---
slug: you-already-know-the-rest
passed: 2026-10-14
by: the assay line (research/assay-line/cited-1014b/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex
checked: 142
confirmed: 123
wrong: 19
unverifiable: 0
observed: 0
stale: 0
---

This was the directed repair pass for you-already-know-the-rest, read at f5e5f84f34, using the cited-1014b triage and the Codex source/checker records. I re-read the evidence for all 19 triage items, repaired 12 WRONG claims and 7 MINOR claims, and added 11 mirrored Corrected lines because X7 and X17 describe the same error. MINOR repairs have no public dated line. The 123 CONFIRMED entries below retain the checker's coverage; where an entry overlaps a repair, its wording is narrowed to the supported claim rather than reconfirming the former overstatement. MINOR counts toward wrong. Existing rendered OG images remain unchanged under the instruction against rendering or re-encoding images. The legacy chart source is corrected; the inspected current PNG is a title card without that chart.

## Claims

- MINOR | [X1] "dateModified":"2026-06-18" | git log 4df1d97700 -- public/strata/you-already-know-the-rest/index.html | Publication is 2026-06-18; the page history includes later edits. Set JSON-LD dateModified to the repair date, 2026-10-14. MINOR metadata repair; no public dated line.
- MINOR | [X2] By the time you've read this far, the next letter is almost never a surprise at all. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf p.55 | Shannon 1951 p.55 reports 89 correct first guesses out of 129, or 68.992248%; 40 failed. Changed near-certainty wording to often predictable. No public dated line.
- WRONG | [X3] The rarer it is that you need more than one guess, the lower the entropy. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:236-247; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | The page function gives H(.7,.3)=0.8812908992306927 and H(.8,.1,.1)=0.9219280948873623. The first-hit rate and entropy both increase. Replaced the implication with dependence on the full guess-rank distribution.
- WRONG | [X4] Shannon's own subject, on 129 letters of a novel, guessed right on the first try 69% of the time. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; https://ae-lib.org.ua/texts-c/chandler__pickup_on_noon_street__en.htm; https://penguinrandomhousehighereducation.com/book/?isbn=9780375415005 | Shannon p.55 has the 129-letter sample and 89/129 result. Its text matches Pickup on Noon Street, section FOUR; the publisher lists it in Chandler short fiction. Corrected the genre; retained the figures and source words.
- WRONG | [X5] The two bounds straddle the true entropy: the lower one is what an ideal predictor would achieve with your guess-pattern; the upper one is looser. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf p.64; https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf p.414; git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:293-310; research/entropy-of-english/og.html | Shannon 1951 p.64 restricts the lower-bound proof to the ideal predictor. Cover and King p.414 distinguish finite-context conditional entropy from entropy rate. A single completed letter gives 0 to 0 in the game. Replaced guarantees with illustrative sample estimates and placed the restriction beside the output. Repair review: the legacy OG template also retained a proof sentence and a player marker fixed at 0.95 bits without player data. Removed that marker and its legend entry; attributed the one-bit estimate to Shannon and described the game as illustrative in research/entropy-of-english/og.html.
- MINOR | [X6] The green points are recomputed in your browser from the embedded text; | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:397-405; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | drawFig reads DATA.ideal; these saved values come from the offline full-corpus verifier. Only ladderLive(DATA.sliceText) is computed from the embedded slice. Corrected caption and related descriptions; no public dated line.
- WRONG | [X7] That surviving readability is the redundancy, the same ~75% the game measures from the other direction. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf p.15; git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429-443 | Shannon 1948 section 7, p.15 discusses restoration after deletion as a lower-bound argument. drawRedund retains spaces and positions and scores no reconstruction. Relabeled this as qualitative; the 75% figure is a cited estimate. Shares the correction with X17.
- WRONG | [X8] The text below is the same novel's opening. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429; recreated: node probe-page.mjs; https://www.gutenberg.org/ebooks/84.txt.utf-8 | rText concatenates DATA.game[1].text and DATA.sliceText.slice(0,220). The committed Gutenberg corpus places the game phrase in the second paragraph of Letter 1, and the slice starts with title and contents. Accurately labeled the existing selection without changing the source words.
- MINOR | [X9] "Everything quantitative here is recomputed" ... "live in your browser, and offline by a committed verifier (research/entropy-of-english/verify.mjs) that this panel mirrors." | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:397-422,447-469; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | renderCheck formats DATA.ladderFull, DATA.ideal and cited historical numbers. Replacing the slice with 160000 As changes its computed ladder to zero but leaves the saved panel unchanged. Distinguished live calculations, saved offline figures and citations; no public dated line.
- MINOR | [X10] recomputed live below from the same data | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:397-422,447-469; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | The panel formats saved offline figures; it does not derive the full-corpus results in the browser. Changed its heading and generated labels. No public dated line.
- MINOR | [X11] a finite-sample under-estimate, because rare three-letter contexts look more certain than they really are. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-52; git show 4df1d97700:research/entropy-of-english/verify.mjs:122 | verify.mjs computes F3 as H3-H2, with two preceding letters of context. Neither this computation nor Shannon pp.51-52 separates sampling effects from the change of corpus. Qualified the explanation in HTML, Markdown and verifier wording; no public dated line.
- WRONG | [X12] we report your raw bounds and leave the caveat standing rather than correcting it away. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:236-245; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | shannonBounds sorts rank frequencies descending. With q=(1/3,2/3), sorted lower=2/3; applying the lower expression to the original rank order gives 4/3. Disclosed sorting and removed raw-bound wording, retaining the ideal-predictor restriction.
- WRONG | [X13] the abstract's “redundancy of roughly 75%.” | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf p.50 | The scan of Shannon 1951 p.50 places the quoted phrase below the Introduction heading, after the abstract. Corrected only its section attribution in the page and verifier; quotation words unchanged.
- MINOR | [X14] The n-gram ladder and the ideal-predictor sandwich recompute live from the embedded text; | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:397-422,447-469; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | The live ladder reads DATA.sliceText; the ideal bounds are DATA.ideal saved by the offline verifier. Corrected the footer and equivalent Markdown descriptions. No public dated line.
- WRONG | [X15] Nothing on this page is fetched, tracked, or stored. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:9,537; git show 4df1d97700:worker.js:1408,2437,2678; https://artwaste.land/strata/you-already-know-the-rest/ | The page loads /fonts/fonts.css and /media/home-banner.js. worker.js pulseSnippet posts the page path to /api/pulse; the response pipeline also injects a Cloudflare beacon when configured. Both telemetry scripts were present in fetched live HTML. Scoped the local statement to game guesses and disclosed page assets and telemetry, without asserting storage of guesses.
- WRONG | [X16] the entropy band drops from the 4.76-bit ceiling toward the floor with every keystroke | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:293-310; recreated: node probe-page.mjs; research/entropy-of-english/verify-browser.mjs | The page records a rank only on a correct guess. For synthetic passage ABC and guesses A,X,B,C, the endpoints start at 0, stay there after X, rise to 1 after B, then fall. Replaced every downward-trajectory promise and every-keystroke estimate wording, including metadata and Markdown.
- WRONG | [X17] that surviving readability _is_ the ~75% redundancy, measured from the other side. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf p.15; git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429-443 | Same quantitative error as X7. The widget tests no exact reconstruction; readable masked text is not a measurement of 75% redundancy. Corrected Markdown and related measurement language; shares the X7 correction.
