Ruled Out
55 hypotheses tested and ruled out. 4 LSDP-strippedPatterns that looked real in raw data but disappeared when confounds (social contagion, media influence, etc.) were removed by the LSDP pipeline. patterns, 5 refuted investigations, 16 dead-listVariable pairs that were tested for correlation but showed no significant relationship — permanently removed from further screening. pairs from ADPAnomaly Detection Protocol — 20 statistical screening methods applied across all variable pairs. screening, 11 sign-reversalVariables where the effect was real but went the OPPOSITE direction from what the hypothesis predicted. E.g., triggered reports had MORE fear language, not less. content variables from Layer 7, 19 linguistic featuresAll linguistic style features tested in Phase 9B — every one killed after corrections and source controls. from Phase 9B, plus 4 closed investigative approaches. Ruling things out IS progress.
| Hypothesis | Method | Result | Why It Failed |
|---|---|---|---|
| NUFORC summer peak is real | Hawkes L2 decomposition | STRIPPED | Social contagion explains summer clustering |
| NUFORC weekend effect is real | Hawkes L2 decomposition | STRIPPED | Reporting convenience, not phenomena |
| NUFORC exponential growth is real | Hawkes L2 + L5 + L6 | STRIPPED | Website launch + media + instruments |
| NUFORC 2014 spike is real | All layers | STRIPPED | Not present in raw data either |
| MPI stress proxy correlates with geomag | Partial correlation + Granger | REFUTED | Confounded by secular trends; no causal link after detrending |
| Armed conflict × geomagnetic activity | ADP lag analysis | REFUTED | No significant correlation at any lag; temporal structure explained by retaliation dynamics |
| FAERS adverse events × geomagnetic activity | ADP correlation + MI | REFUTED | No signal above noise floor; reporting artifacts dominate FAERS temporal patterns |
| Dream ESP accuracy × mortality rates | ADP screening | REFUTED | Insufficient temporal overlap; dataset too small for meaningful analysis (N=63) |
| Whale strandings × geomagnetic storms | Coincident extremes + lag | REFUTED | Published correlations do not replicate with proper temporal controls |
| Geomag × NDE correlation | Partial correlation | KILLED | Confounded by Kp self-correlation |
| Alcohol consumption → birth rate | ADP screening | KILLED | Session 1 validation failed — spurious temporal correlation |
| Volcano EVT distribution anomaly | Extreme value analysis | KILLED | Distribution matches standard GEV after monitoring correction |
| FAERS weekend death pattern | ADP screening | KILLED | Reporting artifact — deaths reported on Monday for weekend events |
| Helios butterfly × ocean temperature | Wavelet coherence | KILLED | Coherence vanishes after detrending; shared secular trend only |
| Solar flare × earthquake triggering | Granger causality + lag | KILLED | No significant lag correlation at any offset (1h–90d) |
| Lunar phase × NUFORC sightings | Phase correlation | KILLED | Observer availability (darkness) fully explains apparent lunar signal |
| Geomag Kp × hospital admissions | Partial correlation | KILLED | Confounded by day-of-week effects and seasonal patterns |
| CME impact × Bigfoot sightings | Coincident extremes | KILLED | No temporal alignment; BFRO is pure social contagion |
| Sunspot cycle × conflict intensity | Wavelet coherence | KILLED | Apparent 11-year cycle is artifact of shared trend |
| Fireball rate × seismic activity | Pairwise correlation | KILLED | Independent processes; no physical mechanism |
| Hum reports × infrasound stations | Lag analysis | KILLED | Early-period Hum is social contagion; late-period shows no infrasound correlation |
| NDE reports × solar wind speed | Mutual information | KILLED | NDE submission timing is media-driven; no physical pathway |
| Ganzfeld hit rate × local Kp | Partial correlation | KILLED | Branching ratio explains temporal clustering; Kp adds nothing |
| Storm events × UFO sightings | Coincident extremes | KILLED | Storm events suppress outdoor observation; negative correlation is trivial |
| Reddit paranormal × full moon | Phase correlation | KILLED | Reddit posting is pure social contagion (BR=0.986); no lunar signal |
| Witness count higher in independent reports | Content analysis (20k) | KILLED | Sign reversal: triggered reports mention more witnesses (social validation) |
| Duration mentions enriched in independent | Content analysis (20j) | KILLED | Sign reversal: triggered reports have longer duration claims |
| Distance estimates differ by group | Content analysis (20j) | KILLED | No significant difference after FDR correction |
| Fear language marks independent reports | Content analysis (20d) | KILLED | Sign reversal: triggered reports use more fear language |
| Technical vocabulary marks independent | Content analysis (20m) | KILLED | No significant difference in vocabulary diversity metrics |
| Report complexity higher for independent | Content analysis (20l) | KILLED | Sign reversal: triggered reports are more verbose |
| Sound descriptions mark independent | Content analysis (20e) | KILLED | Sound descriptors distributed equally across groups |
| Movement patterns differ by group | Content analysis (20i) | KILLED | No significant group difference after FDR correction |
| Location specificity marks independent | Content analysis (20e) | KILLED | Both groups provide similar location detail levels |
| Time-of-day language differs | Content analysis (20g) | KILLED | Time references reflect actual observation time, not group membership |
| Weather mentions enriched in independent | Content analysis (20e) | KILLED | No significant enrichment after cloud cover correction |
Linguistic Style Features Phase 9B — ALL KILLED
All 19 linguistic features tested. All killed after corrections and source controls. The “absence and restraint” pattern (less sensory language, more negation) was entirely explained by reporting platform.
circular (target leakage), removed
dataset composition artifact
killed by source controls (ΔR² = 0.001)
killed by source controls
killed by source controls
killed by source controls
killed by source controls (0/4 domains)
REVERSES direction between domains
killed by source controls
killed by source controls
killed by source controls
dataset composition artifact
killed by source controls
killed by source controls
killed by source controls
killed by source controls
killed by source controls
killed by source controls
killed by source controls
Geomagnetic-Anomalous Experience Hypothesis CLOSED
Case-crossover design with matched control windows. No geophysical variable (Kp, Ap, F10.7, Dst, seismic activity, lunar phase) predicts when independent anomalous reports are filed. This closes decades of speculation with the strongest test ever run.
Simpson's Paradox in Cross-Database Analysis CONFOUND IDENTIFIED
Global correlation between independence probability and content uniqueness (ρ=−0.31) was entirely a dataset-level confound. Within each dataset: ρ=−0.02. All naive cross-database comparisons without source controls are suspect.
Monthly Cross-Domain Correlations APPROACH CLOSED
360 pairwise Spearman correlations across 28 monthly time series from 6 domains. After detrending: all cross-domain correlations were artifacts of shared secular trends (population growth, internet adoption, improved instrumentation). This approach is permanently closed.
Quadratic Detrending Lesson METHODOLOGICAL
Upgrading from linear to quadratic detrending killed 51% of cross-domain correlations. All solar × UFO correlations died. ~Half of published cross-domain research using linear detrending may be reporting curve-shape artifacts.
Null results have the same scientific value as positive results. Every hypothesis ruled out narrows the search space and prevents other researchers from duplicating effort.