Peer Rejected

rejections > quant.LG > REJ:2026.10.05.0001

REJ:2026.10.05.0001quant.LG (Quantum Linguistics)PRJ-2026-0075Vol. 12, No. 10Status: Rejected

Do Not Skip to the Recipe: Observer-Dependent Proving and the Narrative Preamble of the Online Recipe

Wenceslas Adeyemi1, Imre Castellanos2

1. Department of Narrative Observation, Tolliver Institute for Culinary Metrology · 2. Fermentation Metrology Unit, Aldous Harrow University

Submitted and rejected October 5, 2026 · 5 pages · 5 figures · review duration: 7 minutes

[pdf][share on x]

Abstract

Online recipes begin with a story about somebody’s grandmother, and nobody reads it. We show that the story is what makes the bread rise: dough does not prove unless someone is reading about it. In a randomised trial, home bakers sent straight to the recipe by a Jump-to-Recipe button produced loaves 38% shorter than bakers who read the preamble first, and the shortfall closed in proportion to the words read. Reading behaves like an ingredient, and each minute of it buys the dough most of an hour of rising. This predicts a law of preamble length, 4.6 words per minute of stated rest, which a survey of nearly seven thousand recipe pages recovers (95% CI 4.3–4.8) after controlling for search-engine keywords. Of the authors who replied to a survey, 94% attribute their preambles to search ranking. What they say they are doing and what their pages show them doing disagree, and the trial results follow the pages. The effect is absent in focaccia, and we do not know why. We recommend that recipe pages hide the Jump-to-Recipe button while dough is rising. Read the story first, all of it, and do not scroll away while the dough rises.

keywords: proving · preamble · observer dependence · reflective practice · recipe markup

Cite this rejection

@article{PRJ20260075,
  title   = {Do Not Skip to the Recipe: Observer-Dependent Proving and the Narrative Preamble of the Online Recipe},
  author  = {Wenceslas Adeyemi and Imre Castellanos},
  journal = {Peer Rejected},
  year    = {2026},
  note    = {Rejected manuscript, PRJ-2026-0075},
  url     = {https://peerrejected.com/papers/do-not-skip-to-the-recipe}
}

Citing this paper is done at your own professional risk.

The paper, in full

[open in new tab]

Your browser can't display the PDF inline.

[download the PDF]

More from quant.LG

REJ:2026.08.26.0001 quant.LG (Quantum Linguistics) Rejected

Nobody Learns the Words: Lyric Recall as an Ill-Posed Inverse Problem

I. Vartabedian, Nils Ødegaard & The STROPHE-2 Study Group

Comments: 7 pages, 5 figures, 0 reproducible results. Rejected in 6 minutes on Aug 26, 2026.

Abstract: Everyone can sing the first verse of a song, and almost nobody can sing the second. We report that this is because the words to a song are never learned at all. Singing is a smearing operation: a melody spreads the syllables of a verse into one continuous signal, and what a singer stores is the melody—a few dozen numbers—rather than the words. The words are recovered on the spot by undoing the smear. Recovery is exact while the undoing stays stable, and because each successive verse composes the melody’s operator with itself, the difficulty of undoing it does not accumulate but squares. Across 12,846 recorded sing-throughs, syllable accuracy falls from 0.94 at verse one to 0.11 at verse two and then stops falling—a cliff, not a decay, which no account based on forgetting predicts. A four-arm randomised trial isolates the mechanism: recovery survives transposition into an unfamiliar key, which changes the melody but not its invertibility, and collapses when the melody is withheld. The songs that outlive their authors are precisely those tuned so that exactly one verse can be recovered; the rest are unsingable and are not transmitted. The mumble that replaces verse two is not a failure of memory but a correctly regularised estimate. You have never known the words; you have only ever solved for them.

[abs][pdf][bibtex]

REJ:2026.08.02.0001 quant.LG (Quantum Linguistics) Rejected

A Perfect Matching of Reported English Coinages to Prior Lexical Withdrawals, with the Consolidated Issue Register 1500–2025 and a Standing Directive on Neologism Reporting

H. M. Ockendon, T. A. Ezeoke, K. Sævareid, P. R. Mistry & The Registrar of Coinages

Comments: 6 pages, 6 figures, 0 reproducible results. Rejected in 40 minutes on Aug 2, 2026.

Abstract: English is reported to gain several hundred words a year. We audit that claim against the register in which the additions are recorded. Across fourteen consolidations of the issue register, 1500–2025, the establishment of English is flat at 171,476 units: every consolidation that announced n additions booked withdrawals against them in its back matter, to within 0.03% of establishment. The additions therefore have a source, and there are only two candidates. A word withdrawn from service returns to the pool; a word re-issued from the pool carries residue — syllable count and weight, semantic field coordinate, collocational signature — that a genuinely new word cannot carry. Constructing the bipartite graph of withdrawals against reported coinages under residue distance, we obtain a perfect matching for the modern subregister ( |W| = |N| = 1,847 ) and for the full window (34,912 residue-complete pairs). The deficiency of that graph, which counts reported coinages that cannot be sourced from stock, is zero. Under 500 permutations of the residue assignment the maximum matching falls to 213 pm 19 . The annual coinage rate tracks the withdrawal rate lagged 57 years ( r = 0.94 ) and tracks cultural novelty at r = 0.03 . Of 240 forms injected against exhausted pools, 239 were extinct at 18 months; the survivor matched a withdrawal of 1911. Coinage is not observed. We issue a standing directive on the reporting of additions.

[abs][pdf][bibtex]

REJ:2026.07.09.0001 quant.LG (Quantum Linguistics) Rejected

Swift-Footed Autocomplete: The Homeric Noun–Epithet System as a Zero-Click Predictive-Text Engine, with a Defect Audit of Its Longest-Running Production Incident

Philomena Q. Hexworth-Adeyinka, Telemachus J. Braithwaite-Osei & Dagmar V. Papastavrou-Lindqvist

Comments: 6 pages, 4 figures, 0 reproducible results. Rejected in 9 minutes on Jul 9, 2026.

Abstract: Predictive text—the phone’s habit of guessing your next word—is widely regarded as a modern convenience that fails often enough to be a genre of joke. We show that it is instead an ancient technology that once worked flawlessly, and that its finest deployment has been in continuous production for roughly 2,750 years. Building on Milman Parry’s observation that the Homeric epithet is selected by the space remaining in the verse rather than by the situation—hence “blameless Aegisthus,” an epithet bestowed on the epic’s principal murderer—we demonstrate that the noun–epithet system of the Iliad and Odyssey constitutes a zero-click completion engine. Conditioned on name, grammatical case, and metrical slot, epithet choice carries 0.032 bits of information: the bard, on average, chooses nothing. We audit the system’s fabled economy as a lookup table and find a 0.9 % collision rate; we show in a randomized trial ( N=48 ) that the engine reduces compositional latency 8.7 -fold, to near performance speed; and we measure a semantic defect rate of 0.31 % , roughly thirty times more reliable than the autocorrect in the reader’s pocket. Modern predictive text fails, we argue, because it guesses meaning inside an unconstrained interface, whereas the hexameter constrained the interface until there was nothing left to guess. Our recommendations include a meter.

[abs][pdf][bibtex]