Skip to content

Latest commit

 

History

History
408 lines (283 loc) · 16.6 KB

File metadata and controls

408 lines (283 loc) · 16.6 KB

Methodology: Correlation Verification and Standards

Purpose: Documents the statistical methodology, verification standards, and how to interpret the r-values and significance tests used throughout this repository.


v10.3 Precision Upgrade: From "14-Day Lag" to "7-Day Median Lag"

This is not a methodology change — it is a precision upgrade.

The underlying correlation (r = 0.6196, p = 0.0004) and all robustness tests are unchanged. What changed is the descriptive label applied to the lag:

Version Resolution Lag Label Basis
v10.2 Legacy 2-week index bins (n=30 rows, n=28 after lag) "14-day lag" Peak Pearson r at lag=2 bins
v10.3 High-Resolution Calendar-day backfill (n=66 pairs, 2017–2024) "7-day median lag" Actual measured median = 7 days (mean = 6.5)

The "14-day" figure was an artifact of the 2-week binning resolution used in the original 30-row index dataset. When the same events are measured at calendar-day resolution using the 66-pair historical backfill, the actual median response time is 7 days.

Evidence: 04_Testing_and_Counters/ROBUSTNESS_AUDIT_v10.2.md, 04_Testing_and_Counters/placebo_calendar_results.json


Key Claims

Claim Verification Evidence Location
Primary correlation r = 0.6196 is reproducible ✅ VERIFIED Run_Correlations_Yourself/run_original_analysis.py
p-value = 0.0004 (n = 28) ✅ VERIFIED Same script
Multiple robustness tests pass (16 scripts) ✅ VERIFIED Project_Trident/Copilot_Opus_4.6_Analysis/Statistical_Tests/
Independent verification by Opus 4.6 completed ✅ VERIFIED Project_Trident/Copilot_Opus_4.6_Analysis/Statistical_Tests/README.md

How to Reproduce the Analysis

Quick Start

cd Run_Correlations_Yourself/
pip install -r requirements.txt
python run_original_analysis.py

What the Script Produces

Part 1: r = 0.6196 (2-week lag, n = 28) - Primary correlation Part 2: Mann-Whitney U p = 0.002 - Project Trident Part 3: χ² = 330.62 (14-day periodicity) - Cross-validation


Understanding the Statistics

Pearson Correlation (r)

r Value Interpretation
0.0 No relationship
±0.1-0.3 Weak
±0.3-0.5 Moderate
±0.5-0.7 Strong
±0.7-1.0 Very strong

Our finding (r = +0.6196): Strong positive correlation - when friction events spike, compliance events follow ~7 days later (median; measured at 2-week index resolution).

p-value Interpretation

p-value Interpretation
p > 0.05 Not statistically significant
p < 0.05 Statistically significant
p < 0.01 Highly significant
p < 0.001 Extremely significant

Our finding (p = 0.0004): Less than 0.05% probability this occurred by chance.

Granger Causality

Tests whether one time series helps predict another.

Lag F-statistic p-value Significant?
1 32.49 < 0.0001 YES
2 14.74 < 0.0001 YES
3 8.68 < 0.0001 YES
4 6.43 < 0.0001 YES

Interpretation: Friction events significantly predict compliance events at all lags tested.


Robustness Tests

Independent Verification by Opus 4.6

After the repository owner established the original correlations, GitHub Copilot (Claude, Opus 4.6) independently wrote and ran a suite of 16 statistical test scripts to stress-test the findings. Opus 4.6 did not build the datasets or compute the original correlations — it received the data and designed its own tests to challenge them.