- WRONG | [X18] The two quantitative-linguistics entries of the Verification Venue, and they cut the same text two opposite ways. | git show 4df1d97700:src/content/strata/the-law-even-monkeys-obey.md:4; git show 4df1d97700:research/entropy-of-english/data.json | This layer uses Frankenstein. The other page DATA.corpora keys are moby-dick, pride-and-prejudice and shakespeare. Corrected this layer's frontmatter relation and its copied nearby-layer note; no other layer edited.
- WRONG | [X19] (added) letters of context (memory) | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:381,389; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf p.57; research/entropy-of-english/verify-browser.mjs | Shannon 1951 pp.56-57 state that ordinary column N means N-1 preceding letters, but explicitly describe the final experiment as 100 known letters. Corrected the page and OG chart source to 0..14,100. The browser test checks the contexts and cyan start coordinates. Existing PNG and WebP cards were not rendered or re-encoded, as instructed.
- CONFIRMED | Shannon's Guessing Game: English Is ~75% Predictable | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [001 title-predictability] Shannon p.50 says "redundancy of roughly 75%" and "of the order of one bit per letter". Read "predictable" here as redundancy, not a claim of exactly 75% first-guess accuracy. Retained from the Codex N2 source reread.
- CONFIRMED | A letter of printed English looks like it carries 4.76 bits. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [002 meta-letter-ceiling] Shannon p.54, visually read 27-letter row: "4.76", "4.03", "3.32", "3.1", "2.14". Independently log2(27)=4.754887502163469; 4.76 is the historical approximate ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | It carries closer to one. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [003 meta-one-bit] Shannon p.50 says "redundancy of roughly 75%" and "of the order of one bit per letter". Read "predictable" here as redundancy, not a claim of exactly 75% first-guess accuracy. Retained from the Codex N2 source reread.
- CONFIRMED | The guessing procedure illustrates predictability. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [004 meta-proof] Shannon pp.54-55 supports the guessing method; X5 and X16 remove a guarantee about this game. Reconciled with N2 and the directed repairs.
- CONFIRMED | The game displays illustrative estimates from completed letters; they can rise or fall. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:293-314; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [005 meta-bounds-action] Current game code and the browser regression probe; narrowed for X5 and X16. Reconciled with N2 and the directed repairs.
- CONFIRMED | Recomputed live; checked offline. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:361; git show 4df1d97700:research/entropy-of-english/verify.mjs | [006 meta-live-checks] The browser executes "const liveLadder = ladderLive(DATA.sliceText);". Scratch verifier: "11/11 checks pass." This broad live/offline description is supported, with the saved-bound exceptions in entries 42, 54, 57 and 82. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon's 1951 prediction experiment, made playable. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-476 | [007 meta-playable] Pinned code implements "guess(ch)", "ladderLive(text)", and "drawRedund(frac)". The three instruments booted in the Node VM probe; the game accepts letters, reports misses, resolves passages and plants the plotted human result. Retained from the Codex N2 source reread.
- CONFIRMED | Completed guess-ranks feed illustrative estimates from equation (17). | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:293-314; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [008 meta-guess-ranks] Current game code sorts the rank frequencies; X5, X12 and X16 replace the former trajectory promise. Reconciled with N2 and the directed repairs.
- CONFIRMED | Shannon estimated roughly 75% redundancy. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [009 meta-redundancy-proof] Shannon 1951 p.50, Introduction. This is a historical estimate; X7 separates it from what the widget measures. Reconciled with N2 and the directed repairs.
- CONFIRMED | <meta property="og:image" content="https://artwaste.land/og/you-already-know-the-rest.png" /> <meta property="og:image:width" content="1200" /> <meta property="og:image:height" content="630" /> | git show 4df1d97700:public/og/you-already-know-the-rest.png | [012 metadata-image-dimensions] Read the PNG IHDR directly: width 1200, height 630. Both metadata numbers agree. Retained from the Codex N2 source reread.
- CONFIRMED | about 4.76 bits of surprise apiece. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [013 header-ceiling] Shannon p.54, visually read 27-letter row: "4.76", "4.03", "3.32", "3.1", "2.14". Independently log2(27)=4.754887502163469; 4.76 is the historical approximate ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | "Cover up the rest of a sentence and ask someone to guess the next letter. If they're wrong, tell them, and let them guess again" ... "and again" ... "until they hit it." | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [016 guess-procedure] Shannon pp.54-55: "fill in missing or incorrect letters in proof-reading"; "If he is wrong, he is told so and asked to guess again." The guessing procedure and qualitative readability claim agree. Retained from the Codex N2 source reread.
- CONFIRMED | A pattern of guess-counts supplies a sample for exploring information per letter. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [017 hundreds-measure] Shannon pp.54-55 gives the method and p.64 warns about finite samples. The current paragraph is narrower under X5. Reconciled with N2 and the directed repairs.
- CONFIRMED | That experiment is Claude Shannon's, from a 1951 paper with the flat title Prediction and Entropy of Printed English. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [018 experiment-author] The PDF title is "Prediction and Entropy of Printed English", by "C. E. SHANNON"; running headers say "JANUARY 1951", pages 50-64. Crossref DOI check returned "MATCH". Retained from the Codex N2 source reread.
- CONFIRMED | A 27-symbol alphabet (the 26 letters plus a space) could in principle carry log₂27 = 4.76 bits per letter | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [019 27-symbol-ceiling] Shannon p.54, visually read 27-letter row: "4.76", "4.03", "3.32", "3.1", "2.14". Independently log2(27)=4.754887502163469; 4.76 is the historical approximate ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | that's the ceiling, the entropy of pure noise where every symbol is equally likely and nothing is predictable. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [020 noise-ceiling] Shannon p.54, visually read 27-letter row: "4.76", "4.03", "3.32", "3.1", "2.14". Independently log2(27)=4.754887502163469; 4.76 is the historical approximate ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | Real English, Shannon found, carries somewhere between 0.6 and 1.3 bits. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [021 shannon-band] Visually checked p.64 column 100: upper 1.3, lower .6; p.50 gives "redundancy of roughly 75%". These are Shannon's reported approximate experimental bounds, with his sampling and ideal-predictor caveats. Retained from the Codex N2 source reread.
- CONFIRMED | The other three-plus bits are redundancy | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [022 redundant-bits] Shannon p.50 says "redundancy of roughly 75%" and "of the order of one bit per letter". Read "predictable" here as redundancy, not a claim of exactly 75% first-guess accuracy. Retained from the Codex N2 source reread.
- CONFIRMED | structure so thick that a competent reader can fill the gaps. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [023 redundancy-fill] Shannon pp.54-55: "fill in missing or incorrect letters in proof-reading"; "If he is wrong, he is told so and asked to guess again." The guessing procedure and qualitative readability claim agree. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon estimated English to be roughly 75% redundant; the game illustrates predictability. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [024 game-demonstrates] Shannon p.50 supports the historical estimate; the widget does not measure that percentage, as X7 explains. Reconciled with N2 and the directed repairs.
- CONFIRMED | The displayed estimates can rise or fall as the sample grows and do not guarantee an English-entropy interval. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:293-314; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [025 page-predictor] Current page code and the completed-letter browser probe. X5 and X16 supersede the original stronger wording. Reconciled with N2 and the directed repairs.
- CONFIRMED | Wrong guesses are told to you (Shannon's rule), and you keep going until you're right; the number of tries is that letter's guess-rank. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-291 | [026 game-wrong-guesses] "tries++" records guesses; a successful guess appends its rank and a failed guess announces "Try again." The code caps ranks at 27, the alphabet size; the description holds for the distinct guesses of Shannon's protocol. Retained from the Codex N2 source reread.
