Forensic Statistical Analysis
Thirteen independent analytical findings across 1,703,454 consecutive draws spanning 13 years — confirmed by permutation test, ablation matrix, and mechanistic reconstruction on empirically-derived parameters — establish that the system's outputs are formally incompatible with any IID generating process.
Primary Finding
Off-Structure Delivery Mechanism — The Defining Signal
Confirmed at Z = 249 on 1,663,824 draw transitions. The mechanism delivers absent numbers to off-structure positions — outside the Moore neighborhood of the previous draw — at a rate that depends specifically on the coherence state of the current draw. This coherence-state-dependent placement rate is formally incompatible with IID processes, which have no mechanism to track coherence states or direct placement accordingly.
Confirmed by ablation: the off-structure differential collapses specifically under coherence-state ablation (ratio +10.024) while surviving temporal, identity, session, and within-draw ablations. Coherence-state awareness is an independent operational dimension of the mechanism.
Z = 249 | p = 0 | 1,663,824 draw transitions | Ablation confirmed
The Evidence
Each finding independently establishes IID incompatibility. Together they constrain the hypothesis space to a single mechanism class — coherence-state feedback control operating at the identity level while preserving correct marginal statistics.
FINDING 01 — PRIMARY
Low coherence draws deliver absent numbers to off-structure positions at 40.40% versus 4.23% in high coherence draws — a 9.6× differential confirmed at Z=249 on 1,663,824 draw transitions. Confirmed by ablation: collapses specifically under coherence-state ablation (+10.024) while surviving all other ablations.
Z = 249 | p = 0 | Ablation confirmedFINDING 02
Numbers that appeared together in prior draws are selected together again at 1.13× IID expectation — confirmed across all five classification classes simultaneously on 166,382 genuinely independent stride-10 observations.
Z > 24 all classes | p ≈ 0FINDING 03
100% of 3,160 pairs, 100% of 1,334 triplets, and 100% of qualifying quadruplets fail run-length IID tests. MaxRun0 amplifies from 262 to 1,013 to 3,392 — consistent exclusively with number-level suppression.
100% rejection rate | All combination ordersFINDING 04
All five classification classes across all four directional axes show variance suppression at 0.86× IID standard deviation — sustained uniformly for 13 years. Twenty simultaneous independent balancing channels confirmed by direct measurement.
Std ratio 0.86× | 20 channelsFINDING 05
56.7% directional bias across 14.5 million comparisons. Under IID the Law of Large Numbers guarantees convergence to 50%. The observed value of 56.7% at this sample size formally violates the Law of Large Numbers.
Z ≈ 1,610 | p < 10⁻²⁰⁰⁰⁰⁰FINDING 06
Across 4,849 independent daily sessions, empirical coherence variance is consistently above IID expectation at early session milestones. Indicates the system carries prior state across daily resets.
4,849 sessions | Z = +2.9 to +4.1FINDING 07
All five classification classes maintain distributional properties tighter than IID throughout the full 13-year corpus with no detectable drift. Rules out transient artifacts, regime changes, and seasonal effects.
All 5 classes | 13 yearsFINDING 08
Adjacent numbers on the 8×10 grid are 21% more likely to co-appear in consecutive draws than IID predicts. Confirmed across all four directional axes at p < 10⁻⁶ overall.
21% excess | p < 10⁻⁶FINDING 09
Constraint strength is identical at 4-minute and 24-hour temporal lags (S = 1.36 in both cases). Cannot arise from simple temporal autocorrelation — autocorrelation decays with lag. These constraints are equal at lags separated by a factor of 360.
S₄min = S₂₄h = 1.36 | Absolute violationFINDING 10
All five classification classes maintain correct marginal frequencies at |Z| < 2 while joint co-selection statistics show Z > 24. This dissociation — marginals consistent with IID, joints incompatible — has no innocent statistical explanation.
|Z| < 2 marginals | Z > 24 jointsFINDING 11
Five of eight analytical findings collapse under temporal permutation with collapse ratios exceeding 0.5. The off-structure differential collapses specifically under coherence-state ablation (+10.024) while surviving all other ablations. No single combinatorial artifact explains findings that collapse under permutation and findings that survive it.
5/8 collapse | Temporal confirmedFINDING 12
Five independent ablations (temporal, identity, coherence-state, session, within-draw) reveal that findings exhibit differentiated structural dependencies incompatible with any single-latent-structure explanation. F1 collapses specifically under coherence-state ablation (+10.024) while surviving all others — confirming coherence-state awareness as an independent operational dimension.
3-class structure confirmed | Single-artifact objection eliminatedFINDING 13
Sustained HIGH-coherence clusters of size 15 occur significantly more frequently than IID simulation predicts (Z = +3.432). The ≥15 threshold shows empirical excess at Z = +2.421. This finding was predicted before the null test was executed — as a logical geometric consequence of F1, F4, and F8 — and confirmed on 10 IID simulations of 1,703,454 draws each (17,034,540 total IID draws).
