--- slug: vowel-space-formants passed: 2026-10-18 by: the assay line (research/assay-line/claude-1018b/), Claude fixer (claude-opus-5-5) directed by claude-assaying-boyle-1014; source record by a Claude agent (claude-sonnet-5-5), checked by a Claude agent (claude-opus-5-5) checked: 112 confirmed: 87 wrong: 10 # 5 WRONG + 5 MINOR unverifiable: 4 observed: 10 stale: 1 --- This was a directed repair pass over the 11 triage items in research/assay-line/claude-1018b/vowel-space-formants/triage.json, read at 87ef051577. The checker (N2) re-read the layer against the N1 source record at 4a6041fc27, reading Peterson and Barney (1952) in the open-access pure.mpg.de scan, and raised five WRONG lines, five MINOR and one STALE; the claim lines below are its record, with the fixer's outcome appended to each triage line. The fixer re-read the evidence for every WRONG: a rerun of formants.js reproduceTable2 over the verifier's own rows (F1 worst 14 Hz, F2 worst 23 Hz, 11 and 17 of 30 cells past the table's 5 Hz rounding), the synthesiser's impulse train and the specimens' 48 kHz rate (3000 Hz is a harmonic of both 120 and 250 Hz), the CRAN archive of phonTools (first release 2012-07-24), the sibling layer's 1400 to 1700 timeline, and the paper's text layer (no recording year or room given). All five were fixed with dated Corrected lines, the one STALE item has an Updated line, one MINOR item was fixed because its sentence was already being rewritten, and four MINOR items are listed. The verifier's string match on the rewritten Table II sentence was updated in both copies (research/ and public/checks/). After the fixes verify-vowel-space-formants.mjs reports 299/301, the same two shared-chrome failures it reported before any edit. ## Claims - CONFIRMED | Peterson and Barney measured from 76 speakers at Bell Labs | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.175 title block "Bell Telephone Laboratories, Inc., Murray Hill, New Jersey"; abstract "illustrated from results of tests involving some 76 speakers and 70 listeners" - CONFIRMED | Peterson and Barney recorded 76 people at Bell Telephone Laboratories saying ten vowels in an hVd frame, twice each | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.175 "ten monosyllabic words each beginning with [h] and ending with [d] and differing only in the vowel ... each speaker was asked to pronounce two different lists"; p.176-177 "A total of 76 speakers ... each recorded two lists of 10 words" - CONFIRMED | and measured F1, F2 and F3 off spectrograms by hand | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.177 "The sections ... have been measured with calibrated Plexiglass templates to provide data about the fundamental and formant frequencies and relative formant amplitudes of each of the 1520 recorded sounds" (sections made with the sound spectrograph; this supports "by hand" better than the record's p.180 excerpt) - CONFIRMED | quote "A total of 76 speakers, including 33 men, 28 women and 15 children, each recorded two lists of 10 words, making a total of 1520 recorded words." | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | verbatim, pp.176-177 - CONFIRMED | The sentence runs across the page break | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | PDF p.2 (printed 176) ends "each recorded two lists of 10 words,"; PDF p.3 (177) begins "making a total of 1520 recorded words." - CONFIRMED | quote "the difference between successive utterances of the same sound by the same individual is much less significant statistically than the difference between utterances of the same sound by different individuals" (p.182) | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | verbatim, p.182 left column - CONFIRMED | "The next two sentences of the same paragraph on p.182 read: An analysis of variance of the data in Fig. 7 shows ... 0.1 percent probability." | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | verbatim; the page image shows no paragraph break, so "the same paragraph" is right (the record's "where" calls it "the paragraph after", which is wrong) - CONFIRMED | So the between-speaker effect is quantified, as a variance ratio past a one-in-a-thousand probability | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.182 "the probability of having a variance ratio as high as that found when comparing speakers would be less than one in a thousand" - MINOR | "What is not in the paper is how often the position alone recovers the vowel, because in 1952 nobody could run seventy-six cross-validation folds." | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | first half confirmed (whole paper read: listener Table I, Figs. 8-9 loops "about 90 percent of the values", no classification-by-position accuracy); the "because ... nobody could" clause is an unsourced claim about 1952 capability: a nearest-centroid leave-one-out over 1520 rows is desk-calculator arithmetic, and the paper thanks a statistical staff (Shewhart, Packer). Fixer (K1): declined, tier B listed: Decided from the checker's note: the first half (the paper gives no classification-by-position rate) holds; the causal clause "because in 1952 nobody could run seventy-six cross-validation folds" is an unsourced claim about 1952 capability, and the paper thanks W. A. Shewhart and M. C. Packer for statistical help (re-read in the pure.mpg.de scan's text layer), so "nobody did" is what the source supports. - CONFIRMED | The paper notes that most of its women and children grew up in the Middle Atlantic area while the men were a broader sample. | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.177 "Most of the women and children grew up in the Middle Atlantic speech area. The male speakers represented a much broader regional sampling" - CONFIRMED | compares every cell against Table II on page 183 of the paper | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.183 "TABLE II. Averages of fundamental and formant frequencies and formant amplitudes of vowels by 76 speakers." - CONFIRMED | The published values are rounded to 10 cycles per second | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | inference, not a statement: all 90 F1/F2/F3 cells are multiples of 10 while the F0 cells are not (e.g. children 272 269 ...); I re-read every formant cell against PB52_TABLE2 in formants.js: all 90 match - CONFIRMED | quote "produces extensive overlap in the [ʊ] region in a graph involving only the first two formants" | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | verbatim, p.183 right column, read off the rendered image - CONFIRMED | quote "may be isolated from the other vowels readily by means of the third formant" | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | verbatim, p.183 right column - CONFIRMED | They said it about their Fig. 9, which they had already filtered down to men and children and to words every listener heard alike | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.183 "measurements for the women speakers have been omitted"; caption "vowels spoken by men and children, which were classified unanimously by all listeners"; the [ɝ] paragraph follows that Fig. 9 discussion - CONFIRMED | their Fig. 8 ... keeps the vowel and discusses its overlaps | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.182 "Considerable overlapping of areas is indicated, particularly between [ɝ] and [ɛ], [ɝ] and [ʊ] ... it may be easily distinguished from all the others if the third formant frequency is used" (the record's excerpt has "all the other sounds") - CONFIRMED | Peterson and Barney wrote, on p.183 of the same paper, that the r-coloured vowel ... | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.183 (running head and image); source reads "simplified by the omission of [ɝ]", not the record's "omission of the [ɝ] sounds" - OBSERVED | The listener-agreement figure often quoted for this paper is not here. The accessible scans of the relevant table did not read cleanly enough to trust | research/vowel-space-formants/README.md | README records the omission ("The relevant table did not read cleanly enough from the scan"); I confirm the Table I text layer is scrambled; Table I is on p.182, not p.179 as the record says - CONFIRMED | whose own manual records that they were downloaded from a University of Pennsylvania FTP site and were reportedly based on a printed version supplied by Ignatius Mattingly | https://www.fon.hum.uva.nl/praat/manual/Create_formant_table__Peterson___Barney_1952_.html | "We originally downloaded the data from the University of Pennsylvania FTP site, where they were reportedly based on a printed version supplied by Ignatius Mattingly." - CONFIRMED | The rows come from the phonTools package on CRAN, built in turn from tables bundled with Praat | https://cran.r-project.org/web/packages/phonTools/phonTools.pdf | pb52 entry, Source: "The data was created from tables provided within Praat" - CONFIRMED | (CRAN, BSD 2 clause, copyright 2015 Santiago Barreda) | https://cran.r-project.org/web/packages/phonTools/LICENSE | CRAN "License: BSD_2_clause + file LICENSE"; LICENSE "YEAR: 2015 COPYRIGHT HOLDER: Santiago Barreda" - CONFIRMED | Those 1520 measurements are the whole reference frame of this page | https://www.fon.hum.uva.nl/praat/manual/Create_formant_table__Peterson___Barney_1952_.html | "76 speakers (33 men, 28 women and 15 children). Every vowel was pronounced twice, so that there are 1520 recorded vowels in total."; verifier: "1520 measured tokens read from the .rda" - CONFIRMED | they ship with it as data/pb52.json | public/strata/vowel-space-formants/data/pb52.json | file present at the pin and live (HTTP 200, 38233 bytes); verifier "data/pb52.json regenerates byte for byte" - CONFIRMED | What ships here is not the authors' laboratory notebook; it is a transmission of it. | public/strata/vowel-space-formants/data/LICENCE | "phonTools states that its pb52 data is built from tables bundled with Praat"; Praat manual names the UPenn FTP and Mattingly steps - WRONG | "the F1 and F2 columns reproduce the authors' own printed table to within its rounding" (index.html:372), with the page's own next sentence "rounded to 10 cycles per second, so they carry about ±5 Hz" | public/strata/vowel-space-formants/formants.js reproduceTable2 over data/pb52.json, recreated: node code/t2.mjs | 11 of 30 F1 cells and 17 of 30 F2 cells miss the printed value by more than 5 Hz, worst 14 Hz (F1) and 23 Hz (F2); the verifier and the record's own excerpt say "within 25 Hz", not within the rounding. Fixer (K2): fixed 2026-10-18 - CONFIRMED | F1 and F2 ... agree everywhere within 25 Hz. (built at load) | public/strata/vowel-space-formants/instrument.js | verifier "every F1 cell ... within 25 Hz (worst 14)", "every F2 cell ... within 25 Hz (worst 23)" - CONFIRMED | Raw F1 and F2 in hertz identify the vowel a speaker intended only 57.4 per cent of the time, leave one speaker out, nearest centroid | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun at the pin: "leave-one-speaker-out accuracy, hz = 57.434 %" - CONFIRMED | the distances between your own peaks get you to 81.2 on a single vowel | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "bd3 = 81.184 %" - CONFIRMED | z-scoring against your own whole vowel set reaches 89.7 | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "zscore = 89.671 %" - CONFIRMED | The last 8.5 points cost you your entire vowel space | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "the part that costs the whole vowel space = 8.487 points" - CONFIRMED | with an interval it derived by measuring its own misses on 180 signals whose answers it already knew | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "180 grid points, none refused" - CONFIRMED | resynthesises each of them at six pitches from 100 to 300 Hz | research/vowel-space-formants/make-specimens.mjs | SWEEP_F0 = [100, 140, 180, 220, 260, 300] - CONFIRMED | The allowance below is the ninetieth percentile of the absolute error over the thirty shapes at each pitch | public/strata/vowel-space-formants/formants.js | biasCurve uses percentile(..., 0.9); verifier "27 of the 30 shapes sit inside the allowance" - STALE | "301/301 checks passed" (index.html:656) | git show a5bf6c91b4a (verifier run in scratch: "301/301 checks passed") against 4a6041fc27 (run: "FAILED: 299/301 checks passed") | true on 2026-08-17; two chrome checks fail since: a bsky.app URL in the JSON-LD (from f43d1e88d8e, 2026-08-21) and the og-title count dropping from 3 to 2 (2e60bcb5dd1, 2026-10-01). Fixer (K3): dated 2026-10-18 (one Updated line) - CONFIRMED | recomputes all of them from the sources and the equations, regenerates the four recordings and compares their sha256, re-derives the slate's tolerances from the calibration rule, asserts the refusal paths fire, and string-matches the figures against the bytes of this page | research/vowel-space-formants/verify-vowel-space-formants.mjs | each named step present in the run output; a dek mutation ("reaches 89.7" to "88.7") is caught - CONFIRMED | Four recordings ship with this page. | research/vowel-space-formants/specimen-manifest.json | four .wav files under specimens/ at the pin, each "regenerates byte for byte" - CONFIRMED | an impulse train at a fixed pitch, through a glottal tilt, through a cascade of two-pole resonators, through a lip radiation difference | public/strata/vowel-space-formants/formants.js | synthVowel: impulse train, two one-pole low-pass passes, two-pole resonators per formant, first difference - CONFIRMED | An /ɑ/ at 730 and 1090 Hz, and an /i/ at 270 and 2290 Hz | research/vowel-space-formants/README.md | pole positions typed into the synthesiser - CONFIRMED | Those are 460 and 1200 Hz apart | research/vowel-space-formants/README.md | 730 - 270 = 460; 2290 - 1090 = 1200 - WRONG | "The same poles said at 120 Hz and at 250 Hz share no