Test Suite Location

Project_Trident/Copilot_Opus_4.6_Analysis/Statistical_Tests/

Available Scripts

Script Purpose
permutation_test.py Shuffle-based significance
autocorrelation_adjusted_test.py Block bootstrap
normalized_correlation.py Per-year normalized correlation
cross_validation_dec2025.py Dec 2025 exclusion test
rolling_window_correlation.py Sliding-window analysis
event_study_framework.py Compliance response analysis
granger_causality_test.py Predictive direction test

Results Summary

Test Result Verdict
Permutation (10K shuffles) p < 0.0001 — observed r beat 10,000 random shuffles ✅ Pass
Autocorrelation adjustment Pearson p = 0.008 (block-bootstrap), Spearman ρ = 0.61 ✅ Both survive
Dec 2025 exclusion Pearson r drops 6%, Spearman ρ = 0.60 ✅ Signal survives
Normalized (binary) r = 0.59 (p < 0.0001) ✅ Presence/absence holds
Event-study Friction dates attract 20–42x more compliance ✅ Strong colocation
Granger (hand-scored) F→C at lag 1 (p = 0.0008), lag 2 (p = 0.027) ✅ Predictive
Partial correlation (political calendar) < 1% of correlation explained by congressional session ✅ Not a confound
Granger (first-differenced) Direction consistent after stationarity correction ✅ Robust
Rolling window (13/26/52 wk) Correlation present across multiple time periods ✅ Not driven by one cluster
Historical backfill (66 pairs) Δr = +0.0012 ✅ Negligible impact

Dataset Scope and Classification

Friction Events (attention-consuming)

  • Epstein-related releases and coverage
  • Political events and media reactions
  • Crisis events
  • DOGE/FDA conflict events

Compliance Events (policy/financial positioning)

  • Policy and geopolitics
  • Government ties
  • Strategic shifts
  • Crypto pivots
  • FDA/regulatory changes
  • Financial performance indicators

What's Excluded from Primary Correlation

High_Growth_Companies_2015_2026.csv (1,049 records) is excluded because:

  • Contains operational events (clinical milestones, earnings)
  • These follow medical/market schedules, not strategic calendar exploitation
  • Including them dilutes r = 0.6685 to r = 0.5268

Location: 14_Files/TRANSPARENCY_NOTE_FOR_2026_ANALYSIS.md


Verification Levels

Throughout the repository, claims are marked:

Mark Meaning
✅ VERIFIED Confirmed through multiple independent sources
⚠️ PARTIALLY VERIFIED Some evidence supports; needs more verification
🔍 HYPOTHESIS Proposed but not yet verified
❌ FAILED Prediction did not materialize

Scout Methodology

Core Approach: Observe and report patterns without claiming intent or coordination.

What it does:

  • Documents patterns
  • Notes correlations
  • Flags timing
  • Verifies through multiple sources

What it doesn't do:

  • Claim conspiracy
  • Assert intent
  • Accuse individuals of coordination

Multi-AI Verification

Process: Cross-checking findings across multiple AI systems (Claude, Grok, Gemini) to identify blind spots or biases.

Why: Different AI systems have different training data. Convergent findings are more robust.


Source Triangulation

Standard: Major claims require verification from at least two independent sources.

Source Types Used:

Type Examples
Government primary DHS.gov, SEC filings, DOJ releases
Wire services AP, AFP, Reuters
Major outlets NPR, CNN, Bloomberg, WSJ
Investigative ProPublica, Byline Times
International France24, Al Jazeera

Known Limitations

  1. Event classification subjectivity: What counts as "friction" vs "compliance" involves judgment
  2. Autocorrelation present: High temporal clustering (r = 0.67 at lag 1 for friction)
  3. Outlier sensitivity: December 2025 disproportionately influential
  4. Dataset mixing errors: See Run_Correlations_Yourself/Wrong_Correlations/ for archived mistakes

Common Questions

Why r = 0.6196 vs r = 0.6685?

Correlation Dataset Scope
r = 0.6196 30-week hand-scored Original pre-2026 data
r = 0.6685 1,027 strategic events 2026 raw event counts
r = 0.5268 2,069 total events Including operational events

All three are valid for their scopes. The 0.6196 is the canonical reference.

What about autocorrelation?

High autocorrelation (r = 0.67) means friction events cluster temporally. This is controlled for via block bootstrap (p = 0.008) and first-differenced Granger tests.

What about December 2025?

Removing December 2025 drops Pearson r by 6% but Spearman ρ remains 0.60. The pattern survives exclusion.