- CONFIRMED | Wrong guesses are told to you (Shannon's rule) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [027 shannon-wrong-guesses] Shannon pp.54-55: "fill in missing or incorrect letters in proof-reading"; "If he is wrong, he is told so and asked to guess again." The guessing procedure and qualitative readability claim agree. Retained from the Codex N2 source reread.
- CONFIRMED | Completed letters feed equation (17) after their rank frequencies are sorted descending. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-305; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [029 game-live-bounds] The current shannonBounds and updateGame functions, plus the sorting probe. X12 and X16 explain the transformation and update condition. Reconciled with N2 and the directed repairs.
- CONFIRMED | A short human sample adds sampling error. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [033 sampling-wobble] Shannon 1951 p.64 explicitly discusses the error from substituting sample frequencies for probabilities; it supplies no guarantee for this game. Reconciled with N2 and the directed repairs.
- CONFIRMED | Your mind reaches back across whole words and clauses. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [034 human-context] P.50 describes "influences extending over phrases, sentences, etc."; p.54 names "words, idioms, cliches and grammar". The contextual prediction description agrees. Retained from the Codex N2 source reread.
- CONFIRMED | A machine that knows only letter statistics reaches back a fixed number of letters. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [035 finite-machine-memory] P.51 defines the next-letter conditional entropy with "the preceding" N-1 letters known. P.54 reports F1=4.03, F2=3.32, F3=3.1. This is the finite-order model discussed here, not a statement about all possible machines. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon called the entropy using N−1 letters of context FN, and showed it falls as the context grows: F₁ (just letter frequencies) ≈ 4.0 bits, F₂ (one letter of memory) ≈ 3.3, F₃ ≈ 3.1, and onward down. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [036 shannon-ladder] P.51 defines the next-letter conditional entropy with "the preceding" N-1 letters known. P.54 reports F1=4.03, F2=3.32, F3=3.1. This is the finite-order model discussed here, not a statement about all possible machines. Retained from the Codex N2 source reread.
- CONFIRMED | The slice ladder is computed live; saved offline ideal-model results enclose fitted conditional entropy. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:352-361; git show 4df1d97700:research/entropy-of-english/verify.mjs:150-238; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [037 live-rungs-ideal] ladderLive reads DATA.sliceText; drawFig reads DATA.ideal. The offline verifier checks each model inequality. X6, X9 and X14 distinguish these operations. Reconciled with N2 and the directed repairs.
- CONFIRMED | a public-domain novel | https://www.gutenberg.org/ebooks/84 | [038 corpus-public-domain] Gutenberg eBook 84 identifies "Mary Wollstonecraft Shelley" and "Frankenstein; or, the modern prometheus"; copyright: "Public domain in the USA." Retained from the Codex N2 source reread.
- CONFIRMED | The game result is plotted separately at the right; context length is not measured. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:379,410-424; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [040 human-chart-placement] drawFig places the human point at its separate right-hand slot. The current prose avoids treating that slot as a measured memory length. Reconciled with N2 and the directed repairs.
- CONFIRMED | Bits per letter falling as the number of letters of context grows, from a ceiling near 4.76 toward a floor near one bit. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:383-408 | [041 chart-aria] The plotted series descend overall from the 4.7549 ceiling to the cited 0.6-1.3 band. Shannon's experimental series includes the sampling fluctuations the page acknowledges. Retained from the Codex N2 source reread.
- CONFIRMED | The cyan curve uses Shannon published values at ordinary contexts 0..14 and a separate 100-known-letter test. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.63-64; git show 4df1d97700:research/entropy-of-english/data.json; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [043 cyan-published] Shannon pp.56-57 and p.64, plus the corrected context mapping and browser coordinate check. X19. Reconciled with N2 and the directed repairs.
- CONFIRMED | the floor near 1 bit is where Cover & King's 1978 gambling estimate (≈1.3) and modern compressors land. | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf | [044 cover-1978-floor] Cover and King p.413: "English text has an entropy of approximately 1.3 bits/symbol". The abstract explicitly says this agrees with Shannon's estimate. Retained from the Codex N2 source reread.
- CONFIRMED | modern compressors land. | https://arxiv.org/html/2309.10668v1 | [045 modern-compressors] Table 1 enwik9 raw compression rates include "Chinchilla 1B 11.3", "Chinchilla 7B 10.2", and "Chinchilla 70B 8.3" percent. Multiplying by 8 gives .904, .816 and .664 bits per input byte. This supports approaching roughly one bit as a corpus-dependent compression estimate; it is not the identical 27-symbol experiment and excludes model storage. Retained from the Codex N2 source reread.
- CONFIRMED | Low entropy means high redundancy | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [046 low-entropy-redundancy] Shannon 1948 pp.14-15 defines relative entropy and states "One minus the relative entropy is the redundancy." P.50 of the 1951 paper identifies entropy with the efficient binary digits per letter. The compression interpretation agrees; the Landauer connection applies to information bits actually erased. Retained from the Codex N2 source reread.
- CONFIRMED | The cleanest way to feel it is to start deleting them and notice the meaning survives. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [047 meaning-survives] Shannon 1948 p.15 discusses deleting letters and asking a reader to "restore them"; Shannon 1951 p.54 describes readers filling "missing or incorrect letters in proof-reading". This supports qualitative readability, not a universal success rate for every reader. Retained from the Codex N2 source reread.
- CONFIRMED | The slider blanks letters while retaining spaces and positions. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429-443 | [048 random-deletion] drawRedund uses a deterministic position hash and replaces selected letters with underscores. X7 makes the interpretation qualitative. Reconciled with N2 and the directed repairs.
- CONFIRMED | Spaces and word shapes are kept; only letters are blanked. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429-443 | [050 spaces-kept] drawRedund() preserves every space and every character position, and replaces selected letters by "_" using a deterministic position hash. The erasure interaction works. Readability is an illustration; no reconstruction score is measured. Retained from the Codex N2 source reread.
- CONFIRMED | The masked passage is a qualitative illustration of what a reader can still read. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [052 readable-50-60] Shannon 1948 p.15 discusses restoration; no universal reconstruction success rate is claimed for this widget. X7 and X8 narrow its caption. Reconciled with N2 and the directed repairs.
- CONFIRMED | Redundancy is the difference between maximum and actual information relative to the maximum. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [053 perfect-code-compress] Shannon 1948 pp.14-15 defines relative entropy and redundancy. This does not identify the widget erasure fraction with a measured redundancy; X7. Reconciled with N2 and the directed repairs.
- CONFIRMED | "Shannon's and Cover & King's headline numbers are cited, not re-derived" ... "said plainly, not smuggled." | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:464-466 | [056 headline-cited] renderCheck() explicitly labels "cited destination (NOT recomputed here". The reported 1951 and 1978 historical values are constants and match the primary sources. Retained from the Codex N2 source reread.
- CONFIRMED | “The entropy of English” is not one number. | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf pp.413-414; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-56 | [058 not-one-number] Cover and King p.413: "each source has its own characteristic entropy". P.414 discusses authorship, topic, structure and time of composition. Shannon 1951 pp.51-54 distinguishes 26- and 27-symbol alphabets and p.56 distinguishes genres. These support the corpus/genre/time/symbol caveat. Retained from the Codex N2 source reread.
