--- slug: ppg-pulse 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: 175 confirmed: 145 wrong: 14 unverifiable: 4 observed: 12 stale: 0 --- This pass was the assay line's source-first re-read of The Variability the Camera Invents (batch claude-1018b, tier B): a Claude agent recorded the sources behind every claim, a second Claude agent checked each claim against those sources and the shipped code, and a Claude fixer re-read the evidence for the two WRONG findings before fixing them (the contact geometry described as transmission, and an open-access paper described as paywalled). The twelve MINOR findings are loose wording or overstatements of the page's own checking; at tier B they are listed here and the page is left as it was. The claim lines below are the checker's, with the fixer's decision added to every WRONG and MINOR line. ## Claims - CONFIRMED | after first showing the same code recover 54 bpm from a real smartphone recording whose answer came from a 1000 Hz electrocardiogram running alongside it | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [dek-recovers-54] section 3 "anchor 133007 measures 54.70 bpm", inside the declared 3.0; the reference is the 1000 Hz ECG median (PhysioNet description) - CONFIRMED | at 30 frames a second the frame grid alone typically manufactures about 23.6 ms of heart rate variability out of nothing | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [dek-236] section 9 "sqrt(h^2/2) at 30 fps is 23.57 ms" and the 40 to 200 bpm sweep median is the closed form; arithmetic redone: 33.33^2/2 = 555.6, root 23.57 - CONFIRMED | more than the 17.5 ms measured from ten seconds of electrocardiogram in the recording it ships | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [dek-175] section 10 "the RMSSD of the eight intervals between them is 17.48 ms"; 23.6 > 17.5 - CONFIRMED | at rates whose beats land a whole number of frames apart the same grid invents nothing at all | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [dek-wholeframes] section 9: 50, 60, 72, 75, 90, 100, 120 bpm read 0.00 for a steady heart; my own sweep with the shipped quantiseTimes gives 0.00 at every multiple of 10 that divides 1800 (40, 50, 60, 90, 100, 120, 150, 180, 200) - MINOR | a 30 frames a second grid usually invents about 23.6 ms of it, and at some rates hides the real thing | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive); instrument.js line 804 (panel meanMs 833.33 ms, which is 72 bpm, 25 whole frames) | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. The checker's note holds on its face: the page shows the camera hiding real variability by frame phase (section 10), and at a whole-frame rate its panel inflates it; the plain card's 'at some rates' is loose wording, not a false fact about the world. (checker: [plain-usually-236] The first half holds (section 9 median 23.57). "and at some rates hides the real thing" is not what the page shows: the page demonstrates under-reading only by FRAME PHASE on the anchor ECG (section 10: "over 2000 frame phases the smallest answer is 0.00 ms", "the camera reads high on 89 per cent of phases"), and at a whole-frame RATE its own panel inflates real variability instead (the panel runs at 833.33 ms = 25 frames; section 9: true 5 ms reads "about 15.7", true 20 ms reads 29.99). Suggest "and sometimes, depending on where the frames fall, hides the real thing". Same) - CONFIRMED | each beat drives more blood into the finger, the blood absorbs more of the light, and the level dips | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=4440202 | [dek-dip] Kamshilin: "the intensity of the back-reflected light is maximal in the end-diastole phase, which is well-established fact" and "The general consensus is that the PPG waveform originates from pulsatile variation in the tissue optical density caused by arterial pulsations"; the shipped data agree (verifier section 10: red minimum a median 311 ms after the R wave) - CONFIRMED | Press a fingertip over your phone's rear camera and the light coming back through your skin | https://physionet.org/content/butppg/2.0.0/ | [dek-torch-hint] BUT PPG Methods: "The measured subjects put their 1) index finger to the rear side of the smartphone to cover the camera and the lit LED light ... The video of the finger/ear was recorded"; Kamshilin abstract: light "after interaction with the biological tissue in vivo is modulated at the heartbeat frequency" - CONFIRMED | that page derives heart rate variability from the baroreflex delay that produces it | src/content/strata/six-breaths-a-minute.md | [rel-six-breaths] its dek: "running a live model of the heart-rate baroreflex ... the resonance the delay sets" - CONFIRMED | haemoglobin absorption and light transport in skin, which is why a pulse shows up in a colour channel at all | src/content/strata/why-veins-look-blue.md | [rel-veins] its dek: "a live skin light-transport model driven by real haemoglobin and melanin spectra" - CONFIRMED | That one is time against frequency and depends on window length | src/content/strata/the-window-you-cannot-win.md | [rel-window] its plain card: "make the window longer and you can tell two close pitches apart but no longer see exactly when things happened" - CONFIRMED | a published dataset re-measured from the released signals, with the refusals reported as a result rather than hidden | src/content/strata/when-the-room-becomes-the-sound.md | [rel-room] its dek: "this page measures all of them again from the released audio ... a refusal wherever the recording cannot support an answer (728 of 2,160 octave bands, said out loud)" - CONFIRMED | This page asks for the torch in the constraints it opens the camera with | public/strata/ppg-pulse/instrument.js | [torch-constraint] captureLive passes video: { advanced: [{ torch: true }, ...] } to AW.sensors.captureFrames, which merges o.video into the getUserMedia constraints (public/_kit/sensors.js lines 127 to 130); no applyConstraints call anywhere in instrument.js - UNVERIFIABLE | and most browsers ignore that | MDN browser-compat-data raw files for MediaTrackConstraints/Settings/Capabilities (404 at those paths); chromestatus.com API search for torch and for image-capture constraints in getUserMedia (found Pan/Tilt in getUserMedia, a positive control, nothing on torch); the WebKit test | [torch-most-ignore] No source states how browsers treat torch passed to getUserMedia. The WebKit test itself passes torch:true to getUserMedia and then asserts only the capability, switching the light on with applyConstraints; that is suggestive, not a survey of "most browsers" - CONFIRMED | the other route the API allows and the one WebKit's own test for the feature uses | https://github.com/WebKit/WebKit/commit/9c0b105d80212675839a00325b58df55f408f1c0.patch | [webkit-test-route] mediastreamtrack-video-torch.html: getUserMedia({ video: { width: 640, torch : true } }) then asserts only getCapabilities().torch, then "await stream.getVideoTracks()[0].applyConstraints({ torch : true }); assert_equals(...getSettings().torch, true". The setting is asserted only after applyConstraints, so "the one WebKit's own test uses" is fair - CONFIRMED | red passes through flesh far more easily than green does | https://en.wikipedia.org/wiki/Near-infrared_window_in_biological_tissue | [red-through-flesh] Wikipedia: the window "from 650 to 1350 nanometre (nm) where light has its maximum depth of penetration in tissue ... primarily limited by the light absorption of blood at short wavelengths"; Kamshilin (PMC4440202): "the penetration depth of the green light is less than 1 mm"; Prahl: HbO2 319.6 at 660 nm against 53236 at 540 nm. N1 found no source; these support the sentence - CONFIRMED | near 540 nm both forms absorb | https://omlc.org/spectra/hemoglobin/summary.html | [hb-540] Prahl table rows "540 53236 46592" and "542 53292 48148" (HbO2, Hb): both forms absorb strongly near 540 nm - MINOR | Each of those is something readers reasonably expect from a fingertip and each of them needs either a second wavelength, a calibration against a cuff, a clinical rhythm standard, or all three | https://en.wikipedia.org/wiki/Pulse_oximetry; index.html line on HRV "dressed up as recovery or stress or readiness" | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Loose grouping: three of the four expectations plausibly need a second wavelength, a cuff calibration or a rhythm standard; the fourth (a stress or recovery score) is refused for the frame-grid reason the page gives elsewhere. Not false about any source. (checker: [four-need-more] Page: "Each of those is something readers reasonably expect from a fingertip and each of them needs either a second wavelength, a calibration against a cuff, a clinical rhythm standard, or all three". Holds for blood oxygen (Wikipedia: "A standard pulse oximeter passes two wavelengths of light through tissue"), plausibly for blood pressure and rhythm, but not for the fourth item: a stress or recovery score needs none of the three; the page itself says that score is heart rate variability "dressed up as recovery or stress or readiness", which it refuses for the frame-grid reaso) - CONFIRMED | real fingertip videos taken on real phones | https://physionet.org/content/butppg/2.0.0/ | [brno-real-fingertip] Abstract: "PPG data were collected by smartphones Xiaomi Mi9 and Huawei P20 Pro"; Methods: "index finger to the rear side of the smartphone ... The video of the finger/ear was recorded" - CONFIRMED | each shipped alongside a simultaneous single lead electrocardiogram sampled at 1000 Hz | https://physionet.org/content/butppg/2.0.0/ | [brno-30hz-1000hz] "Each record contains PPG signal and one-lead ECG recorded with a sampling frequency of 30 Hz and 1,000 Hz, respectively"; Methods "Concurrently, the single-lead ECG signal was recorded" - CONFIRMED | The reference rate is the median rate over the ten second window, from QRS detection on that electrocardiogram | https://physionet.org/content/butppg/2.0.0/ | [brno-median-hr] "the QRS complexes were found using robust QRS detector ... and manually verified. Then, the median HR was calculated (one number for the whole 10-second signal) and this number was considered the reference HR" - CONFIRMED | The third specimen we did generate, and it is there to break things | public/strata/ppg-pulse/specimens/synth-specs.json | [third-generated] synth-trap is declared synthetic, generator research/ppg-pulse/make-specimens.mjs; verifier section 1 regenerates it to its recorded digest - CONFIRMED | The same estimator recovered every known answer and stayed silent on both records that have none | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [slate-recovers] section 3: anchor 54.70, control 82.90, trap 72.006, each inside tolerance; section 6: 130078 and still dark frames return no value - CONFIRMED | These pass by returning no rate, and the check is for the absence of the value, not for a flag beside it | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refusals-absence] section 6 "quality-0 record 130078 returns NO value at all, not a hidden one" - CONFIRMED | one that always returns the anchor's published 54 bpm with a tight error bar | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [stub-constant] section 7 "NEGATIVE: a constant estimator returning the anchor's 54 does not arm" - CONFIRMED | Brno record 140067, a different person at 1.5 times the rate | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [control-1p5] section 2: anchor 54, control 82; 82/54 = 1.52 - CONFIRMED | 82 is 28 bpm away from the anchor | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [control-28away] section 2 "anchor butppg-133007 and control butppg-140067 are separable (28 > 6)" - CONFIRMED | own R waves put the instantaneous rate between 53.7 and 55.5 bpm | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [anchor-range] section 10 "the slowest beat is 53.74 bpm", "the fastest is 55.45 bpm" - CONFIRMED | manually verified beat marks for this record give 54.50 | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [anchor-5450] section 8b "so the dataset's own marks put it at 54.50 bpm against the column's integer 54"; the shipped 133007 .qrs bytes are identical to PhysioNet's (my cmp) - CONFIRMED | chosen by a rule fixed before it was run: the first quality-0 record in the declared fitting split | public/strata/ppg-pulse/specimens/butppg-sample.json | [refusal-130078-rule] the first record in the sample list is {"id":"130078","quality":0,"split":"fit"}, so it is the first quality-0 record of the fitting split in its order; section 2 confirms quality 0. "fixed before it was run" is the author's account - CONFIRMED | It has a sharp systolic upstroke and a second, smaller bump as the aortic valve closes | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=9280335 | [pulse-notch] "If onset is related to the beginning of systolic ejection, then diastolic denotes the end of ejection as the aortic valve closed". The same paper attributes the notch and diastolic peak to "wave reflections of the periphery"; the page says only when the bump comes, which the source supports - CONFIRMED | and a waveform with a bump like that has real energy at twice the heart rate | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [bump-twice-hr] section 8 "the second harmonic carries a median of 10.4 per cent of the fundamental's power" over 432 channel-records: measured energy at twice the rate on real recordings. N1 found no outside source; the page's own measurement supports it - CONFIRMED | hands a perfectly calm person a rate of 144 | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [trap-144] section 4 "an argmax of the spectrum reports 144.001 bpm for this specimen" - CONFIRMED | the generator was handed a second harmonic larger than its fundamental | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [trap-built] section 4 "the generator really was given a larger second harmonic than fundamental", ratio 1.67 - CONFIRMED | The second harmonic is the taller of the two on 20 of those channels, and 0 of the 20 belong to a record these quality gates accept. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [harm-432-20-0] section 8 "432 channel-records have twice their reference rate inside the search band", "it is the taller line on 20 of the 432", "and on none of the records these gates accept" - CONFIRMED | on this corpus the condition never once arises in a recording good enough to report | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [harm-rare] section 8 "and on none of the records these gates accept"; largest accepted ratio "under 30 per cent (25)" - OBSERVED | What it cannot tell you is how often the case arises, and the page used to assert that too | research/ppg-pulse/README.md; git log -S "commonest bug" 9ddc1e9045 -- public/strata/ppg-pulse/index.html | [harm-used-to-assert] README: "The page used to say the second harmonic is 'often' taller ... Neither was measured"; git history shows the "commonest bug" text entering and leaving index.html (571a21f44da, a618079f847, 2026-08-17) - CONFIRMED | Across all 200 shipped records the same test fired on 27 channels in 20 of them, and it is not free: on 26 of those 27 channels | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [halved-27-20-26] section 8 "the half-harmonic test fires on 27 channels in 20 of the 200 records", "on 26 of those 27 channels it moved the rate AWAY from the reference" - CONFIRMED | Every one of those records was refused by the quality gates anyway, so on this sample the test never changed a number the page would have printed | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [halved-no-change] section 8 "while none of those 20 records survived the quality gates, so no printed number moved" - CONFIRMED | This page subtracts a two second centred moving average | public/strata/ppg-pulse/ppg.js | [detrend-2s] CFG detrendSeconds 2, detrendPadding reflect; channelRate subtracts movingAverage(x, win, padding) (my dump of P.CFG) - UNVERIFIABLE | Pretend the record is surrounded by silence, which is what a plain same mode convolution does and what most quick implementations do | https://numpy.org/doc/stable/reference/generated/numpy.convolve.html | [same-mode-zero] First half confirmed: numpy convolve mode "same" "Contains boundary effects, where zeros are taken into account". "what most quick implementations do" is a survey of other people's code; no source found. Also in instrument.js line 535 - CONFIRMED | Record 133007, published reference 54 bpm, same estimator, one word changed. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [pad-137] section 5: reflect 54.75/54.72/54.70, zero 41.08/48.60/48.95, "the worst swing between the two conventions is 13.7 bpm"; heading "fourteen" rounds 13.7 - CONFIRMED | leaving a large slow swing that lands at the bottom of the search band and gets reported as a pulse | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [pad-baseline-low] section 5: zero-padded red 41.08 bpm against 54; band starts at 0.6 Hz = 36 bpm (CFG bandLoHz 0.6) - CONFIRMED | The three colour channels must agree with each other to within 3.0 bpm | public/strata/ppg-pulse/ppg.js | [gate-spread] CFG spreadMaxBpm 3 (my dump of P.CFG); verifier section 12 "quoted in its own sentence and is the one in CFG" - CONFIRMED | The in band signal to noise must reach 4.0 | public/strata/ppg-pulse/ppg.js | [gate-snr] CFG snrMin 4 (my dump of P.CFG); verifier section 12 "quoted in its own sentence and is the one in CFG" - CONFIRMED | must stay under 3.0 per cent | public/strata/ppg-pulse/ppg.js | [gate-mod] CFG modMaxPercent 3 (my dump of P.CFG); verifier section 12 "quoted in its own sentence and is the one in CFG" - CONFIRMED | The modulation index, which is the amplitude of the fundamental over the channel's mean level | public/strata/ppg-pulse/ppg.js | [gate-modindex] channelRate: modPercent: 100 * amp / Math.abs(dc), amp the spectral peak amplitude with the window gain divided out - CONFIRMED | correlate with itself one beat later at 0.35 or better | public/strata/ppg-pulse/ppg.js | [gate-repeat] CFG repeatMin 0.35 (my dump of P.CFG); verifier section 12 "quoted in its own sentence and is the one in CFG" - MINOR | a real fingertip pulse moves the light by tenths of a per cent and anything larger is your finger moving | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=%22peripheral%20perfusion%20index%22%20AND%20AUTH:%22Lima%20A%22%20AND%20PUB_YEAR:2002&resultType=core&format=json | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Unscoped generalisation: Lima et al. 2002 report fingertip pulse-oximeter perfusion index 0.3 to 10.0 per cent (a different measure, peak-based, on a different sensor); for this camera instrument the page's own accepted means are under one per cent, so the gate stands and the wording is loose. (checker: [pulse-tenths] Page: "must stay under 3.0 per cent, because a real fingertip pulse moves the light by tenths of a per cent and anything larger is your finger moving." As a statement about fingertips it is too strong: Lima, Beelen, Bakker, Crit Care Med 2002 (doi 10.1097/00003246-200206000-00006), 108 healthy adults, fingertip pulse oximeter: "values ranged from 0.3 to 10.0, median 1.4 (inner quartile range, 0.7-3.0)" per cent. That index is pulsatile over static light (peak-based), not this page's fundamental amplitude, and an oximeter is not a phone camera, so the 3.0 gate may be right for th) - CONFIRMED | The database holds | https://physionet.org/content/butppg/2.0.0/ | [corpus-3888] "The data comprises 3,888 10-second recordings"; my recount of the shipped quality-hr-ann.csv (sha256 identical to PhysioNet's copy fetched 2026-10-19) gives 3888 rows - CONFIRMED | expert consensus judged 830 of them usable for estimating a heart rate | public/strata/ppg-pulse/specimens/quality-hr-ann.csv | [corpus-830] my recount: 830 rows with quality 1; file byte-identical to https://physionet.org/files/butppg/2.0.0/quality-hr-ann.csv - CONFIRMED | a majority of the 3 or 5 annotators landing within 5 bpm of the electrocardiogram) is the definition given in that database's description | https://physionet.org/content/butppg/2.0.0/ | [quality-definition] "Only the 'good' annotations were considered, meaning those HR values that differed from the reference HR by less than or equal to 5 bpm ... When most annotators (2 out of 3 or 3 out of 5) provided good annotations, the signal was considered of 1" - CONFIRMED | it is a quality benchmark built to contain bad recordings on purpose | https://physionet.org/content/butppg/2.0.0/ | [benchmark-bad-on-purpose] "created ... for the purpose of evaluating PPG quality and estimation of heart rate"; "The signals were recorded even at rest or during various activities ( higher pressure ..., walking, coughing, laughing, changing the light , talking)" - CONFIRMED | of the recordings are of an ear held against the front camera rather than a finger on the rear one | public/strata/ppg-pulse/specimens/subject-info.csv | [ear-1962] my recount: 1962 rows with Ear/finger 0 (1926 finger); file byte-identical to PhysioNet's subject-info.csv - CONFIRMED | an ear held against the front camera rather than a finger on the rear one | https://physionet.org/content/butppg/2.0.0/ | [ear-front-camera] "2) ear to the front camera of the smartphone in the telephoning position" - CONFIRMED | were made while the subject was walking, coughing, laughing, talking, pressing harder, moving the skin across the lens or changing the light | public/strata/ppg-pulse/specimens/subject-info.csv | [motion-1164] my recount: 1164 rows with Motion other than 0 (codes 1 to 7, one row "2;3") - CONFIRMED | each an artefact the authors induced deliberately | https://physionet.org/content/butppg/2.0.0/ | [motion-codes] "motion information: 0 (rest), 1 ( higher pressure of finger/ear on camera), 2 (moving the finger/ear on the lens), 3 (walking), 4 (coughing), 5 ( laughing), 6 (changing the light) , 7 (talking)": a protocol of named activities, so "induced deliberately" is fair - CONFIRMED | 877 of 1,354 recordings still could not be read to within 5 bpm of the electrocardiogram by a majority of the three or five experts who tried | public/strata/ppg-pulse/specimens/subject-info.csv | [finger-rest-877] my recount: 1354 finger-at-rest rows, 877 of them quality 0 in quality-hr-ann.csv; quality 0 means a majority of the 3 or 5 annotators missed by more than 5 bpm (PhysioNet description) - CONFIRMED | About two in three | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [two-in-three] section 11 "the finger-at-rest failure rate is 64.8 per cent"; 877/1354 = 0.648 - CONFIRMED | Of 100 held out records the gate accepted 15 and refused 85 | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-15-85] section 8 "the gate accepts 15 of them", "and refuses 85" - CONFIRMED | Of those accepted, 13 were within 5 bpm of the electrocardiogram reference and 2 were not | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-13-2] section 8 "13 of the accepted are within 5 bpm", "and 2 are not" - CONFIRMED | It also refused 24 records the dataset's annotators had called usable | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-24-refused] section 8 "it also refuses 24 records the annotators called usable" - CONFIRMED | One of the fifteen it accepted is record 133007, the anchor this page picked as its known-answer specimen; drop it and the figures are 14 accepted, 12 of them good | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-anchor-in] section 8 "the page's own anchor 133007 is one of the accepted", "without it the headline is 14 accepted and 12 good" - CONFIRMED | 9 of them are recordings of an ear, and the gate accepted 0 of those | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-ear-9-0] section 8 "9 of the held-out hundred are ear recordings", "and the gate accepted none of them" - CONFIRMED | 24 were made under an induced artefact | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-24-artefact] section 8 "24 of the hundred were made under an induced artefact" - CONFIRMED | Take only the 69 that are a finger at rest, which is the condition you will be in: the gate accepted 14 of them and 13 came back within 5 bpm, the single miss by 5.9 bpm. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-fr-69] section 8 "69 are a finger at rest", "of which the gate accepted 14 and 13 were within 5 bpm", "the worst miss among those is 5.9 bpm" - CONFIRMED | it still refused 55 of those 69 | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-fr-refused-55] section 8 "and it refuses 55 finger-at-rest records, which is most of the 85 refusals" - CONFIRMED | Most of them are somebody sitting still with a finger on the lens, getting nothing | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [held-most-sitting] section 8: 55 of the 85 refusals are a finger at rest - OBSERVED | The four gate thresholds were fitted on the declared fitting split | research/ppg-pulse/README.md | [held-sweep-split] README "The four gate thresholds were chosen on the fitting split. Only the held-out split is quoted on the page." Verifier section 8 confirms the split exists and is disjoint ("no record appears in both halves"); the fitting itself is the author's record, not recreated - CONFIRMED | "reference QRS complex annotations from ECG, manually verified". The first is what | https://physionet.org/files/butppg/2.0.0/ANNOTATORS | [annotators-quote] fetched 2026-10-19: "qrs\treference QRS complex annotations from ECG, manually verified", byte-identical to the shipped specimens/ANNOTATORS (cmp) - CONFIRMED | On 44 of the 200 shipped records the two references differ by more than 5 bpm | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refs-44-of-200] section 8b "the two disagree by more than 5 bpm on 44 of them" - CONFIRMED | Scored against the annotation column 13 of the fifteen accepted records are good; scored against the manually verified beat marks it is 14 | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refs-13-14] section 8b "scored against the HR column, 13 of the 15 accepted records are good", "scored against the manually verified marks, 14 are" - CONFIRMED | one of them by 57.8 bpm | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refs-worst-578] section 8 "the worst error among accepted records is 57.8 bpm"; "148023 is reported at 44.2 bpm against a reference of 102" - CONFIRMED | against the dataset's other reference, its manually verified beat marks, it is one of the fifteen | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refs-marks-one-of-15] section 8b "exactly one record changes verdict between them: 149040", so one of 15 misses against the marks - CONFIRMED | Both of the two carried a quality label of unusable, written by expert annotators | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refs-both-unusable] section 