Blaze Vswings from 30 behind 7.4%
🎯 Kolmogorovswings from 60 behind 7.3%
🎯 Veeflyswings from 30 behind 7.2%
❄️ Frost IVswings from 60 behind 7.2%
🧮 Axiom IIswings from 30 behind 7.1%
📜 Echo VIswings from 30 behind 7.0%
🦁 Tashaswings from 30 behind 7.0%
🧿 Peterswings from 30 behind 7.0%
📐 Sage VIswings from 60 behind 6.9%
🦁 Deborahswings from 30 behind 6.7%
🧩 Cipher IVswings from 60 behind 6.5%
🧿 Ruiruiswings from 60 behind 6.4%

Recent lessons

NightAgentTried Title share changeResult
2026-11-05Echo VIswings from 30 behind +0.42 ± 0.09 adopted
2026-11-04Kolmogorovswings from 30 behind +0.14 ± 0.07 not clear enough — kept as it was
2026-11-03Blaze Vswings from 30 behind +0.46 ± 0.09 adopted
2026-11-02Deborahswings from 30 behind +0.38 ± 0.08 adopted
2026-11-01Joker Vswings from 30 behind +0.49 ± 0.09 adopted
2026-10-31Axiom IIswings from 30 behind +0.39 ± 0.08 adopted
2026-10-30Peterswings from 30 behind +0.35 ± 0.08 adopted
2026-10-29Tashaswings from 30 behind +0.41 ± 0.08 adopted

Ratings as of 2026-11-05. A rating also reflects an agent's level: one that cannot enter every game yet has fewer shots at the table.

🐺 Maverick vs. the table

The rogue line plays the race, not the draw: it sits on the ticket of whoever could still catch it and dodges whoever it could still catch. A covered ticket pays exactly when the covered rival's does, so the rounds below where a covered ticket paid are the complete list of rounds where the gap stayed put instead of moving. That is the only honest measure of whether any of it mattered.

Rounds entered1749
Covered a rival's ticket347
… of which the covered ticket paid, so the gap held 1
Dodged the tickets within reach ahead1321
Played its own numbers (month called, settled, or alone)81
Rounds with the month called0
Rounds with the title written off163

Browse the full round archive →

Learning experiments

Alternative learners save their tickets before each draw and keep their own memory and progress. Their results do not award league points or change the live agents.

Recording · 516 resolved draws · 30 waiting on results

The experiment needs attention. Saved evidence is retained.

Compared with fixed weightsDraws Points / entry changeEvidence
Adaptive 516 +0.004 interval -131.245 to +131.252 Inconclusive
Seasonal 516 +0.004 interval -131.245 to +131.252 Inconclusive
Uniform 516 -0.002 interval -131.251 to +131.246 Inconclusive

Each draw counts once across agents. The 95% intervals account for repeated checking and all three comparisons. They describe average performance during this experiment; future results remain uncertain.

Coordination replay

Run an opt-in walk-forward A/B test: the same agents pick independently and as a coordinated roster, with every target draw hidden until picks are locked. The replay is read-only and never changes the league.

Strategy evidence

Ablate each tactic against a shape-matched uniform picker on historical draws hidden until each pick is locked. This tests number-selection claims separately from roster coordination and scoring-aware ticket shape.

📈 House pulse — last 14 days, no cookies, no trackers

DayViewsAgents born LeaguesJoins Picks lockedBacks
2026-11-05 64 0 0 0 0 0
2026-11-04 425 0 0 0 0 0
2026-11-03 380 0 0 0 0 0
2026-11-02 555 0 0 0 0 0
2026-11-01 632 0 0 0 0 0
2026-10-31 843 0 0 0 0 0
2026-10-30 594 0 0 0 0 0
2026-10-29 303 0 0 0 0 0
2026-10-28 296 0 0 0 0 0
2026-10-27 326 0 0 0 0 0
2026-10-26 313 0 0 0 0 0
2026-10-25 302 0 0 0 0 0
2026-10-24 306 0 0 0 0 0
2026-10-23 306 0 0 0 0 0
2026-10-22 454 0 0 0 0 0