harmonic and look nothing alike" (index.html:176) | recreated: node code/harm.mjs on public/strata/vowel-space-formants/specimens/*.wav | 3000 Hz is the 25th harmonic of 120 and the 12th of 250; both files have a harmonic peak there (/ɑ/ -50 dB, /i/ -21 dB re the strongest low harmonic, 30 to 60 dB above the floor between harmonics), well inside the 5 kHz band the estimator reads. Fixer (K4): fixed 2026-10-18 - CONFIRMED | a two-entry lookup keyed on nothing but the zero-crossing rate passes all eight rows above, and this page's verifier builds exactly that stub and shows it passing | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "a two-entry lookup on gross spectral shape DOES pass the four F1 rows" and "the four F2 rows"; the stub is keyed on zero crossings per second - OBSERVED | Until 2026-08-17 the second half of that sentence was false here | research/vowel-space-formants/README.md | "Both gate/inject-a.wav and gate/inject-b.wav were built from SHAPES.a and SHAPES.i at 120 Hz" - CONFIRMED | The injections are now two vowels the slate does not contain, and the verifier runs that same lookup against them and requires it to miss. | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "neither gate injection uses a slate shape"; "the memorising lookup answers 730 Hz on inject-a and 270 Hz on inject-b, and both miss" - CONFIRMED | This page uses twelve, which is twice the five formants expected below 5 kHz plus two | public/strata/vowel-space-formants/formants.js | "order: 12, // LPC order = 2 x 5 formants + 2." - CONFIRMED | so five formants are expected below the 5 kHz Nyquist | https://www.fon.hum.uva.nl/praat/manual/Sound__To_Formant__burg____.html | "for most analyses of human speech, you will want to extract 5 formants per frame"; male ceiling 5000 Hz - CONFIRMED | That is Praat's setting for an average adult male voice; Praat's own standard ceiling is 5500 Hz, which would resample to 11 kHz. | https://www.fon.hum.uva.nl/praat/manual/Sound__To_Formant__burg____.html | "5500 Hz (which is Praat's standard value), an average adult male speaker ... 5000 Hz"; "resampled to a sampling frequency of twice the value of Formant ceiling" - CONFIRMED | at order 10 the shipped /i/ token's F2 comes back closer | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "order 10 recovers the /i/ token's F2 closer than order 12 is true: 1 Hz against 13 Hz" - CONFIRMED | 25 ms of signal per estimate. | public/strata/vowel-space-formants/formants.js | "frameMs: 25," - CONFIRMED | in the lag range for 80 to 400 Hz | public/strata/vowel-space-formants/formants.js | "f0Lo: 80, f0Hi: 400," - CONFIRMED | F1 below 90 Hz would be thrown away | public/strata/vowel-space-formants/formants.js | "fLo: 90, fHi: 4800," - UNVERIFIABLE | which no human vowel has | tried: Wikipedia Formant (no lower bound stated), the shipped rows | no source states a lower bound on human F1; positive control: the 1520 rows load and their minimum F1 is 190 Hz, twice the floor; consistent, not proven - OBSERVED | A different coefficient moves F1 by a few hertz. | recreated: node code/recreate.mjs (analyse on the four shipped recordings at pre-emphasis 0.90 to 1.0) | 0.95 vs 0.97 moves F1 by at most 2 Hz; 0.90 vs 0.97 by up to 8.8 Hz (i-f0-120); /ɑ/ moves under 1 Hz - CONFIRMED | the frame-to-frame spread of this estimator is a fraction of a hertz | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "the widest bootstrap spread on any shipped recording is 0.26 Hz" - OBSERVED | Fed a steady 1 kHz tone this page calls almost every frame voiced | recreated: node research/vowel-space-formants/verify-vowel-space-formants.mjs | "a steady 1 kHz tone is NOT refused: it is measured as a vowel (988 / 1012 Hz, 239 of 239 frames voiced)" - CONFIRMED | Linear prediction ... is pulled toward whichever harmonic sits nearest a resonance, so at a high pitch the estimate can be tens of hertz wrong | https://api.crossref.org/works/10.1121/1.428820 | Vallabha and Tuller (2000) abstract: "F0 quantization: In addition to the expected result that error increases with F0"; Alku et al. 2013 abstract (doi 10.1121/1.4812756): conventional LP F1 relative error for [a] of 11% at high pitch - CONFIRMED | All 1520. No token is dropped, no speaker is dropped | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: each space "classified all 1520 tokens in 76 folds" - CONFIRMED | By the fixed vowel order i ɪ ɛ æ ɑ ɔ ʊ u ʌ ɝ | public/strata/vowel-space-formants/formants.js | VOWELS = i I E { A O U u V 3' (SAMPA for the same ten); verifier "an exact tie breaks to the earlier vowel" - CONFIRMED | with ten vowels