Key Sources

Document Location
Run Correlations README Run_Correlations_Yourself/README.md
Original analysis script Run_Correlations_Yourself/run_original_analysis.py
Verification Report 14_Files/VERIFICATION_REPORT_Jan2026.md
Transparency Note 14_Files/TRANSPARENCY_NOTE_FOR_2026_ANALYSIS.md
Glossary 14_Files/Glossary.md

Cross-References

  • For core theory: 01_CORE_THEORY.md
  • For datasets: 08_KEY_DATASETS.md
  • For statistical tests: Project_Trident/Copilot_Opus_4.6_Analysis/Statistical_Tests/

Lag=5 Negative Oscillation (r = −0.6064)

Finding

The lag sweep reveals a strong negative correlation at lag=5 weeks:

Lag sweep on 30-row index (Pearson r):
  lag=0: r = −0.0323 (p = 0.8653)    — No simultaneous relationship
  lag=1: r = +0.5034 (p = 0.0054)    — Strong positive (emerging)
  lag=2: r = +0.6196 (p = 0.0004)    — PEAK positive ← Canonical finding
  lag=3: r = +0.2849 (p = 0.1497)    — Declining
  lag=4: r = −0.4069 (p = 0.0391)    — Reversal begins
  lag=5: r = −0.6064 (p = 0.0013)    — STRONG NEGATIVE ← Documented here
  lag=6: r = −0.3363 (p = 0.1081)    — Declining

Theoretical Explanation: The Thermostat Cooling Cycle

The negative correlation at lag=5 is consistent with a thermostat oscillation — a self-regulating feedback loop in the friction→compliance system:

  1. Lag 0–2 (Heating phase): A friction event creates an attention window. Compliance actors use this window to execute institutional changes. Friction and compliance move together (positive r).

  2. Lag 2 (Peak): Maximum correlation. The compliance response is at its strongest. This is the 7-day median response window measured at index resolution.

  3. Lag 3–4 (Saturation): The attention window closes. Media moves on. The friction "fuel" is exhausted. The correlation drops toward zero and begins inverting.

  4. Lag 5 (Cooling phase): The system overcorrects. After a major compliance push (policy changes, financial moves), there is a period of institutional quiescence. Simultaneously, the absence of friction coverage may trigger a new friction event to maintain attention management. The result: high friction co-occurs with low compliance (and vice versa), producing a strong negative r.

  5. Lag 6+ (Reset): The oscillation dampens and the system resets for the next cycle.

Significance

  • r = −0.6064 at lag=5 is as statistically significant as the primary finding (p = 0.0013 vs p = 0.0004)
  • This suggests the friction→compliance system is not one-directional but oscillatory
  • The full cycle is approximately 10 weeks (peak-to-trough: lag 2 to lag 5 = 3 index bins)
  • This oscillation is consistent with the "thermostat" metaphor: the system self-regulates to prevent either sustained attention or sustained compliance activity

Implication

The "14-day lag" (now corrected to 7-day median) is only half the story. The full pattern is a ~10-week oscillation where:

  • Weeks 0–2: Friction → Compliance (positive coupling)
  • Weeks 3–4: Transition/decorrelation
  • Weeks 5–6: Friction → Anti-Compliance (negative coupling, "cooling off")

Evidence: Run_Correlations_Yourself/run_original_analysis.py (lag sweep), 04_Testing_and_Counters/ROBUSTNESS_AUDIT_v10.2.md


Business Cycle Artifact Check (v10.3)

Finding

The weekday frequency analysis on the 66-pair backfill dataset reveals:

Metric Value Expected by Chance
Friction & compliance share SAME weekday 30.3% 14.3% (1/7)
Lags that are exact multiples of 7 22.7% 14.3%
Friction events on Wednesday/Friday 47.0% 28.6%
Compliance events on weekdays (Mon–Fri) 97.0% 71.4%

Interpretation

The 7-day median lag is partially a business-cycle artifact. Events cluster on weekdays (compliance events almost never occur on weekends), and 30.3% of friction–compliance pairs share the same weekday (2.1x the expected rate). This "same-weekday" effect inflates the apparent 7-day signal.

