Skip to content

VALIDATION.md

Latest commit

 

History

History
145 lines (105 loc) · 7.03 KB

VALIDATION.md

File metadata and controls

145 lines (105 loc) · 7.03 KB

Computational Validation: IRIS Gate Evo Convergence Data

This document maps each core claim in the manuscript to specific convergence results from the IRIS Gate Evo multi-model protocol.

IRIS Gate Evo fires the same compiled prompt to five independent LLM mirrors (Claude, Gemini, Grok, Mistral, DeepSeek). Models never see each other's outputs. Convergence is measured server-side by semantic embedding + clustering. TYPE is assigned by model count: 5/5 = TYPE 0, 4/5 = TYPE 0, 3/5 = TYPE 1, 2/5 = TYPE 2, 1/5 = TYPE 3 (singular).

Full run data: iris-evo-findings

Validating Runs

Two independent IRIS Gate Evo runs passed the S3 convergence gate on the CBD/VDAC model. Both produced full protocol packages (S4 hypotheses, S5 Monte Carlo, S6 structured output).

Run Session ID S3 Cosine Outcome
EFSA Stress Test evo_20260213_035958_pharmacology 0.8569 PASSED
Chronic Dosing evo_20260213_042930_pharmacology 0.9024 PASSED

Cross-run analysis detected the system's first INDEPENDENT REPLICATION (cosine 0.789): both runs independently converged on sub-Kd hepatic CBD concentrations at 50mg/day oral dosing.

Claim-Level Validation

Core Thesis: CBD Is a Universal Mitochondrial Stressor

Manuscript Claim IRIS Evidence TYPE Models Run
CBD does not discriminate VDAC1 vs VDAC2 >70% transmembrane homology; no published isoform selectivity data TYPE 0 5/5 EFSA
Membrane partitioning (logP 6.3) is cell-type-indiscriminate 100-1000x enrichment in mitochondrial membranes regardless of cell type Singular (Claude) 1/5 EFSA
CBD engages both VDAC1 and VDAC2 in all cells VDAC2 releases BAK in healthy cells too; undermines selectivity argument TYPE 0 5/5 EFSA

Bioenergetic Resilience: GSH Determines Outcome

Manuscript Claim IRIS Evidence TYPE Models Run
Healthy hepatocyte GSH synthesis: 0.5-2.3 umol/g/hr Rate-limited by GCL; depends on ATP and cysteine TYPE 0 4/5 (claude, gemini, grok, mistral) Chronic
NAFLD/alcohol liver GSH synthesis drops >2x GCL downregulation + cysteine depletion in metabolic disease TYPE 0 4/5 (claude, gemini, grok, mistral) Chronic
GSH depletion threshold ~20 uM intracellular CBD Unlikely in healthy liver under oral dosing; possible in compromised cells TYPE 0 4/5 EFSA
ROS scavenging ratio (healthy:HCC) < 2x Modest selectivity at GSH level; not the dramatic gap assumed in older models TYPE 0 4/5 EFSA

Chronic Safety: The Sponge Refills

Manuscript Claim IRIS Evidence TYPE Models Run
CBD-VDAC1 binding is reversible (t_off << 12hr) Non-covalent; VDAC gating dynamics (ns-us) >> drug residence (<hr) TYPE 0 4/5 (claude, gemini, grok, mistral) Chronic
Channel resets between dosing intervals No covalent modification; no conformational memory TYPE 0 4/5 Chronic
7-OH-CBD accumulates 2-5x at steady state Slower clearance (t1/2 ~18-32 hr vs 6-12 hr) TYPE 0 4/5 (claude, deepseek, grok, mistral) Chronic
At 50mg/day, hepatic CBD ~0.1-0.5 uM, VDAC occ <5% ROS load < 2% of daily GSH synthesis capacity TYPE 1 3/5 (claude, deepseek, grok) Chronic
Null defeated: no progressive GSH depletion in healthy liver Safety margin 5-10x for healthy, 2-3x for compromised TYPE 1 3/5 (claude, deepseek, gemini) Chronic

Safety Margin & Regulatory Implications

Manuscript Claim IRIS Evidence TYPE Models Run
Safety margin 5-10x in healthy liver GSH synthesis vastly exceeds oxidative load from sub-Kd VDAC occupancy TYPE 1 3/5 Chronic
Margin narrows to 2-3x in compromised liver NAFLD loses both synthesis capacity and baseline GSH reserves TYPE 1 3/5 Chronic
Risk-stratified dosing > universal milligram limit At-risk populations (NAFLD, hepatitis, alcohol) need lower limits; healthy do not Synthesis Both runs Both

Open Questions and Known Gaps

Gap Status Proposed Experiment Testability
VDAC1 vs VDAC2 binding selectivity (Kd ratio) Unknown; predicted <5x by IRIS MST/ITC in lipid nanodiscs + BAK oligomerization assay 7/10
7-OH-CBD VDAC1 Kd Unknown; predicted ~10-20 uM (Grok) or weaker (DeepSeek) MST/SPR binding + Seahorse OCR 5/10
Cumulative ATP deficit independent of GSH Unresolved; Mistral dissent Longitudinal ATP/ADP with ROS/GSH clamped via NAC/BSO 6/10
VDAC gating persistence post-washout Predicted reversible; no direct measurement Bilayer electrophysiology with washout kinetics 6/10

Monte Carlo Power Analysis

All operationalized hypotheses from both runs achieved statistical power = 1.0 at 300 iterations.

EFSA Stress Test Hypotheses

ID Description Effect Size (d) Power
H1 Membrane enrichment quantification 0.931 1.0
H2 Electrophysiology: gating vs occlusion 1.006 1.0
H3 VDAC1 vs VDAC2 selectivity 0.922 1.0
H4 GSH depletion assay 0.728 1.0

Chronic Dosing Hypotheses

ID Description Effect Size (d) Power
H1 GSH synthesis rate healthy vs NAFLD 1.232 1.0
H2 VDAC1 gating recovery after washout 1.059 1.0
H3 14-day chronic dosing GSH stability 1.349 1.0
H4 7-OH-CBD accumulation & VDAC effects 0.851 1.0

Structural Pattern: Dose-Dependent Isomorphism

The CBD two-pathway model belongs to a structural pattern discovered across 5 independent IRIS Gate Evo runs spanning different molecules:

Molecule Pattern Key Variable
CBD Dose picks pathway; GSH determines outcome VDAC occupancy vs GSH capacity
Lithium Low dose neuroprotective; high dose nephrotoxic GSK-3beta inhibition level
THC Low CB1 occupancy therapeutic; high occupancy harmful G-protein vs beta-arrestin bias

Same abstract structure, discovered independently. The molecule is a stress test; the dose picks the pathway; the tissue context determines the outcome.

How to Reproduce

# Clone the IRIS Gate Evo engine
git clone https://github.com/templetwo/iris-gate-evo.git

# Run data is archived in the findings repo
git clone https://github.com/templetwo/iris-evo-findings.git

# View the specific run data
ls iris-evo-findings/runs/evo_20260213_035958_pharmacology/  # EFSA stress test
ls iris-evo-findings/runs/evo_20260213_042930_pharmacology/  # Chronic dosing

# View gold extractions
cat iris-evo-findings/gold/efsa_vdac1_stress_test.md
cat iris-evo-findings/gold/chronic_dosing_gsh_dynamics.md

# Run cross-run convergence analysis
cd iris-gate-evo
python cross_run.py --all