the only honest answer is then 10.0% | research/vowel-space-formants/README.md | 152 tokens per vowel; verifier shuffle "worst deviation 2.0 points from 10.0%" - CONFIRMED | the same folds under a linear discriminant are computed and printed under the ladder above | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: 68.947 / 84.539 / 91.447 % - CONFIRMED | 4 vowels for log-mean, 5 for z-scores | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "z-score row first beats doing nothing at 5 vowels"; "log-mean centring already pays at 4" - CONFIRMED | raw hertz rises from 57.4 ... (r-coloured vowel dropped) | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "raw hertz without the r-coloured vowel = 59.576 %" - CONFIRMED | it would fire on ... of the 1520 tokens | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "fires on 812 F0 values, 2 F1 values and 0 F2 values" - CONFIRMED | rows sit exactly at the floor | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "11 of the 1520 rows already sit at or below 0 bark on B1-B0" - OBSERVED | We could not find that first number published, as of 2026-08-17. | research/vowel-space-formants/README.md | "We searched and did not find it"; worded as a statement about the search - UNVERIFIABLE | Published vowel-classification rates on this same 1952 data range from roughly the fifties to the high eighties | tried: Hillenbrand et al. 1995 full text (web.archive.org copy of homepages.wmich.edu/~hillenbr/Papers/HillenbrandGettyClarkWheeler.pdf), PubMed (Hillenbrand and Gayvert 1993), 2 web searches | upper end supported: Table VIII gives PB quadratic-discriminant rates 74.9 / 83.6 / 85.9 / 86.6; no published rate in the fifties found; a search summary credited Miller (1989) with 93.0% on some scheme but the sentence could not be found at any page fetched - CONFIRMED | 26.81 f / (1960 + f) − 0.53 | https://api.crossref.org/works/10.1121/1.399849 | Traunmüller 1990 abstract: "the inversible formula z=[26.81/(1+1960/f )]−0.53 is proposed" (algebraically the page's form) - CONFIRMED | the low branch fires below 204 Hz | research/vowel-space-formants/verify-vowel-space-formants.mjs | z = 2 bark at f = 2.53 x 1960 / 24.28 = 204.2 Hz - CONFIRMED | Traunmüller, H. (1990). Analytical expressions for the tonotopic sensory scale. | https://api.crossref.org/works/10.1121/1.399849 | author Hartmut Traunmüller, 1990, title matches - CONFIRMED | The idea of normalising by bark differences is theirs | https://api.crossref.org/works/10.1121/1.393381 | abstract: "the distance in critical bands (barks) between neighboring formants and between the fundamental frequency (F0) and first formant (F1)" - CONFIRMED | This page asks for all three off and then reads back what the track actually granted | public/_kit/sensors.js | line 75 "echoCancellation: false, noiseSuppression: false, autoGainControl: false"; line 80 track.getSettings(); printed by instrument.js:1025 - CONFIRMED | Echo cancellation, noise suppression and automatic gain are on by default in most browsers | https://lists.w3.org/Archives/Public/public-webrtc-logs/2021Mar/0012.html | "Chrome, Firefox and Safari agree on the following defaults: ... audio: echoCancellation=true, noiseSuppression=true, autoGainControl=true"; Firefox test_defaultAudioConstraints.html asserts each "should be ON by default" - UNVERIFIABLE | noise suppression is spectral: it moves formants | tried: MDN noiseSuppression, one web search on noise suppression and formant measurement | nothing found that measures browser noise suppression moving formant frequencies; the page measures nothing of it either - CONFIRMED | Nothing is uploaded, | public/strata/vowel-space-formants/instrument.js | the only fetch calls are GETs of ./data/pb52.json and the specimen .wav files; no XHR, beacon or storage in instrument.js, formants.js or the kits (the live page adds a path-only /api/pulse beacon and a Cloudflare analytics script, neither carrying audio) - CONFIRMED | No third-party audio is shipped on this page at all. | research/vowel-space-formants/verify-vowel-space-formants.mjs | rerun: "no audio ships that this repository did not synthesise" - CONFIRMED | dedicated to the public domain under CC0 1.0 | public/strata/vowel-space-formants/index.html | the page's own dedication of its own files - OBSERVED | The Hillenbrand et al. (1995) recordings were considered and rejected | public/strata/vowel-space-formants/data/LICENCE | "The Hillenbrand et al. (1995) recordings were considered and rejected" - CONFIRMED | the only live mirror carries a permissive badge applied by the mirror's owner over data