However, the weekday effect does not fully explain the lag:

  • Mean lag varies by friction weekday (Monday: 5.0d, Saturday: 9.3d), showing the lag is not a fixed weekday-to-weekday jump
  • The chi-square test rejects uniform weekday distribution for both friction (p = 0.035) and compliance (p = 0.001), confirming systematic weekday clustering

Verdict

The 7-day median lag is a composite of:

  1. A genuine sequential friction→compliance response (~5–7 days)
  2. A business-cycle anchoring effect (events snap to workdays)
  3. Calendar-anchor clustering (71.2% of pairs share the same calendar anchor)

Evidence: 04_Testing_and_Counters/ROBUSTNESS_AUDIT_v10.2.md


Financial Anchor Alignment (v10.3)

Finding (February 2026 Case Study)

Cross-referencing 11 February 2026 compliance events against Tier 1 entity financial anchors (BlackRock Q4 earnings, Apollo Q4 earnings, SEC 13F deadline, Apollo dividend record date):

Metric Value
Mean distance to nearest financial anchor 1.7 days
Median distance to nearest financial anchor 2.0 days
Events within 3 days of a financial anchor 81.8%
Events within 7 days of a financial anchor 100%
Events exactly on a financial anchor date 5 of 11 (45.5%)

Comparison

Clustering Method Mean Distance
Financial anchors (Feb 2026) 1.7 days
7-day sequential lag (backfill) 6.5 days
Calendar anchors (holidays/solstices) 5.6–6.5 days

Financial anchors provide 3.8x tighter clustering than the sequential lag model. The February 2026 compliance window is better explained by the earnings/filing calendar of Tier 1 entities than by sequential friction→compliance reaction.

Key Overlaps

  • Feb 9: Apollo Q4 earnings = Maxwell testimony (same day)
  • Feb 17: SEC 13F filing deadline
  • Feb 19: Apollo dividend record date = Board of Peace inaugural summit (same day)

Implication

For periods with dense financial anchors (earnings season, filing deadlines), the "financial calendar" is a stronger explanatory variable than the friction→compliance sequential lag. The 7-day median lag may represent the typical spacing between financial calendar events during earnings season, not a sequential reaction time.

Evidence: 04_Testing_and_Counters/temporal_node_reconciliation.py, SEC EDGAR, Apollo IR


Quick Facts

Fact Value
Primary correlation r = +0.6196
Sample size n = 28 (after lag)
Significance p = 0.0004
Granger F-statistic (lag 1) 32.49
December 2025 Z-score 2.35 (top 1% of months)

This summary distills content from Run_Correlations_Yourself/, 14_Files/VERIFICATION_REPORT_Jan2026.md, 14_Files/TRANSPARENCY_NOTE_FOR_2026_ANALYSIS.md, and 14_Files/Glossary.md.


Additional Methodology Resources

Resource Location Content
Scout methodology origin 00_Quick_Breakdowns/About_Me.md Author's Army Cavalry Scout background → "Numbers Station" observational concept
16 statistical test scripts Project_Trident/Copilot_Opus_4.6_Analysis/Statistical_Tests/ Permutation, autocorrelation, Granger, rolling window, event-study, partial correlation, and more
Dataset provenance Project_Trident/Copilot_Opus_4.6_Analysis/Findings/dataset_provenance.md Data origin documentation
AI fabrication case study Project_Trident/Copilot_Opus_4.6_Analysis/Findings/AI_Fabrication_Case_Study.md How Grok-fabricated data was identified and retracted (Layers 2-3)
Wrong correlations archive Run_Correlations_Yourself/Wrong_Correlations/ Deprecated scripts that mixed datasets — preserved for transparency
Independent verification 09_Silicon_Sovereignty/CRUCIAL-Cross_Verification_Check.md Cross-AI verification of Silicon Sovereignty claims
Validation report (Feb 24) docs/validation/VALIDATION_REPORT_2026-02-24.md External validation report

Updated March 2, 2026 (v10.3).