8 "both wrong answers are on records the annotators called unusable" - CONFIRMED | the 57.8 bpm miss was recorded while its subject was walking | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [refs-walking] section 8 "the 57.8 bpm miss, 148023, carries motion code 3, which the dataset defines as walking" - CONFIRMED | A 2024 scoping review of resting heart rate from contact-based smartphone photoplethysmography found correlations against electrocardiography of r = 0.98 to 1 | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=10937558 | [mather-corr] "correlations ranging from r = .98 to 1" - CONFIRMED | with limits of agreement running from 0.29 bpm in the best case | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=10937558 | [mather-loa-029] Table 8, Bolkhovsky, Scully and Chon: "iPhone 4s supine ... LoA: 0.29" and tilt 0.29; I read every LoA in Table 8 (0.29, 0.29, 3.20, 1.40, -1.43/1.03, -1.36/1.56, -0.83/0.58, -1.27/1.13, -5.58/5.52, -1.98/2.07): 0.29 is the narrowest - CONFIRMED | to minus 5.58 and plus 5.52 bpm in the widest | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=10937558 | [mather-loa-widest] Table 8, Nam et al.: "LoA: -5.58, 5.52", the widest of the ten tabulated (list in the line above) - CONFIRMED | Those come from supervised recordings with cooperative subjects | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=10937558 | [mather-controlled] "However, agreement was established under highly controlled conditions." - CONFIRMED | Verkruysse and colleagues showed in 2008 that in ambient light, reflected off a face, the green channel carries the strongest plethysmographic signal | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=2717852 | [verk-2008-green] "Plethysmographic signals were measured remotely (> 1m) using ambient light ... Although the green channel featuring the strongest plethysmographic signal, corresponding to an absorption peak by (oxy-) hemoglobin"; "measured on the human face with normal ambient light"; Optics Express 2008 - CONFIRMED | Their own wording is "corresponding to an absorption peak by (oxy-) hemoglobin" | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=2717852 | [verk-quote] verbatim in the abstract: "corresponding to an absorption peak by (oxy-) hemoglobin" - CONFIRMED | The Brno authors annotated from red | https://physionet.org/content/butppg/2.0.0/ | [brno-annotated-red] "From the video, only the red color channel [8] was considered for annotations" - WRONG | the light goes through the finger rather than bouncing off it, and red | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=4440202; https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1371/journal.pone.0091205&resultType=core&format=json; https://physionet.org/content/butppg/2.0.0/ | [fixed] Re-read 2026-10-18. BUT PPG v2.0.0 Methods (physionet.org/content/butppg/2.0.0/): subjects put their "index finger to the rear side of the smartphone to cover the camera and the lit LED light", so lamp and camera are side by side. Kamshilin et al. 2015 (PMC4440202, efetch full text, Introduction), the page's own source: reflection mode is where "both the illuminating light source and photodetector are situated alongside each other"; transmission mode is where they "are situated at the opposite sides of tissue, e.g., across a fingertip". Matsumura et al. 2014 (PMC3949790, abstract): smartphone PPG "has been achieved with reflection mode photoplethysmography (PPG), by using a smartphone's embedded flash as a light source and the camera as a light sensor". So "the light goes through the finger rather than bouncing off it" (index.html) and its copies (instrument.js "a geometry where the light goes straight through", "as light through a finger must be"; README and verifier "transmission through a finger") describe the geometry falsely. The second half, red passing through flesh more easily than green, is untouched. (checker: [red-hb-660] Page (index.html, Green, red, blue section): "Contact geometry is a different problem: the light goes through the finger rather than bouncing off it". A finger on a phone covers "the camera and the lit LED light" side by side (BUT PPG Methods), which the page's own source Kamshilin defines as reflection mode: "iPPG usually operate in reflection mode where both the illuminating light source and photodetector are situated alongside each other", against transmission mode where they "are situated at the opposite sides of tissue, e.g., across a fingertip". Matsumura et al. 2014 (PLoS O) - CONFIRMED | on the anchor's blue channel the amplitude is 0.267 per cent of the level and the peak to peak swing of the same detrended trace is 0.737 | public/strata/ppg-pulse/ppg.js | [chan-blue-0267-0737] recreated: channelRate on 133007 with the shipped CFG gives blue amp 0.267 per cent and detrended peak to peak 0.737 per cent of DC 35.18 (red 0.040/0.124, green 0.178/0.484); verifier section 3 pins 0.267 - CONFIRMED | Red is the brightest channel through a fingertip by a long way and carries the smallest fractional pulse | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [red-brightest] section 3: anchor DC red 238.19, green 52.86, blue 35.18; red fractional pulse 0.040 per cent, the smallest; section 8 mean red level 240.4 - CONFIRMED | Green and blue are within a few per cent of each other | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [blue-largest-11] section 8: mean fractional pulse green 0.713, blue 0.747 per cent (4.8 per cent apart); "blue carries the largest fractional pulse in 11 of 15" - CONFIRMED | is a fifteen year old consumer feature and there are a thousand apps that do it | https://en.wikipedia.org/wiki/Azumio | [fifteen-year-feature] Wikipedia: "Instant Heart Rate, released in October 2010 ... illuminates a user's finger with a phone's flashlight feature and uses the phone's camera to analyze color changes in the finger tip"; 2010 to the page's 2026-08-17 is fifteen years and ten months. "a thousand apps" is a round figure I could not source and read as one - UNVERIFIABLE | Many of them also sell a second number: heart rate variability, usually RMSSD | Mather review text (PMC10937558) searched for app counts; no app-store survey of HRV-selling camera apps found | [apps-sell-hrv] Mather notes only "a considerable increase in the number of mobile apps". "Many of them also sell ... RMSSD" is unsupported but unlikely to be false - CONFIRMED | RMSSD, the root mean square of successive differences between beats | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [rmssd-definition] "The root mean square of successive differences between normal heartbeats (RMSSD)" - CONFIRMED | At 30 frames a second the spacing is h = 33.33 ms | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [h-3333] section 9 "the frame spacing at 30 fps is 33.33 ms"; 1000/30 - CONFIRMED | each has variance h²/12 | https://en.wikipedia.org/wiki/Quantization_(signal_processing) | [uniform-variance] mean squared rounding error Delta^2/12; for uniform on +-h/2 the variance is h^2/12 by direct integration - CONFIRMED | with variance 6σ² | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [variance-6sigma] algebra redone: the second difference of beat times carries error e(i) - 2e(i+1) + e(i+2), variance (1+4+1) sigma^2 = 6 h^2/12 = h^2/2; section 9 checks it numerically (true 20, 30, 50, 80 ms within 1 ms) - CONFIRMED | RMSSD_measured = sqrt( RMSSD_true^2 + h^2 / 2 ) | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [closed-form] follows from the line above: RMSSD_measured^2 = RMSSD_true^2 + h^2/2; section 9 "on a clean signal the frame-locked measurement matches the closed form (30.82)" - CONFIRMED | so a perfectly steady heart, with no variability whatsoever, reads as sqrt(33.33²/2) = 23.6 ms. Usually. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [steady-236] section 9 "sqrt(h^2/2) at 30 fps is 23.57 ms"; "Usually." is honest: section 9 sweep minimum 0.00, maximum above 33 - CONFIRMED | across the 15 studies that reported RMSSD in Nunan's review of 44 studies covering 21,438 healthy adults, the average reported RMSSD was 42 ms and the study averages themselves ran from 19 to 75 ms | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [nunan-42] Table 6 "Nunan et al. ( 17 ) short-term norms": "RMSSD (ms) 42 (15) 19-75 15" (en dash in source) - CONFIRMED | Nunan's review of 44 studies covering 21,438 healthy adults | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1111/j.1540-8159.2010.02841.x&resultType=core&format=json | [nunan-44-21438] Nunan abstract "Forty-four studies met the pre-set inclusion criteria involving 21,438 participants"; Shaffer: "21,438 healthy adult participants" - CONFIRMED | It is a spread of study averages rather than a range that individual people occupy | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [nunan-min-age] Table 6 gives Range beside a "Studies" count (15 for RMSSD), and Nunan's abstract reports "large interindividual variations (up to 260,000%)", so 19 to 75 cannot be a range of people. N1's excerpt (three populations with minimum age 40) is beside the point; the claim stands on Table 6 and the Nunan abstract - CONFIRMED | If its interval happens to be a whole number of frames, every beat lands at the same point between two of them, every rounding error is identical, the differences are exactly zero, and the invented variability is not 23.6 ms but nothing at all | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [comb-whole-frames] section 9: 50, 60, 72, 75, 90, 100, 120 bpm read 0.00 - MINOR | the exceptions are not rare curiosities: they are the round numbers | recreated: the shipped P.quantiseTimes over 300 beats at every multiple of 10 bpm, 30 fps (public/strata/ppg-pulse/ppg.js) | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Overstatement: only round numbers that divide 1800 at 30 fps read zero (the checker's recreation gives 70 and 80 bpm at 25.18 and 33.33 ms). The sentence's point, that exceptions are common rates and not curiosities, stands; the wording is loose. (checker: [comb-round-numbers] Page: "biased upward at most rates, by roughly the amount the closed form says, and the exceptions are not rare curiosities: they are the round numbers." Only round numbers that divide 1800 are exceptions: my run gives 40, 50, 60, 90, 100, 120, 150, 180, 200 bpm at 0.00 ms, but 70 bpm 25.18 ms, 80 bpm 33.33 ms, 110 bpm 28.44 ms (25.71, 22.5, 16.36 frames). 80 bpm, a typical resting round number, reads 33.33 ms, the top of the curve and 41 per cent above the closed form. Suggest "many round numbers, not all: 60 and 90 read zero, 80 reads the most") - CONFIRMED | A series of 300 beats is generated with an exactly known RMSSD | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [rm-300-beats] section 9 "300 beats at seed 3 is what the shipped panel actually runs"; instrument.js rrSeries({ meanMs, rmssdMs: RM.trueMs, beats: RM.beats, seed: 3 }) - OBSERVED | with a realistic dicrotic shoulder the peak picker finds two maxima per beat and what you would be watching is the picker, not the grid | research/ppg-pulse/README.md | [rm-single-hump] README "The frame-grid demonstration uses a single-humped pulse on purpose. With a realistic dicrotic shoulder the peak picker finds two maxima per beat"; the verifier does not test it (N1 attributed the quote to the verifier output; it is the README) - CONFIRMED | they agree closely for a true RMSSD of about 15 ms and above | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [agree-15] section 9: true 20, 30, 50, 80 ms agree within 1 ms; "so the model overstates the damage below a true 15 ms" - CONFIRMED | At a true 5 ms and 30 fps the closed form says 24.1 ms and the actual rounding gives about 15.7, averaged over six independent runs of 600 beats | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [five-ms-241-157] section 9 "at a true 5 ms the closed form predicts 24.1 ms", "averaged over 6 seeds of 600 beats, gives about 15.7 ms" - CONFIRMED | so its rounded row sits at 18.7 at that setting rather than at the average | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [five-ms-187] section 9 "the panel itself, one 300-beat series at seed 3, shows 18.69 ms there" - CONFIRMED | At 10 per cent of the pulse amplitude the parabolic estimate is worse than the floor it was meant to remove, and raising the frame rate hardly helps | recreated: runPipeline as in research/ppg-pulse/verify-ppg-pulse.mjs, with the shipped public/strata/ppg-pulse/ppg.js, at 30/60/120/240 fps | [noise-parabola-worse] noise 0.1, true 20 ms: refined 54.04 / 53.32 / 48.52 / 42.16 ms against floors 23.57 / 11.79 / 5.89 / 2.95; worse than the floor at every rate, and 8x the frame rate cuts it by about a fifth - CONFIRMED | Turn the noise to zero and it works completely, at every frame rate, and the 23.6 ms floor simply disappears | recreated: runPipeline as in research/ppg-pulse/verify-ppg-pulse.mjs, with the shipped public/strata/ppg-pulse/ppg.js, at 30/60/120/240 fps | [noise-zero-parabola] noise 0, true 20 ms: refined 20.00 at all four frame rates (the verifier checks only 30 fps) - CONFIRMED | Parabolic refinement of the fiducial point is the standard remedy for a coarse sampling grid, and the 1996 Task Force report on heart rate variability is the usual citation for it at low sampling rates | https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11278032/fullTextXML | [parabola-standard-remedy] A 