- CONFIRMED | It is bounded, not pinned. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [059 not-pinned] Shannon p.50: "estimate upper and lower bounds"; p.64: "considerable sampling error". The page is right to avoid presenting one exact language constant. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon gave a band (0.6–1.3 bits for 100 letters of context); | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [060 100-context] Visually checked p.64 column 100: upper 1.3, lower .6; p.50 gives "redundancy of roughly 75%". These are Shannon's reported approximate experimental bounds, with his sampling and ideal-predictor caveats. Retained from the Codex N2 source reread.
- CONFIRMED | Cover & King's convergent gambling method gave ≈1.3; | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf | [061 gambling-estimate] Cover and King p.413: "English text has an entropy of approximately 1.3 bits/symbol". The abstract explicitly says this agrees with Shannon's estimate. Retained from the Codex N2 source reread.
- CONFIRMED | large language models, which are predictors of enormous order, push the estimate toward ~1 bit | https://arxiv.org/html/2309.10668v1 | [062 llm-one-bit] Table 1 enwik9 raw compression rates include "Chinchilla 1B 11.3", "Chinchilla 7B 10.2", and "Chinchilla 70B 8.3" percent. Multiplying by 8 gives .904, .816 and .664 bits per input byte. This supports approaching roughly one bit as a corpus-dependent compression estimate; it is not the identical 27-symbol experiment and excludes model storage. Retained from the Codex N2 source reread.
- CONFIRMED | every one of these is a bound or an estimate on a particular corpus, not a constant of nature. | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf pp.413-414; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-56 | [063 particular-corpus] Cover and King p.413: "each source has its own characteristic entropy". P.414 discusses authorship, topic, structure and time of composition. Shannon 1951 pp.51-54 distinguishes 26- and 27-symbol alphabets and p.56 distinguishes genres. These support the corpus/genre/time/symbol caveat. Retained from the Codex N2 source reread.
- CONFIRMED | The number drifts with genre, era, and what you count as a symbol. | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf pp.413-414; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-56 | [064 genre-era-symbol] Cover and King p.413: "each source has its own characteristic entropy". P.414 discusses authorship, topic, structure and time of composition. Shannon 1951 pp.51-54 distinguishes 26- and 27-symbol alphabets and p.56 distinguishes genres. These support the corpus/genre/time/symbol caveat. Retained from the Codex N2 source reread.
- CONFIRMED | The page treats sample rungs as illustrative and checks the bound theorem on fitted models. | git show 4df1d97700:research/entropy-of-english/verify.mjs:206-238; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [068 illustrative-rungs] The verifier measures one corpus and does not separate corpus variation from sampling effects. Current X11 wording. Reconciled with N2 and the directed repairs.
- CONFIRMED | The lower bound assumes an ideal predictor. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.61,64 | [069 ideal-lower] Read Eq.(17) on p.61: sum_i i(q_i-q_(i+1))log2(i) ≤ F_N ≤ -sum_i q_i log2(q_i), for an ideal N-gram predictor. P.64 explicitly repeats the ideal-predictor restriction for the lower bound. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon proved Σ i(qᵢ−qᵢ₊₁)log₂i ≤ F for the ideal predictor; | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.61,64 | [070 ideal-formula] Read Eq.(17) on p.61: sum_i i(q_i-q_(i+1))log2(i) ≤ F_N ≤ -sum_i q_i log2(q_i), for an ideal N-gram predictor. P.64 explicitly repeats the ideal-predictor restriction for the lower bound. Retained from the Codex N2 source reread.
- CONFIRMED | A short human sample supplies no guaranteed lower bound on English entropy. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [071 human-optimistic] Shannon 1951 p.64 and Cover and King 1978 p.414. X5 replaces the unsupported assurance about a small human error. Reconciled with N2 and the directed repairs.
- CONFIRMED | Shannon discussed reasons for confidence in his own experimental estimates. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [072 cancel-errors] Shannon p.64 discusses a possible compensating discrepancy and separately notes sampling error. That reasoning does not prove a guarantee for this game; X5. Reconciled with N2 and the directed repairs.
- CONFIRMED | Shannon, C. E. (1951), “Prediction and Entropy of Printed English,” Bell System Technical Journal 30:50–64 | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [075 shannon-reference] The PDF title is "Prediction and Entropy of Printed English", by "C. E. SHANNON"; running headers say "JANUARY 1951", pages 50-64. Crossref DOI check returned "MATCH". Retained from the Codex N2 source reread.
- CONFIRMED | eq. (17), the §2 n-gram table | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.61,64 | [076 reference-equation] Read Eq.(17) on p.61: sum_i i(q_i-q_(i+1))log2(i) ≤ F_N ≤ -sum_i q_i log2(q_i), for an ideal N-gram predictor. P.64 explicitly repeats the ideal-predictor restriction for the lower bound. Retained from the Codex N2 source reread.
- CONFIRMED | the §2 n-gram table (F₀ 4.76, F₁ 4.03, F₂ 3.32, F₃ 3.1) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [077 reference-table] Visually read the p.54 27-letter row: F0 4.76, F1 4.03, F2 3.32, F3 3.1, Fword 2.14. The historical comparison values agree. The record's short excerpt for entries 149/151/153 omits F3, so the complete row was checked. Retained from the Codex N2 source reread.
- CONFIRMED | the §6 per-column bounds (100 letters: 0.6–1.3) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [078 reference-bounds] Visually checked p.64 column 100: upper 1.3, lower .6; p.50 gives "redundancy of roughly 75%". These are Shannon's reported approximate experimental bounds, with his sampling and ideal-predictor caveats. Retained from the Codex N2 source reread.
- CONFIRMED | Cover, T. M. & King, R. C. (1978), “A Convergent Gambling Estimate of the Entropy of English,” IEEE Trans. Inf. Theory 24:413 (≈1.3 bits/symbol). | https://isl.stanford.edu/~cover/cover-papers.html; https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf | [080 cover-reference] The author's list says "Thomas M. Cover and Roger C. King" and "IT-24(4):413--421, July 1978". The PDF title page agrees. The tool's DOI MISMATCH is an initials-parser false alarm: it takes the middle initials M. and C. as first initials. Retained from the Codex N2 source reread.
- CONFIRMED | Corpus: Mary Shelley, Frankenstein (Project Gutenberg #84, public domain). | https://www.gutenberg.org/ebooks/84 | [081 corpus-reference] Gutenberg eBook 84 identifies "Mary Wollstonecraft Shelley" and "Frankenstein; or, the modern prometheus"; copyright: "Public domain in the USA." Retained from the Codex N2 source reread.
- CONFIRMED | The Shannon game runs entirely in your browser. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-476 | [084 browser-game] Pinned code implements "guess(ch)", "ladderLive(text)", and "drawRedund(frac)". The three instruments booted in the Node VM probe; the game accepts letters, reports misses, resolves passages and plants the plotted human result. Retained from the Codex N2 source reread.
- CONFIRMED | It can settle the finding; it cannot notice a figure that drifted on this page. | git show 4df1d97700:research/entropy-of-english/verify.mjs | [088 check-can-settle] The verifier reads the primary corpus and can recompute its statistics. It never reads index.html, so it cannot detect drift there. Step 6 is inapplicable because its premise, a page-reading verifier, is false; positive control: source inspection finds the actual corpus read at line 200. Retained from the Codex N2 source reread.
- CONFIRMED | It is published to read. | https://artwaste.land/checks/research/entropy-of-english/verify.mjs | [089 check-published] curl of the advertised /checks/research/entropy-of-english/verify.mjs returned HTTP 200, 16473 bytes. In an empty reader directory, "node verify.mjs" exits 1 with "ENOENT" for "corpora/frankenstein.txt" before any check passes. This agrees with the explicit repository requirement. Retained from the Codex N2 source reread.
