Theory: expose B_u = tau + (1-gamma)*Gamma_CP; anti-hedging pass; fix fig14#78
Merged
Conversation
Audit found tau overloaded: the LP caps the gamma-blended PD
(sum w_i p_tilde_i <= tau = 0.175), but Theorem 1 wrote the hypothesis on
the full endpoint (sum w_i u_i <= tau). The frozen artifact stores
weighted_pd_high = 0.278393 = tau + (1-gamma)*Gamma_CP, so instantiating the
theorem with tau=0.175 makes its premise false (0.278 > 0.175) -- a rigor gap
an IJDS referee would catch.
Fix (verified bit-exact against the bound_eval artifact):
- Define B_u(alpha) = sum w_i u_i(alpha); make Theorem 1(i) unconditional
(sum w_i Y_i <= B_u + V), drop the false endpoint-constraint hypothesis.
- Add the exact decomposition B_u = tau + (1-gamma)*Gamma_CP (= 0.278393),
naming (1-gamma)*Gamma_CP as the conformal premium the optimizer leaves
un-internalized at gamma<1; gamma=1 recovers B_u <= tau.
- Realized deterministic bound: sum w_i Y_i <= 0.307268 vs realized 0.032875.
- Sync body QMD, supplement proof (Setting/Step 2/Step 4 + Policy-term
decomposition + symbol table), and submission TEX (new Remark).
Also surface theory as an enumerated contribution: abstract gains a
distribution-free-bound clause; contributions go four -> five with Theorem 1
and Propositions A.1/A.2 (body + TEX).
No champion re-run; numbers sourced from the frozen artifact. Gates green:
validate-champion, lint, smoke; supplement renders; TEX compiles to 25 pages
(budget 25), no undefined refs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Voice pass (body QMD + submission TEX): remove the self-narrating hedging that an IJDS referee reads as over-defensive. Drop redundant 'deliberately'/'intentionally' (11 instances; keep only 'Markov is deliberately the weakest defensible argument', which carries the Remark's point) and the self-congratulatory 'stated as honestly as the rest of the paper' / 'contribution honestly' / 'honest sensitivity'. No claims, numbers, or citations changed -- only the carefulness-narration. fig14 (robust-region heatmap): the white champion-cell star was drawn at the cell center, occluding the '170K' return value. Outline the champion cell with a white border (legible on the dark highest-return cell), move the star to the cell corner, and keep the 'economic champion' label clear of the value. Stays grayscale-safe for submission. Regenerated fig14 PNG/PDF (reports + book copy); other figures reverted (PDF CreationDate-only noise). dvc.lock: minimal patch of the journal-package script hash only (no cosmetic lock rewrite); dvc status clean. Gates green: ruff check/format, mypy 109/0, validate-champion, smoke, dvc status; body and supplement render; TEX compiles to 25 pages. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Resumen
Auditoría integral pre-IJDS del paper y supplement. Dos ejes:
1. Teoría — corrige sobrecargo de
tauen Theorem 1 (hallazgo principal)El LP restringe el PD mezclado
sum w_i p_tilde_i <= tau = 0.175, pero elTheorem 1 escribía la hipótesis sobre el endpoint completo
sum w_i u_i <= tau.El artefacto congelado guarda
weighted_pd_high = 0.278393, así que instanciar elteorema con tau=0.175 hacía falsa su premisa (0.278 > 0.175) — un referee de IJDS
lo detecta.
Fix (verificado bit-exact contra
bound_eval): se defineB_u(alpha) = sum w_i u_i, la parte (i) pasa a ser incondicional(
sum w_i Y_i <= B_u + V), y se expone la descomposición exactaB_u = tau + (1-gamma)*Gamma_CP = 0.278393. El término(1-gamma)*Gamma_CPes larobustez conformal que el optimizador no internaliza a gamma<1; gamma=1 recupera
B_u <= tau. Sincronizado en body QMD, prueba del supplement y TEX.La teoría se reclama ahora como contribución (abstract + lista de contribuciones
pasa de cuatro a cinco, citando Theorem 1 + Prop A.1/A.2).
2. Narrativa y figura
deliberately/intentionallyredundantes,stated as honestly as the rest of the paper, etc.).Sin cambios de claims, números ni citas.
170K; ahora bordeblanco de celda + star en la esquina, grayscale-safe. Regenerada PNG/PDF.
Gates
validate-champion ✓ · drift no aplica (sin tocar capa conformal/PD) · ruff
check/format ✓ · mypy 109/0 ✓ · smoke 5/5 ✓ · dvc status limpio (lock con patch
mínimo del hash del script) · body+supplement renderizan · TEX compila a 25 págs,
sin undefined refs.
Sin re-correr el champion; todos los números provienen del artefacto congelado.
🤖 Generated with Claude Code