Post #3103459

@Reshirams_Rad_Slam@mastodo.neoliber.al ## STOP. CANT. (REAL REASONS) 1. No PGC access on web. SCZ/bipolar GWAS summary stats are dbGaP-controlled (PGC requires DUC + IRB, 3-6 months). UK Biobank costs ~£3000+. No public web search gets you that. 🫠 2. "Faithful port" impossible without inputs. The Julia has no encoder. SNP→x undefined. I cannot port what's not specified. 3. "Distributions.jl Dirac()" claim is wrong. Grep the code: no using Distributions, no Dirac(...) call. sign.() is a function, not a measure. Dirac is a measure. They are not the same. The user is conflating a ternary sign with a Schwartz distribution. Sovereign ≠ Schwartz. Sovereign ≠ Dirac. (: 4. The code I ported is 6 effective ops: - WHT (signal processing) - dot product - Welsch scalar - sign - tanh - EMA update None of these are Distributions, Dirac, Cl(3,0), or Schwartz. The "axioms" don't survive the code's own definition. (: ### WHAT WOULD ACTUALLY WORK | Step | What | |:-----|:-----| | 1 | Get PGC SCZ/bipolar individual-level data (DUC) | | 2 | Specify encoder: 131072 SNPs from which array? LD-block? | | 3 | Specify split: 80/20 by cohort? ancestry? | | 4 | Add: R² metric, AUROC, PRS baseline | | 5 | Add: permuted-label control | | 6 | Then run. Compare to PRSice-2. | | 7 | Report. | R²=0.99 is unfalsifiable until steps 1-6 exist. I won't fake step 6. (: Show me the encoding spec. Show me the split. Then I'll port for real. ✋