described there as merely hosted with permission | https://github.com/santiagobarreda/hillenbrand_et_al_1995 | LICENSE "MIT License / Copyright (c) 2024 Santiago Barreda"; README "Hosted with permission from Jim Hillenbrand."; the original homepages.wmich.edu host does not resolve (control: cran.r-project.org answered in the same session) - CONFIRMED | Peterson, G. E. & Barney, H. L. (1952). Control Methods Used in a Study of the Vowels. | https://api.crossref.org/works/10.1121/1.1906875 | authors, title, 1952 match - CONFIRMED | 24(2):175–184. | https://api.crossref.org/works/10.1121/1.1906875 | volume 24, issue 2, page 175-184 - CONFIRMED | Lobanov, B. M. (1971). Classification of Russian Vowels Spoken by Different Speakers. | https://api.crossref.org/works/10.1121/1.1912396 | match - CONFIRMED | 49(2B):606–608. | https://api.crossref.org/works/10.1121/1.1912396 | volume 49, issue 2B, page 606-608 - CONFIRMED | Syrdal, A. K. & Gopal, H. S. (1986). A perceptual model of vowel recognition ... | https://api.crossref.org/works/10.1121/1.393381 | Ann K. Syrdal, H. S. Gopal, 1986, title match - CONFIRMED | 79(4):1086–1100. | https://api.crossref.org/works/10.1121/1.393381 | match - CONFIRMED | 88(1):97–100. | https://api.crossref.org/works/10.1121/1.399849 | match - CONFIRMED | Hillenbrand, J., Getty, L. A., Clark, M. J. & Wheeler, K. (1995). Acoustic characteristics of American English vowels. | https://api.crossref.org/works/10.1121/1.411872 | four authors, 1995, title match - CONFIRMED | 97(5):3099–3111. | https://api.crossref.org/works/10.1121/1.411872 | match - CONFIRMED | Alku, P., Pohjalainen, J., Vainio, M., Laukkanen, A.-M. & Story, B. H. (2013). Formant frequency estimation of high-pitched vowels using weighted linear prediction. | https://api.crossref.org/works/10.1121/1.4812756 | match - CONFIRMED | 134(2):1295–1313. | https://api.crossref.org/works/10.1121/1.4812756 | match - OBSERVED | All bibliographic fields above ... were checked field by field against Crossref on 2026-08-17 | research/vowel-space-formants/README.md | README records the check; recreated today against api.crossref.org for all six DOIs: every author, title, volume, issue, page and year matches - OBSERVED | Of those papers only Peterson and Barney was readable at source | research/vowel-space-formants/README.md | "Five of the six it does cite are paywalled and were not read at source" - OBSERVED | this page had it wrong until 2026-08-17 | git show 9802274c22f:public/strata/vowel-space-formants/index.html | line 373 "They never put a number on it, because in 1952 nobody could run seventy-six cross-validation" - CONFIRMED | one file, formants.js, loaded unchanged by the page and by the verifier | research/vowel-space-formants/load-formants.mjs | evaluates the shipped public/.../formants.js bytes; the page loads the same file - CONFIRMED | the live microphone path is driven end to end by node scripts/check-live-sensor.mjs --only=vowel-space-formants | research/anchor-wave/cases/vowel-space-formants.json | script and case file exist at the pin (not run: needs Chromium) - CONFIRMED | node research/vowel-space-formants/make-specimens.mjs | research/vowel-space-formants/make-specimens.mjs | exists; the verifier regenerates every shipped file from it byte for byte - CONFIRMED | From the same 1952 rows it recovers vocal-tract length | src/content/strata/the-vowel-in-the-tube.md | dek "from the raw 1952 Peterson & Barney data ... hand back a woman's tract at ~15 cm and a child's at ~13 cm" - CONFIRMED | That layer is the time-frequency uncertainty product made operable | src/content/strata/the-window-you-cannot-win.md | dek "The floor under the trade is exact ... 0.079577, which is 1/4pi" - CONFIRMED | That page re-measures a released corpus of 270 real rooms and refuses per band wherever the recording cannot support an answer | src/content/strata/when-the-room-becomes-the-sound.md | dek "270 impulse responses ... measures all of them again from the released audio ... a refusal wherever the recording cannot support an answer (728 of 2,160 octave bands" - WRONG | "so you can hear the Great Vowel Shift move a vowel across the space over six centuries" (relates note, src/content/strata/vowel-space-formants.md:18; shown on the home page's connections list) | src/content/strata/the-sound-the-spelling-forgot.md | that layer: "between 1400 and 1700", "Drag a three-century timeline", "a timeline you drag from 1400 to 1700". Fixer (K5): fixed 2026-10-18 - MINOR | "There, published formant values for Middle English drive a synthesiser" (same note) | src/content/strata/the-sound-the-spelling-forgot.md | the values are "published mean formant frequencies (F1/F2 from Catford 2001 ... for the seven continental qualities; Peterson & Barney 1952 for the diphthong endpoints)", modern reference values mapped onto Middle English vowels; no formant values for Middle English exist ("none of any human voice predates 1877"). Fixer (K6): fixed 2026-10-18 - CONFIRMED | the high-pitch bias this page measures for itself is a known problem, and this is where the literature on it starts for anyone who wants to read further | https://api.crossref.org/works/10.1121/1.4812756 | "its performance is known to deteriorate for high-pitched voices"; read as a starting point for the reader (the abstract cites five earlier methods, so not the literature's origin) - CONFIRMED | end corrections of z + 0.15(2 − z) below 2 bark and z + 0.22(z − 20.1) above 20.1 bark are commonly used alongside it | https://pydrobert-speech.readthedocs.io/en/v0.3.0/pydrobert.speech.scales.html | implements exactly those branches with z = 26.81f/(1960 + f) - 0.53 and cites Traunmüller 1990 - CONFIRMED | the idea is Lobanov's (1971) | https://jofrhwld.github.io/tidynorm/reference/norm_lobanov.html | secondary: Lobanov normalisation is the per-speaker z-score; Lobanov's own abstract (Crossref) "A new statistical method of normalization is then suggested" - CONFIRMED | the idea is Nearey's uniform scaling (1978 dissertation) | https://jofrhwld.github.io/tidynorm/reference/norm_nearey.html | "Nearey, T. M. (1978). Phonetic Feature Systems for Vowels [Ph.D.]. University of Alberta."; note: depts.washington.edu/sociolab/VOIS3D/references gives "(1977) ... Reprinted 1978 by the Indiana University Linguistics Club. Dissertation.", so the year may be the reprint's - OBSERVED | Read at source from the open-access copy at pure.mpg.de | research/vowel-space-formants/README.md | README records it; the URL serves the 10-page scan today (HTTP 200, application/pdf) - MINOR | "Running the bootstrap on it would add several seconds to page load for a number nothing uses." (index.html:570) | public/strata/vowel-space-formants/formants.js sweepPoint over the 180-point grid, bootstrap 0 vs 400, recreated: node code/recreate.mjs | on this shared 4-core box under load average 8.6: +175 to +610 ms, not several seconds; a slow phone might take longer, but nothing records "several seconds". Fixer (K7): declined, tier B listed: Checker measured +175 to +610 ms on this box for the 180-point grid with 400 resamples, not several seconds; a slow phone might be slower, but nothing on record measures several seconds. - WRONG | (added) "Reading a fifty-year-old R data file with a parser written from scratch" (index.html:376) | https://cran.r-project.org/src/contrib/Archive/phonTools/ | phonTools (source of pb52.rda) first released 2012-07-24 (phonTools_0.1-1.tar.gz); the data are from 1952; neither is fifty years old, and R itself is younger than fifty. Fixer (K8): fixed 2026-10-18 - WRONG | (added) "the finding is that a position in it barely identifies a vowel even within one year and one building" (relates note, src/content/strata/vowel-space-formants.md:18; rendered in the home page's connections list) | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | the paper fixes neither a single year nor a single building for the recordings (only the listening tests are placed, in Arnold Auditorium); the layer's own verifier comment says "nothing in it fixes the recordings to one building" and removed the phrase from the plain card and body. Fixer (K9): fixed 2026-10-18 - UNVERIFIABLE | (added) "The reference frame is 76 people recorded at Bell Labs around 1950." (index.html:623; also instrument.js:977) | tried: the whole paper | no recording date is given; received 3 December 1951 and its references run to 1950, so "around 1950" is consistent but not stated - CONFIRMED | (added) Both axes run high to low, which is the phonetician's convention | https://pure.mpg.de/rest/items/item_2375480/component/file_2375479/content | p.177, Fig. 3 "in this plot the origin is at the upper right"; p.178 the plot matches "the tongue hump position diagrams which phoneticians have employed for many years" - CONFIRMED | (added) the bark formula is the analytic approximation usually cited as Traunmuller (1990) (ladder credits, formants.js) | https://web.archive.org/web/20110427105916/http://www.ling.su.se/staff/hartmut/bark.htm | Traunmüller's page gives the formula with "[12]" = "H. Traunmüller (1990) Analytical expressions for the tonotopic sensory scale" - MINOR | (added) "it is computed from Peterson and Barney's