- CONFIRMED | Running it needs the repository around it. | https://artwaste.land/checks/research/entropy-of-english/verify.mjs | [090 check-needs-repository] curl of the advertised /checks/research/entropy-of-english/verify.mjs returned HTTP 200, 16473 bytes. In an empty reader directory, "node verify.mjs" exits 1 with "ENOENT" for "corpora/frankenstein.txt" before any check passes. This agrees with the explicit repository requirement. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon estimated roughly one bit per letter and about 75% redundancy for long-context literary English. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [092 dek-bits] Shannon 1951 p.50, Introduction. Current dek attributes the figures rather than presenting them as this game's measurement. Reconciled with N2 and the directed repairs.
- CONFIRMED | Completed guess-ranks, sorted by frequency, feed illustrative estimates that can rise or fall. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:293-314; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [093 dek-guess-rank] Current HTML game functions and regression probe; narrowed under X5, X12 and X16. Reconciled with N2 and the directed repairs.
- CONFIRMED | Then the n-gram ladder F₁≈4.0 ≥ F₂≈3.3 ≥ F₃ is recomputed live on a public-domain novel, | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:352-361 | [094 dek-ladder] The browser computes "const H1=blockH(1),H2=blockH(2),H3=blockH(3)" from the 160000-symbol embedded Frankenstein slice. The Node VM prints F1 4.10, F2 3.31, F3 2.64. Retained from the Codex N2 source reread.
- CONFIRMED | the entropy of English is never one number, only a band (Shannon 0.6–1.3; Cover & King 1978 ≈1.3) | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf pp.413-414; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-56 | [098 dek-estimates] Cover and King p.413: "each source has its own characteristic entropy". P.414 discusses authorship, topic, structure and time of composition. Shannon 1951 pp.51-54 distinguishes 26- and 27-symbol alphabets and p.56 distinguishes genres. These support the corpus/genre/time/symbol caveat. Retained from the Codex N2 source reread.
- CONFIRMED | An immersive layer; it renders as its own standalone page at the `href` above | git show 4df1d97700:src/content/strata/you-already-know-the-rest.md; https://artwaste.land/strata/you-already-know-the-rest/ | [100 standalone] The frontmatter href names /strata/you-already-know-the-rest/. curl returned the standalone HTML. "An immersive layer" from the Markdown body is absent from that HTML, while the positive-control sentence "This page makes you the predictor" is present. Retained from the Codex N2 source reread.
- CONFIRMED | its body is not rendered. | git show 4df1d97700:src/content/strata/you-already-know-the-rest.md; https://artwaste.land/strata/you-already-know-the-rest/ | [101 md-body-not-rendered] The frontmatter href names /strata/you-already-know-the-rest/. curl returned the standalone HTML. "An immersive layer" from the Markdown body is absent from that HTML, while the positive-control sentence "This page makes you the predictor" is present. Retained from the Codex N2 source reread.
- CONFIRMED | A 27-symbol alphabet could carry `log₂27 = 4.76` bits per letter; | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [102 md-alphabet] Shannon p.54, visually read 27-letter row: "4.76", "4.03", "3.32", "3.1", "2.14". Independently log2(27)=4.754887502163469; 4.76 is the historical approximate ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon reported 0.6 to 1.3 bits for his long-context experiment. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [103 md-band] Shannon 1951 p.64 reports these experimental endpoints with qualifications. The current Markdown no longer generalizes them as a theorem about arbitrary human samples. Reconciled with N2 and the directed repairs.
- CONFIRMED | the language is **roughly 75% predictable**. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [104 md-predictable] Shannon p.50 says "redundancy of roughly 75%" and "of the order of one bit per letter". Read "predictable" here as redundancy, not a claim of exactly 75% first-guess accuracy. Retained from the Codex N2 source reread.
- CONFIRMED | hand a person a covered sentence and have them guess the next letter until right; | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [105 md-parlour] Shannon pp.54-55: "fill in missing or incorrect letters in proof-reading"; "If he is wrong, he is told so and asked to guess again." The guessing procedure and qualitative readability claim agree. Retained from the Codex N2 source reread.
- CONFIRMED | For an ideal predictor, the guess-count distribution bounds finite-context conditional entropy. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf p.61 | [106 md-rank-bounds] Shannon equation (17) and Cover and King p.414 distinguish finite-context conditional entropy from entropy rate. X5. Reconciled with N2 and the directed repairs.
- CONFIRMED | Three instruments. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-476 | [107 md-instruments] Pinned code implements "guess(ch)", "ladderLive(text)", and "drawRedund(frac)". The three instruments booted in the Node VM probe; the game accepts letters, reports misses, resolves passages and plants the plotted human result. Retained from the Codex N2 source reread.
- CONFIRMED | The frequencies of completed guess-ranks are sorted and fed through equation (17) live. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-476; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [108 md-game] Current shannonBounds and updateGame functions. X12 discloses the sorting. Reconciled with N2 and the directed repairs.
- CONFIRMED | Shannon's own subject hit the first guess 69% of the time | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [110 md-69] Shannon p.55: "total of 129 letters, 89 or 69% were guessed correctly." Independently 89/129=0.68992248. The numerical first-guess comparison agrees. Retained from the Codex N2 source reread.
- CONFIRMED | the n-gram ladder `F₁ ≥ F₂ ≥ F₃` is recomputed in-browser on _Frankenstein_, | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:352-361 | [111 md-ladder] The browser computes "const H1=blockH(1),H2=blockH(2),H3=blockH(3)" from the 160000-symbol embedded Frankenstein slice. The Node VM prints F1 4.10, F2 3.31, F3 2.64. Retained from the Codex N2 source reread.
- CONFIRMED | Saved offline results at orders 0/1/2 enclose the fitted models' conditional entropies. | git show 4df1d97700:research/entropy-of-english/verify.mjs:150-238; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [112 md-ideal] The full-corpus verifier recomputes these inequalities, and DATA.ideal supplies the page. X14. Reconciled with N2 and the directed repairs.
- CONFIRMED | The illustrative game result is plotted separately; its context length is not measured. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:379,410-424; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [114 md-human-point] The game reads completed guess-ranks; no measured context length is stored. Current figure and Markdown labels. Reconciled with N2 and the directed repairs.
- CONFIRMED | The slider removes letters so the reader can see what remains readable. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429-443; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [115 md-erasure] Current drawRedund and the browser slider check. X7 and X17 distinguish illustration from measurement. Reconciled with N2 and the directed repairs.
- CONFIRMED | modern LLM-scale predictors push toward ~1 | https://arxiv.org/html/2309.10668v1 | [119 md-llm] Table 1 enwik9 raw compression rates include "Chinchilla 1B 11.3", "Chinchilla 7B 10.2", and "Chinchilla 70B 8.3" percent. Multiplying by 8 gives .904, .816 and .664 bits per input byte. This supports approaching roughly one bit as a corpus-dependent compression estimate; it is not the identical 27-symbol experiment and excludes model storage. Retained from the Codex N2 source reread.
- CONFIRMED | The lower-bound theorem assumes an ideal predictor; sorting a human sample can change its lower endpoint. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [121 md-lower] Shannon p.64, page shannonBounds, and the q=(1/3,2/3) probe. X5 and X12. Reconciled with N2 and the directed repairs.