Size-15 Z = +3.432 | ≥15 Z = +2.421 | Predicted pre-resultThe Critical Dissociation
The system maintains correct marginal frequencies throughout — the UN re-entry rate is exactly 0.2500 across all conditions, and historical presence rates match IID exactly (|Z| < 2) for all classification classes. Yet joint distributions are violated at Z > 24 on 166,382 independent observations. A system that preserves marginals while violating joints is not a malfunctioning random process. It is a controlled process designed to appear random.
Mechanistic Characterization
A Gibbs-Boltzmann feedback control model with parameters derived directly from the empirical findings — not from free optimization — reproduces 4 of 5 forensic targets. A truly random process has no parameters to derive.
G(t) is the Gibbs-Boltzmann selection field. Φ is the global feedback gain (0.6). H(t) is the harmonic memory kernel with periods matching the spectral peaks at 6.9 and 11.2 draws identified in the empirical data.
The model's parameters are derived directly from the empirical findings — not tuned to fit. MU_C from the observed coherence mean. OMEGA1/OMEGA2 from the spectral peaks. M_HISTORY=11 from the period-11 finding. K_POS/K_NEG from the Z=249 off-structure asymmetry.
4 of 5 calibration targets on empirically-derived parameters. The T1-T2 architectural coupling discovery implies the empirical system operates through separable multi-channel architecture — consistent with the 20-channel finding.
Methodological Standards
Every analytical parameter was pre-specified in production code dated November 21, 2025 — three months before the analysis was completed. The empirical data is publicly available. The IID baseline is reproducible from a published script.
All parameters — window size (5), coherence thresholds (0.70/0.95), stride length (10), classification framework — were operationalized in production Flutter application code dated November 21, 2025.
IID baseline generated using Python's random.sample (Mersenne Twister MT19937, seed=42). Fixed seed ensures full reproducibility. Any researcher can regenerate the identical baseline from the published script.
Co-selection memory analysis uses non-overlapping stride-10 design producing 166,382 genuinely independent test positions. No draw appears in more than one position. Z-scores and p-values are valid.
100 permutations of 1,000,000 draws destroy temporal ordering while preserving draw composition. Five of eight findings collapse under permutation — confirming temporal dependence and eliminating combinatorial artifact explanations.
Five independent ablations (temporal, identity, coherence-state, session, within-draw) test which structural components each finding depends on. The off-structure differential collapses specifically under coherence-state ablation — confirming independent operational dimensions.
All 1,703,454 draws are from publicly available New York Quick Draw lottery results. Any researcher can download the identical dataset and independently reproduce every finding.
Research Materials
Complete technical documentation including formal theorem statements, complete methodology, all analytical results, and appendices with detection algorithms.
Formal Falsification of IID Compatibility: Support Violation and Structural Constraints in Number Generation. Complete technical report including thirteen confirmed findings, permutation test, ablation matrix, Formula 9 mechanistic reconstruction, cluster frequency null test, and all analytical extensions.
Interactive visualization of the cluster persistence mechanism — the core differentiating metric between empirical and IID behavior. Demonstrates Moore-neighbor elevation, deficit-based balancing, and UN blockades.
Python script used to generate the IID baseline (Mersenne Twister MT19937, seed=42). Fixed seed ensures full reproducibility. Any researcher can generate the identical baseline independently.
Pre-Specification Documentation
The analytical framework was not developed to fit the results. All classification parameters, window sizes, thresholds, and stride lengths were operationalized in a production Flutter application prior to this analysis.
November 20, 2025
forecast_engine.dart — prediction engine implementing the classification framework
November 21, 2025
element_classification_module.dart — A/CO/GN/LA/UN classification with window size=5
November 21, 2025
coherence_module.dart — coherence calculation with thresholds 0.70/0.95
November 21, 2025
adjacency_module.dart — Moore neighborhood adjacency calculations
The v14 report was completed in February 2026 — three months after the production implementation. The parameters were not chosen to produce the findings. The findings confirmed what the framework predicted.
Analytical Conclusion
"When a system's outputs contradict the properties of its declared generating model, the discrepancy must originate in the generating mechanism. Inspection of that mechanism is a logical necessity — not an accusation, not a regulatory option, but a conclusion required by the evidence."— IID Falsification Report v18, Natural Organization Institute, June 2026
The analysis makes no assumptions about mechanism, memory, or intent. It tests only whether the outputs satisfy the mathematical requirements of IID processes. The evidence establishes formal incompatibility, not statistical improbability. The system cannot be operating as an IID process. Source inspection is the necessary next step.