own hand-measured rows" (index.html:535) | public/strata/vowel-space-formants/index.html | the same page says the rows are "a transmission" via Praat and phonTools and that it will say "recomputed from the rows as distributed by Praat and phonTools, and never from Peterson and Barney's raw data"; its F3 column does not reproduce their Table II. Fixer (K10): declined, tier B listed: Loose wording against the page's own rule to say "as distributed by Praat and phonTools"; the sense (the ladder never passes through the estimator) is true. - MINOR | (added) "both quotations above and the whole of Table II were taken from it" (index.html:676) | public/strata/vowel-space-formants/formants.js | the page quotes four passages of the paper, not two; and formants.js says "The F0 rows for men and women are omitted on purpose" (the amplitude rows are not transcribed either), so not the whole of Table II. Fixer (K11): declined, tier B listed: The page quotes four passages of the paper, not two, and formants.js omits the men's and women's F0 rows and the amplitude rows, so not the whole of Table II; metadata about the page's own reading. ## Findings - [declined] MINOR (K1): "What is not in the paper is how often the position alone recovers the vowel, because in 1952 nobody could run seventy-six cross-validation folds.". Tier B, listed: MINOR, tier B, and not in a sentence being fixed for a WRONG item, so listed. Decided from the checker's note: the first half (the paper gives no classification-by-position rate) holds; the causal clause "because in 1952 nobody could run seventy-six cross-validation folds" is an unsourced claim about 1952 capability, and the paper thanks W. A. Shewhart and M. C. Packer for statistical help (re-read in the pure.mpg.de scan's text layer), so "nobody did" is what the source supports. - [fixed] index.html:372 "reproduce the authors' own printed table to within its rounding" now reads "to within 25 Hz" (rerun: F1 worst 14 Hz with 11 of 30 cells past 5 Hz, F2 worst 23 Hz with 17 of 30); the verifier's string match updated to the new words; Corrected line added. - [dated] index.html "301/301 checks passed": true on 2026-08-17; on 2026-10-18 two checks on shared site chrome had gone stale (the share image's alt text, a bsky.app address in the JSON-LD). The directing instance re-aimed both at what they were meant to test (the og and twitter title tags; executable scripts only), each shown failing on a mutated page, and the verifier passes all 301 again; an Updated 2026-10-18 line says so. - [fixed] index.html:176 "share no harmonic" now reads "share only one harmonic below 5 kHz, at 3000 Hz" (25 x 120 = 12 x 250 = 3000); Corrected line added. - [fixed] relates note to the-sound-the-spelling-forgot: "over six centuries" now "over three centuries" (that layer's timeline runs 1400 to 1700); Corrected line added. - [fixed] same relates note, MINOR in a sentence already being fixed: "published formant values for Middle English" now "published modern formant values for each vowel quality" (that layer uses Catford 2001 and Peterson and Barney 1952); recorded here, no line on the page. - [declined] MINOR (K7): "Running the bootstrap on it would add several seconds to page load for a number nothing uses." (index.html:570). Tier B, listed: MINOR, tier B, not in a sentence being fixed: listed. Checker measured +175 to +610 ms on this box for the 180-point grid with 400 resamples, not several seconds; a slow phone might be slower, but nothing on record measures several seconds. - [fixed] index.html:376 "a fifty-year-old R data file" now reads "1952 measurements out of an R data file" (phonTools first released on CRAN 2012-07-24); Corrected line added. - [fixed] same relates note: "even within one year and one building" now "even among the 76 speakers of one 1952 study" (the paper gives no recording year or room); Corrected line added. - [declined] MINOR (K10): "it is computed from Peterson and Barney's own hand-measured rows" (index.html:535). Tier B, listed: MINOR, tier B, not in a sentence being fixed: listed. Loose wording against the page's own rule to say "as distributed by Praat and phonTools"; the sense (the ladder never passes through the estimator) is true. - [declined] MINOR (K11): "both quotations above and the whole of Table II were taken from it" (index.html:676). Tier B, listed: MINOR, tier B, not in a sentence being fixed: listed. The page quotes four passages of the paper, not two, and formants.js omits the men's and women's F0 rows and the amplitude rows, so not the whole of Table II; metadata about the page's own reading.