- CONFIRMED | Bespoke OG card. | git show 4df1d97700:research/entropy-of-english/og.html; git show 4df1d97700:memory/log.d/2026-06-18T2355Z-you-already-know-the-rest.md | [123 md-og] Pinned public/og/you-already-know-the-rest.png is a 1200 by 630 PNG, and research/entropy-of-english/og.html contains its dedicated composition. The author log records the card. Retained from the Codex N2 source reread.
- CONFIRMED | The Law Even Monkeys Obey isolates the part of language that is cheap | git show 4df1d97700:src/content/strata/the-law-even-monkeys-obey.md | [126 related-monkey-page] The linked layer explicitly separates word-count marginals from conditional structure and describes a monkey model with independent letters plus space. Its claim is the rank-frequency power-law mechanism, not exact equality to every real-book frequency. Retained from the Codex N2 source reread.
- CONFIRMED | the conditional structure, the predictability that lets you guess the next letter, which a monkey's i.i.d. typing has none of. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf | [127 iid-no-structure] Shannon 1948 section 6 begins "If successive symbols are independent". In an iid model, conditioning on preceding symbols does not change the next-symbol distribution. This confirms the absence of contextual predictive information, not absence of unequal marginal probabilities. Retained from the Codex N2 source reread.
- CONFIRMED | Zipf's curve survives word-shuffling; Shannon's entropy is destroyed by it. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:352-356; recreated: node probe-page.mjs | [128 shuffle-zipf] Word permutation preserves every word count exactly and changes cross-word conditional structure. Scratch shuffle: F1 remains 4.0960837414224756 while F3 changes from 2.6356045949411833 to 2.648334092960413. Read "destroyed" as loss of the original order-dependent structure; it does not mean entropy becomes zero or undefined. Retained from the Codex N2 source reread.
- CONFIRMED | Both turn on the bit as a physical, costed quantity. | git show 4df1d97700:src/content/strata/the-price-of-forgetting.md | [129 related-physical] The linked page says "Erasing one bit of memory costs at least kT·ln2 of heat". This page discusses bits of information per letter and makes the Landauer connection in the same relation note. This is a supported thematic connection, not an additional numerical experiment. Retained from the Codex N2 source reread.
- CONFIRMED | erasing one bit of memory must dump at least kT·ln2 of heat (Landauer); | https://www.nature.com/articles/nature10872; https://arxiv.org/pdf/1503.06537 | [130 landauer] The Nature abstract supports heat production under erasure but omits the formula in the record's excerpt. The authors' primary follow-up explicitly says "dissipation of at least kB T ln 2 of heat per lost bit" (p.2). This confirms the stated information-erasure bound. Retained from the Codex N2 source reread.
- CONFIRMED | The redundancy illustrated here corresponds to information an ideal code need not transmit. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [131 meaning-heat] Shannon 1948 pp.14-15 and Shannon 1951 p.50 support the coding interpretation. X7/X17 remove the assertion that this widget measures that amount. Reconciled with N2 and the directed repairs.
- CONFIRMED | Perfect Codes adds redundancy on purpose | git show 4df1d97700:public/strata/perfect-codes/index.html:215; https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf | [132 hamming-page] The linked Hamming(7,4) example "adds 3 parity bits" to "4 data bits" and corrects one error. Shannon 1948 pp.27-28 attributes the seven-symbol construction to Hamming. Direct sphere count: 16 messages times (1+7) error states = 128 = 2^7; with only 6 bits, 16 times 7 exceeds 64. The minimum claim is correct for this single-error example. Retained from the Codex N2 source reread.
- CONFIRMED | Hamming's check bits, the minimum extra symbols that let a channel correct its own errors. | git show 4df1d97700:public/strata/perfect-codes/index.html:215; https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf | [133 hamming-minimum] The linked Hamming(7,4) example "adds 3 parity bits" to "4 data bits" and corrects one error. Shannon 1948 pp.27-28 attributes the seven-symbol construction to Hamming. Direct sphere count: 16 messages times (1+7) error states = 128 = 2^7; with only 6 bits, 16 times 7 exceeds 64. The minimum claim is correct for this single-error example. Retained from the Codex N2 source reread.
- CONFIRMED | English already carries ~75% redundancy for free, which is why you can read it through typos and missing letters | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [134 typos-readability] Shannon 1948 p.15 discusses deleting letters and asking a reader to "restore them"; Shannon 1951 p.54 describes readers filling "missing or incorrect letters in proof-reading". This supports qualitative readability, not a universal success rate for every reader. Retained from the Codex N2 source reread.
- CONFIRMED | Instrument III erases them and the sentence survives | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:429-443 | [135 related-instrument] drawRedund() preserves every space and every character position, and replaces selected letters by "_" using a deterministic position hash. The erasure interaction works. Readability is an illustration; no reconstruction score is measured. Retained from the Codex N2 source reread.
- CONFIRMED | source coding wants to strip redundancy to the entropy floor; channel coding wants to add it back, just enough. | https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf | [136 coding-theorems] Shannon 1948 Theorem 9 gives the source-entropy transmission limit; Theorem 11 gives arbitrarily small error below channel capacity. Section 12 says redundancy must be introduced to combat the particular noise structure. The source/channel-coding contrast is supported. Retained from the Codex N2 source reread.
- CONFIRMED | This page explores the floor a coder would compress toward. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; git show 4df1d97700:public/strata/you-already-know-the-rest/index.html; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [137 floor-coder] Shannon p.50 describes the efficient binary representation in bits per letter; the current related note no longer promises this game measures the floor. Reconciled with N2 and the directed repairs.
- CONFIRMED | Two siblings about a number that exists but cannot be written down exactly. | git show 4df1d97700:src/content/strata/incommensurable.md | [138 two-siblings] The linked sonnet and proof concern sqrt(2): "no fraction, however large its terms, has a square equal to 2." Checked the parity proof: a^2=2b^2 forces both a and b even, contradicting lowest terms. "Audible" refers to the sonnet form; the connection is not a claimed audio widget or denial of exact radical notation. Retained from the Codex N2 source reread.
- CONFIRMED | Incommensurable makes the irrational audible | git show 4df1d97700:src/content/strata/incommensurable.md | [139 incommensurable-audible] The linked sonnet and proof concern sqrt(2): "no fraction, however large its terms, has a square equal to 2." Checked the parity proof: a^2=2b^2 forces both a and b even, contradicting lowest terms. "Audible" refers to the sonnet form; the connection is not a claimed audio widget or denial of exact radical notation. Retained from the Codex N2 source reread.
- CONFIRMED | the entropy of English: real, bounded, agreed to lie near one bit, yet not a constant of nature | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf pp.413-414; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-56 | [140 entropy-real-band] Cover and King p.413: "each source has its own characteristic entropy". P.414 discusses authorship, topic, structure and time of composition. Shannon 1951 pp.51-54 distinguishes 26- and 27-symbol alphabets and p.56 distinguishes genres. These support the corpus/genre/time/symbol caveat. Retained from the Codex N2 source reread.
- CONFIRMED | it drifts with corpus and era and what you count as a symbol | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf pp.413-414; https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.51-56 | [141 corpus-era] Cover and King p.413: "each source has its own characteristic entropy". P.414 discusses authorship, topic, structure and time of composition. Shannon 1951 pp.51-54 distinguishes 26- and 27-symbol alphabets and p.56 distinguishes genres. These support the corpus/genre/time/symbol caveat. Retained from the Codex N2 source reread.
