[Grok retrieval] P3 projective residual: document ratio-testing, novelty, PRE fit

[LightChainr]

Parent: #650. Retrieval-only. No Monte Carlo, no rescoring, no Huawei.

Why

P3 is the complete draft at docs/manuscripts/p3-projective-inference/ (on main: manuscript.md, README.md, tables.md). Target printed in the draft: Physical Review E. The statistic is the covariance-weighted distance from a measured amplitude vector to a model ray/subspace; on two entries it is exactly Fieller's z². A prior Grok venue note exists. This ticket is an independent literature retrieval that must confirm or contradict that note with primary sources.

The draft's only marked hole is reference [6]: it claims ratio-testing of aspect-ratio amplitudes is standard in the conformal-invariance finite-size literature, from familiarity, not from a primary reading. That claim is load-bearing. If you cannot document it, the "we correct a literature practice" framing collapses.

Deliverable

One notes file, draft PR against main:

notes/venue-retrieval-p3-YYYYMMDD.md

Do not edit the P3 manuscript's scientific claims. You may list the exact bibliographic replacements for [6] as a proposed patch (citation keys + quoted sentences), but do not silently rewrite §8. No docs/STATUS.md. Do not close this ticket or #579. Do not merge.

What to read first (in the clone)

1. docs/manuscripts/p3-projective-inference/manuscript.md — abstract, §1, §2.3 Fieller identity, §4 application, §5 established/not, §8 related work.
2. docs/manuscripts/p3-projective-inference/README.md — open items 1–3.
3. Do not rerun scripts/projective_inference.py except as a no-op read. Do not reopen the N=580 ladder.

Retrieval questions

Q1. The [6] hole (highest priority). Find at least three primary papers that form a ratio of two measured finite-size amplitudes (aspect ratio, boundary condition, or lattice-family) and test it against a conformal/FSS predicted ratio. Cardy 1984/1986 is a theory paper; it is not enough by itself. Prefer percolation or Ising/Potts wrapping or crossing-amplitude papers (Ziff, Watts, Simmons, etc.) if they actually ratio-test. For each: quote the sentence where they form Y/X or equivalent, with venue and year. If you cannot find three, say the practice is not documented and downgrade the paper's "corrects a literature" claim.

Q2. Is the statistic new? Fieller 1954, Gleser–Hwang 1987, Rao–Mitra generalized inverses, multivariate GLS residual / Hotelling-style distance to a subspace. Find whether anyone in FSS/percolation already uses a projective/χ² distance to a ray instead of a ratio. If yes, the draft overclaims novelty. If no, the novelty is the recommendation plus a worked inversion of a scientific verdict, not the linear algebra.

Q3. Annihilating functionals / divided differences as class-level tests of amplitude ladders. Known in this literature, or only in statistics/econometrics?

Q4. Venue fit, independently.

• Physical Review E (statistical physics / computational methods)
• JSTAT / J. Stat. Phys.
• Computer Physics Communications
• The American Statistician / other statistics journals
• J. Phys. A

For each: fit / stretch / wrong. Cite the journal's about/scope sentence from the publisher page, and one analogue paper if you find one.

Q5. Application weakness. The worked example is one frozen three-rung ladder at N=580 with a missing cov(r2,r4) and an undetermined bare_aspect_ratio verdict. Would PRE accept a methods paper whose physics example cannot name a surviving law? Find PRE methods papers that similarly reanalyse one dataset and invert a conclusion.

Prior Grok assessment to stress-test (do not copy)

1. PRE is the right first venue; the amplitude-ladder literature lives there.
2. The math is not new; novelty is diagnosing a design pathology (nominating a denominator inherits that rung's conditioning; dropping a distrusted rung removes the only check of the systematic bound).
3. Without 3–4 primary ratio-testing citations, the literature-practice claim dies.
4. Statistics journals will call this a note, not a methods paper (no simulation study, Fieller generalization too small).
5. Do not claim a percolation-threshold or modular-weight result.

Agree / disagree / refine with sources.

Claim boundary

• No claim about p_c. No claim that any of the eight competitor rays is the law.
• Mark every source PRIMARY_TEXT_READ / ABSTRACT_ONLY / [LIT].
• If arXiv is blocked, record it; use publisher HTML, Semantic Scholar, OpenAlex. Do not invent quotations.

Tripwire

If §8's "ratio-testing is ubiquitous" is still unsupported by three fetched primary quotations, the note must weaken that sentence rather than repeat it. A PR that "confirms" Grok while leaving [6] as [LIT] is incomplete.

Wait for ASSIGNED_MACHINE.

[LightChainr]

ASSIGNED_MACHINE: local
BASE_BRANCH: main

Retrieval/notes only. Do not power-on Huawei. If a step looks like NEED_HUAWEI, comment and stop — this ticket should never need it.

Do not merge. Do not edit docs/STATUS.md. Do not close this issue.

[LightChainr]

Retrieval complete: draft PR #701 (branch notes/venue-retrieval-p3-20260909, file notes/venue-retrieval-p3-20260909.md). Retrieval/notes only — nothing merged, nothing closed, no STATUS/ROADMAP edits, no reruns.

Headline findings:

• Q1, the [6] hole: three primary ratio-testing papers documented with verbatim quotations — Simmons, Kleban & Ziff (2007, J. Phys. A 40, F771: measured multiple-crossing fraction 0.07018 vs predicted ratio 0.069189 of integrated conformal densities, extrapolated in 1/L), Izmailian & Yeh (2009, Nucl. Phys. B 814, 573: BC amplitude ratios vs Cardy's perturbed-CFT scenario; exact amplitudes, not MC), and Kamieniarz & Blöte (1993, J. Phys. A 26, 201; ABSTRACT_ONLY — universal moment ratio from finite-size transfer-matrix data incl. arbitrary aspect ratios). However, LPS 1994 and Watts 1996 (both PRIMARY_TEXT_READ) show the dominant practice is full-curve comparison against the conformal formula, not ratio-testing — so the tripwire fired: the note weakens §8's "ubiquitous" sentence rather than repeating it, and proposes exact bibliographic replacements for [6]. Pinson (1994) was attempted and is blocked at Springer (recorded in §8); Langlands et al. (1992) and Pinson are listed as the next primaries to read.
• Q2, novelty: the statistic as a χ² test survives as new to FSS, but narrowly — Parisen Toldin (2011, PRE 84, 025703(R)) already uses a covariance scalar product with an orthogonal/parallel decomposition inside FSS (for error optimization, not hypothesis testing), and von Luxburg & Franz (2009) already generalize Fieller geometrically in statistics. Inspire HEP has 0 hits for Fieller confidence intervals; the standard data-collapse tool (pyfssa) uses a covariance-free statistic. Quotes verified.
• Q3: honest null — no divided-difference/annihilating-functional class tests found in FSS; the draft's "we have not seen" phrasing stands.
• Q4: PRE fit (scope sentence quoted from journals.aps.org/pre/about); J. Phys. A and JSTAT fit; CPC and The American Statistician stretch; J. Stat. Phys. mild stretch (Springer scope page bot-blocked, fallback noted).
• Q5: PRE does accept verdict-inverting reanalyses of a single dataset — Wang et al. (2013, PRE 87, 052107), Mertens (2022, J. Phys. A), and Jensen & Ziff (2006, PRE 74, 020101(R)), each with a verbatim revision sentence in the note.

Every source is marked PRIMARY_TEXT_READ / ABSTRACT_ONLY / [LIT] / API_QUERY; all quotations were read in fetched documents. The prior Grok assessment is stress-tested in §6 of the note (agree/refine per point), not copied.