LotteryOracle
GamesForecastPlaygroundCase filesMy numbers
← Cash Pop Evening
pool inferredsingle source
Frequency ExplorerGap & “Overdue” TrackerSum DistributionConsecutive Numbers SurpriseHarmony MatrixNumber ConstellationCarry-over / Déjà VuDelta SpaceCoverage RaceDraw FingerprintsRandomness ObservatoryPositional RibbonsCross-Lottery MirrorSix true storiesCultural Numbers AtlasThe Unpopular Numbers EdgeWheel WorkshopMyth RebuttalPattern CourtDraw Rank StreamAnytime Bias WatchWhole-Draw DriftConditional HarmonyResidue KaleidoscopeDigit LaboratoryCrowd Calibration LabCalibration ForgeJohnson SpectrumCompression LabForecast ArenaExchangeability WatchGap UniformizerSharing MapChernoff MapReturn LedgerRule-Change LedgerMachine RoomCalendar CourtBalls versus RNGWorld Fairness BoardRegression LedgerLottery Design StudioChance DetectiveThe Headline FactoryThe Crowd Through TimeThirty Years of SaturdaysPattern Recipe LabArithmetic Progression GardenSame Odds, Different WaitsWrong Formula LabEuler’s Lottery DeskCasanova’s Risky TreasuryThe Lottery That Chose LivesThe Museum Built by Chance

Loading…

Reading the draws and working out what chance predicts.

← PreviousForecast ArenaNext →Gap Uniformizer
LotteryOracle

LotteryOracle reports what happened; it does not predict what is next. Every draw is independent — no number is ever "due".

GamesPlaygroundCase filesThe weekDataAboutMethodologyPrivacy
Built byejoyful

Responsible play.A lottery costs more than it pays back; if it stops being fun, help is free and confidential:🇺🇸 1-800-GAMBLERall helplines →

Exchangeability Watch

Let each complete draw ask whether the stream still looks like itself.

A draw is unusual when its overlap geometry differs from the history around it. Online conformal ranks turn that surprise into p-values; a predeclared betting mixture accumulates evidence without invalidating repeated looks.

Whole-draw evidence capital

Cash Pop Evening1/15 era · since Nov 2022600 draws

Current evidence
0.04×
Maximum evidence
2.20×
First 20× crossing
none
Smallest conformal p
0.0015
0–0.158
0.1–0.270
0.2–0.360
0.3–0.438
0.4–0.551
0.5–0.664
0.6–0.769
0.7–0.873
0.8–0.943
0.9–155

Most structurally isolated draws

2025-12-04 · 60.600
2025-11-05 · 60.600
2025-09-24 · 60.600
2025-09-16 · 60.600
2025-09-12 · 60.600
2025-09-07 · 60.600
2025-07-12 · 60.600
2025-06-26 · 60.600

Online full-conformal power martingalewhole-draw-conformal-martingale-v2seed 202608273 pre-declared bandwidths x 3 betting powers, all 9 mixed before inspection

The 20× line is an anytime-valid investigation threshold, not a prediction or allegation.
Analysis ledger
Draw stream
midday
Usable evidence
600 recent / 1286 draws
Trust filter
all canonical draws
Source omitted
none
Conformity score
mean Johnson-overlap kernel distance
Stopping rule
Ville threshold; valid under repeated inspection
Kernel bandwidth
0.250, 0.500 and 1.000 mixed with equal prior, so the alarm stays one declared anytime-valid reading
∑The maths, in plain language

Conformal surprise and betting capital

For every prefix, each draw receives the permutation-equivariant score

ai=1−1t−1∑j≠iexp⁡(−γ∣Si△Sj∣).a_i=1-\frac{1}{t-1}\sum_{j\ne i}\exp(-\gamma|S_i\triangle S_j|).ai​=1−t−11​∑j=i​exp(−γ∣Si​△Sj​∣).

The newest score is ranked among all scores in that prefix. Random tie-breaking turns the rank into a conformal p-value ptp_tpt​. Under exchangeability these p-values are uniform, without specifying the draw distribution.

For a fixed 0<ϵ<10<\epsilon<10<ϵ<1, ϵptϵ−1\epsilon p_t^{\epsilon-1}ϵptϵ−1​ is a betting factor. Products of those factors form a test martingale. The watch averages three predeclared martingales, so Ville's inequality makes a 20× crossing an anytime-valid 5% alarm.

✦The story behind it

A monitor that does not need to name the fault first

Frequency tests look for frequency faults; drift windows look for a chosen kind of boundary. Conformal monitoring asks a broader question: does the newest complete object still exchange places naturally with those before it?

That makes it useful as a smoke alarm. It can justify investigation, but the same evidence cannot diagnose the fire. The construction follows research on conformal test martingales.