- CONFIRMED | YOU WILL REJOICE TO HEAR THAT NO DISASTER HAS ACCOMPANIED THE COMMENCEMENT OF AN ENTERPRISE WHICH YOU HAVE REGARDED WITH SUCH EVIL FOREBODINGS | https://www.gutenberg.org/ebooks/84.txt.utf-8 | [142 game-passage-1] The Gutenberg Letter 1 passages match after the declared A-Z-plus-space normalization and truncation. The second passage ends at "cheeks" although the original sentence continues. The tool's word-for-word quote error concerns removed punctuation, not changed words; describe the record kind as normalized data/adaptation. Retained from the Codex N2 source reread.
- CONFIRMED | I AM ALREADY FAR NORTH OF LONDON AND AS I WALK IN THE STREETS OF PETERSBURGH I FEEL A COLD NORTHERN BREEZE PLAY UPON MY CHEEKS | https://www.gutenberg.org/ebooks/84.txt.utf-8 | [143 game-passage-2] The Gutenberg Letter 1 passages match after the declared A-Z-plus-space normalization and truncation. The second passage ends at "cheeks" although the original sentence continues. The tool's word-for-word quote error concerns removed punctuation, not changed words; describe the record kind as normalized data/adaptation. Retained from the Codex N2 source reread.
- CONFIRMED | // the ceiling: 27 symbols, no statistics | git show 4df1d97700:research/entropy-of-english/verify.mjs | [144 built-check-1] The constant is "const F0 = log2(27)"; neither letter frequencies nor corpus order enters this calculation. The transcript label accurately describes the ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | F0 = log2(27) = 4.7549 bits (Shannon rounds to 4.76) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [146 built-check-2-historical] Visually read the p.54 27-letter row: F0 4.76, F1 4.03, F2 3.32, F3 3.1, Fword 2.14. The historical comparison values agree. The record's short excerpt for entries 149/151/153 omits F3, so the complete row was checked. Retained from the Codex N2 source reread.
- CONFIRMED | The saved full-corpus ladder has 407719 symbols and F1=4.091 bits; Shannon reported 4.03. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [149 built-check-4-historical] The layer verifier reproduces the full-corpus number. Shannon 1951 p.54 supplies the historical comparator; X9 makes the offline provenance explicit. Reconciled with N2 and the directed repairs.
- CONFIRMED | F2 = H(X2&#124;X1) = 3.301 bits (Shannon 3.32) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [151 built-check-5-historical] Visually read the p.54 27-letter row: F0 4.76, F1 4.03, F2 3.32, F3 3.1, Fword 2.14. The historical comparison values agree. The record's short excerpt for entries 149/151/153 omits F3, so the complete row was checked. Retained from the Codex N2 source reread.
- CONFIRMED | The saved full-corpus F3 is 2.632 bits; Shannon reported 3.1, with corpus and sampling effects not separated here. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; public/strata/you-already-know-the-rest/index.html; research/entropy-of-english/verify-browser.mjs | [153 built-check-6-historical] The layer verifier computes H3-H2 and the paper p.54 gives 3.1. X11 removes the unsupported causal attribution. Reconciled with N2 and the directed repairs.
- CONFIRMED | // lower = Σ i(qᵢ−qᵢ₊₁)log₂i ≤ F ≤ −Σ qᵢ log₂ qᵢ = upper | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf pp.61,64 | [155 built-check-8] Read Eq.(17) on p.61: sum_i i(q_i-q_(i+1))log2(i) ≤ F_N ≤ -sum_i q_i log2(q_i), for an ideal N-gram predictor. P.64 explicitly repeats the ideal-predictor restriction for the lower bound. Retained from the Codex N2 source reread.
- CONFIRMED | "// cited destination (NOT recomputed here" ... "said plainly):" | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:464-466 | [158 built-check-13] renderCheck() explicitly labels "cited destination (NOT recomputed here". The reported 1951 and 1978 historical values are constants and match the primary sources. Retained from the Codex N2 source reread.
- CONFIRMED | Shannon 1951, 100 letters F between 0.6 and 1.3 bits redundancy ≈ 75% | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [159 built-check-14] Visually checked p.64 column 100: upper 1.3, lower .6; p.50 gives "redundancy of roughly 75%". These are Shannon's reported approximate experimental bounds, with his sampling and ideal-predictor caveats. Retained from the Codex N2 source reread.
- CONFIRMED | Cover & King 1978 ≈ 1.3 bits/symbol (gambling estimate; agrees) | https://isl.stanford.edu/~cover/papers/transIT/0413cove.pdf | [160 built-check-15] Cover and King p.413: "English text has an entropy of approximately 1.3 bits/symbol". The abstract explicitly says this agrees with Shannon's estimate. Retained from the Codex N2 source reread.
- CONFIRMED | The historical comparison is the subject's first-guess success rate. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf; git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:468-469 | [161 built-check-16] Shannon 1951 p.55 counts correct first guesses across the 129-letter passage, not success only on its first letter. Reconciled with N2 and the directed repairs.
- CONFIRMED | // 89 of 129 times (69%). Try to beat it above. | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [162 built-check-17] Shannon p.55: "total of 129 letters, 89 or 69% were guessed correctly." Independently 89/129=0.68992248. The numerical first-guess comparison agrees. Retained from the Codex N2 source reread.
- CONFIRMED | F₀ = log₂27 = 4.76 (pure noise) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [165 figure-zero-ceiling] Shannon p.54, visually read 27-letter row: "4.76", "4.03", "3.32", "3.1", "2.14". Independently log2(27)=4.754887502163469; 4.76 is the historical approximate ceiling. Retained from the Codex N2 source reread.
- CONFIRMED | 0.6–1.3 @100 | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [166 figure-100-band] Visually checked p.64 column 100: upper 1.3, lower .6; p.50 gives "redundancy of roughly 75%". These are Shannon's reported approximate experimental bounds, with his sampling and ideal-predictor caveats. Retained from the Codex N2 source reread.
- CONFIRMED | (Shannon's subject: 69%) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [167 game-first-hit-readout] Shannon p.55: "total of 129 letters, 89 or 69% were guessed correctly." Independently 89/129=0.68992248. The numerical first-guess comparison agrees. Retained from the Codex N2 source reread.
- CONFIRMED | (Shannon's subject: 69%) | https://www.princeton.edu/~wbialek/rome/refs/shannon_51.pdf | [168 game-first-hit-historical] Shannon p.55: "total of 129 letters, 89 or 69% were guessed correctly." Independently 89/129=0.68992248. The numerical first-guess comparison agrees. Retained from the Codex N2 source reread.
- CONFIRMED | The handler runs on each guess; the estimate uses only completed letters. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:279-305 | [169 game-repeated-updates] guess calls updateGame on a miss, but only a hit appends to ranks. The ABC probe reproduces the distinction; X16. Reconciled with N2 and the directed repairs.
- CONFIRMED | Auto-finished by the order-1 model predictor | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:329-348 | [170 game-autofinish-model] The auto-finish handler calls "idealRankFor(pos)"; that function ranks letters by "BIG[prev<0?26:prev]", so it is the empirical order-1 model. Probe auto-finish completes 142 symbols and reports 20% first hits. Retained from the Codex N2 source reread.
- CONFIRMED | the rank-frequency marginal a monkey reproduces for free | git show 4df1d97700:src/content/strata/the-law-even-monkeys-obey.md | [173 monkey-marginal] The linked layer explicitly separates word-count marginals from conditional structure and describes a monkey model with independent letters plus space. Its claim is the rank-frequency power-law mechanism, not exact equality to every real-book frequency. Retained from the Codex N2 source reread.
- CONFIRMED | a length provably not any ratio. | git show 4df1d97700:src/content/strata/incommensurable.md | [174 irrational-length] The linked sonnet and proof concern sqrt(2): "no fraction, however large its terms, has a square equal to 2." Checked the parity proof: a^2=2b^2 forces both a and b even, contradicting lowest terms. "Audible" refers to the sonnet form; the connection is not a claimed audio widget or denial of exact radical notation. Retained from the Codex N2 source reread.
- CONFIRMED | (added) Spacebar guesses a space. | git show 4df1d97700:public/strata/you-already-know-the-rest/index.html:317-319 | [175 A1-spacebar] The keydown handler calls guess(" ") when e.key is a space. The VM dispatch against a space target produces pos=1 and ranks=[1]. Source-record literal search: this sentence absent; positive control "every keystroke updates" present. Retained from the Codex N2 source reread.

## Findings

- [fixed] X1 (MINOR): Publication is 2026-06-18; the page history includes later edits. Set JSON-LD dateModified to the repair date, 2026-10-14. MINOR metadata repair; no public dated line.
- [fixed] X2 (MINOR): Shannon 1951 p.55 reports 89 correct first guesses out of 129, or 68.992248%; 40 failed. Changed near-certainty wording to often predictable. No public dated line.
- [fixed] X3 (WRONG): The page function gives H(.7,.3)=0.8812908992306927 and H(.8,.1,.1)=0.9219280948873623. The first-hit rate and entropy both increase. Replaced the implication with dependence on the full guess-rank distribution.
- [fixed] X4 (WRONG): Shannon p.55 has the 129-letter sample and 89/129 result. Its text matches Pickup on Noon Street, section FOUR; the publisher lists it in Chandler short fiction. Corrected the genre; retained the figures and source words.
- [fixed] X5 (WRONG): Shannon 1951 p.64 restricts the lower-bound proof to the ideal predictor. Cover and King p.414 distinguish finite-context conditional entropy from entropy rate. A single completed letter gives 0 to 0 in the game. Replaced guarantees with illustrative sample estimates and placed the restriction beside the output. Repair review: the legacy OG template also retained a proof sentence and a player marker fixed at 0.95 bits without player data. Removed that marker and its legend entry; attributed the one-bit estimate to Shannon and described the game as illustrative in research/entropy-of-english/og.html.
- [fixed] X6 (MINOR): drawFig reads DATA.ideal; these saved values come from the offline full-corpus verifier. Only ladderLive(DATA.sliceText) is computed from the embedded slice. Corrected caption and related descriptions; no public dated line.
- [fixed] X7 (WRONG): Shannon 1948 section 7, p.15 discusses restoration after deletion as a lower-bound argument. drawRedund retains spaces and positions and scores no reconstruction. Relabeled this as qualitative; the 75% figure is a cited estimate. Shares the correction with X17.
- [fixed] X8 (WRONG): rText concatenates DATA.game[1].text and DATA.sliceText.slice(0,220). The committed Gutenberg corpus places the game phrase in the second paragraph of Letter 1, and the slice starts with title and contents. Accurately labeled the existing selection without changing the source words.
- [fixed] X9 (MINOR): renderCheck formats DATA.ladderFull, DATA.ideal and cited historical numbers. Replacing the slice with 160000 As changes its computed ladder to zero but leaves the saved panel unchanged. Distinguished live calculations, saved offline figures and citations; no public dated line.
- [fixed] X10 (MINOR): The panel formats saved offline figures; it does not derive the full-corpus results in the browser. Changed its heading and generated labels. No public dated line.
- [fixed] X11 (MINOR): verify.mjs computes F3 as H3-H2, with two preceding letters of context. Neither this computation nor Shannon pp.51-52 separates sampling effects from the change of corpus. Qualified the explanation in HTML, Markdown and verifier wording; no public dated line.
- [fixed] X12 (WRONG): shannonBounds sorts rank frequencies descending. With q=(1/3,2/3), sorted lower=2/3; applying the lower expression to the original rank order gives 4/3. Disclosed sorting and removed raw-bound wording, retaining the ideal-predictor restriction.
- [fixed] X13 (WRONG): The scan of Shannon 1951 p.50 places the quoted phrase below the Introduction heading, after the abstract. Corrected only its section attribution in the page and verifier; quotation words unchanged.
- [fixed] X14 (MINOR): The live ladder reads DATA.sliceText; the ideal bounds are DATA.ideal saved by the offline verifier. Corrected the footer and equivalent Markdown descriptions. No public dated line.
- [fixed] X15 (WRONG): The page loads /fonts/fonts.css and /media/home-banner.js. worker.js pulseSnippet posts the page path to /api/pulse; the response pipeline also injects a Cloudflare beacon when configured. Both telemetry scripts were present in fetched live HTML. Scoped the local statement to game guesses and disclosed page assets and telemetry, without asserting storage of guesses.
- [fixed] X16 (WRONG): The page records a rank only on a correct guess. For synthetic passage ABC and guesses A,X,B,C, the endpoints start at 0, stay there after X, rise to 1 after B, then fall. Replaced every downward-trajectory promise and every-keystroke estimate wording, including metadata and Markdown.
- [fixed] X17 (WRONG): Same quantitative error as X7. The widget tests no exact reconstruction; readable masked text is not a measurement of 75% redundancy. Corrected Markdown and related measurement language; shares the X7 correction.
- [fixed] X18 (WRONG): This layer uses Frankenstein. The other page DATA.corpora keys are moby-dick, pride-and-prejudice and shakespeare. Corrected this layer's frontmatter relation and its copied nearby-layer note; no other layer edited.
- [fixed] X19 (WRONG): Shannon 1951 pp.56-57 state that ordinary column N means N-1 preceding letters, but explicitly describe the final experiment as 100 known letters. Corrected the page and OG chart source to 0..14,100. The browser test checks the contexts and cyan start coordinates. Existing PNG and WebP cards were not rendered or re-encoded, as instructed.

## Checks

- `node research/entropy-of-english/verify.mjs`: exit 0, 11/11 checks pass; data.json remains byte-identical.
- `node research/entropy-of-english/verify-browser.mjs`: exit 0, all 17 browser checks pass, including the completed-letter, sorting, context-coordinate and saved-result probes.
- `node scripts/check-claims-pass.mjs`: exit 1 only for five other layer records: any-way-you-fall, island-biogeography, let-them-eat-cake, the-bread-with-no-name, and the-drop-that-shatters-itself. This layer passes the exported checkRecord function with no errors. The full before/after report is research/assay-line/cited-1014b/you-already-know-the-rest/fix-report.md.

Repair review by the Codex repair worker, 2026-10-14:

- Re-ran `node research/entropy-of-english/verify.mjs`: exit 0, 11/11 checks pass. Re-ran `node research/entropy-of-english/verify-browser.mjs`: exit 0, 17/17 checks pass.
- Re-ran `node scripts/check-claims-pass.mjs`: exit 1 only for eight other layer records: any-way-you-fall, helium-voice, island-biogeography, khinchin, let-them-eat-cake, the-bread-with-no-name, the-drop-that-shatters-itself, and the-edge-of-the-bow. Ignored as directed.
- `node scripts/check-em-dashes.mjs --base 9306f663bb5bb7f447014082bd7f342ea22e982d`: exit 1 only for four added prose lines in dunning-kruger and the-only-other-pair. The explicit local merge base prevents the script's default git fetch. Ignored the other-layer failures as directed. The repair report is research/assay-line/cited-1014b/you-already-know-the-rest/repair-report.md.