2024 J Clin Med paper (PMC11278032): "Parabolic interpolation refined the R wave fiducial point [ 16 ]", ref 16 the Task Force, Circulation 1996;93:1043-1065. That is the pairing the page describes; "usual" rests on this instance and I did not read the report itself (OpenAlex: closed access, no repository copy) - OBSERVED | its full text sits behind publisher paywalls that returned 403 to every retrieval we attempted on 2026-08-17 | research/ppg-pulse/README.md | [taskforce-403] README: "On 2026-08-17 ahajournals.org (article and PDF), academic.oup.com and escardio.org all returned HTTP 403." The second round ("again ... while writing this sentence") is not kept. OpenAlex lists the DOI as closed access, consistent with "paywalls" - CONFIRMED | The anchor recording ships with its own electrocardiogram at 1000 Hz | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [ecg-1000hz] section 10 "the shipped electrocardiogram is sampled at 1000 Hz", "runs for 10000 samples"; header 133007_ECG.hea "133007_ECG 1 1000 10000", byte-identical to PhysioNet's - CONFIRMED | At 1000 Hz the sample grid can only put sqrt(h²/2) = 0.71 ms into a beat time | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [ecg-sample-grid] section 10 "sqrt(h^2/2) at 1000 Hz is 0.71 ms"; sqrt(0.5) = 0.707 - CONFIRMED | the whole of the improvement is the move onto the R wave | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [ecg-parabola-worth] section 10: marks 19.04, R wave 17.40, parabola 17.48, "the parabola on top of that is worth 0.08 ms, upward" - CONFIRMED | rests on eight intervals | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [ecg-eight-intervals] section 10 "nine beats are detected", "the RMSSD of the eight intervals between them is 17.48 ms", "its standard error on eight intervals is about 4.67 ms" (about a quarter of 17.48) - CONFIRMED | "may be potentially as reliable as that derived from the analysis of the electrocardiogram, provided that fs ≥ 25 Hz sampling frequency is used" | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1088/1361-6579/aa5efa&resultType=core&format=json | [choi-quote] abstract: "may be potentially as reliable as that derived from the analysis of the electrocardiogram, provided that f s >=25 Hz sampling frequency is used" (source uses U+2A7E) - CONFIRMED | Choi and Shin took finger recordings sampled at 10 kHz, decimated them, and measured what the sampling rate did to the variability statistics | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1088/1361-6579/aa5efa&resultType=core&format=json | [choi-10khz] abstract: "We generated 5000, 2500, 1000, 500, 250, 100, 50, 25, 20, 15, 10, 5 Hz down-sampled photoplethysmograms from 10 kHz-sampled PPGs". "finger" is in the full text only (N1 quotes "recorded at left index finger"); iopscience gave me a Radware captcha, so that word is unchecked by me - CONFIRMED | with RMSSD among the variables that differed significantly only at 15 Hz and below | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1088/1361-6579/aa5efa&resultType=core&format=json | [choi-rmssd-15] "the NN50 and pNN50, SDSD and RMSSD, LF/HF and nHF, SDNN, TP and nLF analysis had significant differences at f s <= 20 Hz, f s <= 15 Hz, f s <=10 Hz; f s = 5 Hz, respectively": RMSSD at 15 Hz and below - UNVERIFIABLE | theirs is a significance test on 28 subjects | iopscience article and PDF by curl (Radware captcha, 14 KB challenge page); WebFetch (redirect to validate.perfdrive.com); Wayback availability API (HTTP 429); the Europe PMC abstract (no subject count) | [choi-28] The abstract has no subject count (positive control: the same abstract text gives "10 kHz"), yet the page says it quotes only the abstract; N1 quotes the full text: "the dataset from 28 subjects". The figure may well be right; its provenance is not the abstract - WRONG | the full text sits behind a paywall and this page quotes only the abstract | https://api.crossref.org/works/10.1088/1361-6579/aa5efa; https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1088/1361-6579/aa5efa&resultType=core&format=json | [fixed] Re-read 2026-10-18. Crossref api.crossref.org/works/10.1088/1361-6579/aa5efa, license list: {"content-version": "vor", "URL": "http://creativecommons.org/licenses/by/3.0/", start 2017-02-23}. Europe PMC REST record for DOI:10.1088/1361-6579/aa5efa: license "cc by", fullTextUrl availability "Open access" at https://iopscience.iop.org/article/10.1088/1361-6579/aa5efa/pdf. The article is open access under CC BY 3.0 from publication, so "the full text sits behind a paywall" (index.html Sources) and README "The full text is paywalled" are false; that the page quotes only the abstract stays true. (checker: [choi-paywall] Page (Sources): "the full text sits behind a paywall and this page quotes only the abstract". The article has been open access since publication: Crossref license {"content-version":"vor","URL":"http://creativecommons.org/licenses/by/3.0/","start":2017-02-23}; the Europe PMC record the page says it read gives license "cc by" and fullTextUrl "availability":"Open access" (iopscience PDF); OpenAlex oa_status "hybrid", license "cc-by"; Semantic Scholar "HYBRID", "CCBY"; N1's own excerpt from the article page reads "The following article is Open access". iopscience shows scripts a Ra) - CONFIRMED | "short-term variability is somewhat overestimated by PRV, which reflects coupling effects between respiration and the cardiovascular system" | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.ijcard.2012.03.119&resultType=core&format=json | [schafer-quote1] verbatim in the abstract - CONFIRMED | "often to an inacceptable extent" | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.ijcard.2012.03.119&resultType=core&format=json | [schafer-quote2] "Physical activity and some mental stressors seem to impair the agreement of PRV and HRV, often to an inacceptable extent." - CONFIRMED | the agreement is good enough at rest | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.ijcard.2012.03.119&resultType=core&format=json | [schafer-good-rest] "Results speak in favor of sufficient accuracy when subjects are at rest" - CONFIRMED | and that movement breaks the agreement | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.ijcard.2012.03.119&resultType=core&format=json | [schafer-movement] "Physical activity and some mental stressors seem to impair the agreement" (the page says "movement"; the source says physical activity, close enough) - CONFIRMED | The intervals a camera measures are pulse intervals, not the heart's own: the time from a beat to the arrival of its pressure wave at your fingertip varies with blood pressure and with breathing | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.ijcard.2012.03.119&resultType=core&format=json; https://en.wikipedia.org/wiki/Pulse_wave_velocity | [pulse-not-heart] Schafer: PRV "reflects coupling effects between respiration and the cardiovascular system"; Wikipedia: "PWV intrinsically varies with blood pressure", so transit time to the fingertip varies with pressure - CONFIRMED | the conventional short term standard for a variability measurement is five minutes | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [five-minute-standard] "While the conventional minimum recording is 5 min" - CONFIRMED | this page requires at least 60 seconds and 30 accepted beats before it would consider one | public/strata/ppg-pulse/ppg.js | [rmssd-min-60-30] CFG rmssdMinSeconds 60 (my dump) - CONFIRMED | this page requires at least 60 seconds and 30 accepted beats | public/strata/ppg-pulse/ppg.js | [rmssd-min-beats] CFG rmssdMinBeats 30 (my dump) - MINOR | a twenty second fingertip capture gives neither | public/strata/ppg-pulse/ppg.js | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Loose wording: a twenty second capture never gives 60 seconds and gives 30 beats only at 90 bpm or more, so 'neither' holds at resting rates; the page's refusal stands either way. (checker: [capture-20s] Page: "this page requires at least 60 seconds and 30 accepted beats before it would consider one, and a twenty second fingertip capture gives neither." 20 s never gives 60 s, but it gives 30 or more beats whenever the rate is 90 bpm or above (20 s x 90/60 = 30; at 100 bpm about 33). Suggest "gives the first never and the second only above 90 beats a minute", or "gives neither at a resting rate") - MINOR | Nothing on this page is a stored result | public/strata/ppg-pulse/index.html; public/strata/ppg-pulse/instrument.js | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Overclaim about the page's own checking (the brief's MINOR class): a few prose figures (57.8, 0.267, 0.737, 15.7) are typed, though true by the offline verifier. (checker: [check-stored] Page: "Every figure above was computed in your browser from the files in ./specimens/ by ppg.js ... Nothing on this page is a stored result." Several data figures in the prose are typed and never recomputed or audited in the browser: "one of them by 57.8 bpm" (the browser audits fig-* spans, not this sentence), "0.267 per cent of the level and the peak to peak swing ... is 0.737", "about 15.7, averaged over six independent runs of 600 beats" (the browser never runs six seeds of 600; grep finds none of 57.8, 0.737, 0.267, 15.7 in instrument.js). They are true (offline verifier, m) - CONFIRMED | The interval on a live rate combines half the width of the spectral line with the spread of the estimate across overlapping sub windows | public/strata/ppg-pulse/ppg.js | [check-interval] estimate(): terms = [widthBpm / 2, sd(subs)], uncertainty = root sum of squares - MINOR | A ten second record has a raw spectral bin of 6.0 bpm | public/strata/ppg-pulse/ppg.js | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Attribution slip inside the page's own method note: ppg.js zero-pads to CFG.fftPoints before bandPeak's parabola, so 'only because of parabolic interpolation' names one of two interpolations; the caution (no added resolution) stands. (checker: [check-spectral-bin] Page: "A ten second record has a raw spectral bin of 6.0 bpm and the estimates are finer than that only because of parabolic interpolation on the line." 6.0 is right (1/10 s = 0.1 Hz). "only because of parabolic interpolation" is not what the code does: spectrumOf zero-pads to CFG.fftPoints 32768 ("The amplitude spectrum of a record, zero-padded to CFG.fftPoints"), about 109x for a 300-sample record, so the bins are already about 0.055 bpm apart before bandPeak's parabola. Both are interpolation and neither adds resolution, so the caution stands; the attribution is wrong) - CONFIRMED | the binomial interval is printed with it | public/strata/ppg-pulse/instrument.js | [check-binomial] renderSweep prints "That is X per cent, with a 95 per cent interval of A to B per cent" - CONFIRMED | The 200 record sample covers only the three channel layout | public/strata/ppg-pulse/specimens/butppg-sample.json | [check-three-channel] sample minId 130000; every record from 112001 on is three-channel (headers fetched: 112001, 121076, 125001, 129108, 133007 carry PPG_R/G/B; 111004 and 100001 are the older layout) - CONFIRMED | which replaces the camera with a generated video of exactly known rate and checks that the page reports it, and that a second video at a different rate moves the answer | scripts/check-live-sensor.mjs | [check-live] header: "4. an injected signal of a DIFFERENT known value moves the reported number with it"; research/anchor-wave/cases/ppg-pulse.json injects 72 and 120 bpm Y4M files - CONFIRMED | It does not cover a torch, an automatic exposure loop fighting the signal, or a real finger | research/ppg-pulse/README.md | [check-not-covered] README: "The live gate cannot test a torch, an automatic exposure loop fighting the signal, or a real finger"; the harness uses Chromium fake capture - MINOR | reruns every number above from the shipped files, regenerates the synthesised specimens and compares their hashes | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive); mutation copies mut-a/b/c in my scratch | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Overstatement of the verifier's reach: the checker's mutation shows one figure (57.8 in one sentence) is pinned by a token another sentence also satisfies; every other rerun held. (checker: [verifier-claims] It does rerun the figures (sections 1 to 13 all ran for me). Step 6 mutation: changing one sentence's figure in a scratch copy, "one of them by 57.8 bpm" to "47.8 bpm", left the verifier green (425 passed; only the og-image failure), because its check is html.includes("57.8 bpm"), satisfied by the second sentence "the 57.8 bpm miss was recorded while its subject was walking". The same edit to 0.737 and to 15.7 turned it red (positive controls). So one printed figure is unpinned, and the browser does not audit it either (entry check-stored). Suggest pinning the sentence, not t) - CONFIRMED | Nemcova A, Smisek R, Vargova E, Marsanova L, Vitek M, Smital L, Filipenska M, Sikorova P, Galik P. | https://api.datacite.org/dois/10.13026/tn53-8153 | [cite-brno-dataset] DataCite (Crossref 404s because it is a DataCite DOI): creators Nemcova, Smisek, Vargova, Marsanova, Vitek, Smital, Filipenska, Sikorova, Galik; PhysioNet 2024, version 2.0.0, url physionet.org/content/butppg/2.0.0/ - CONFIRMED | the associated paper in BioMed Research International 2021, article 3453007 | https://api.crossref.org/works/10.1155/2021/3453007 | [cite-brno-paper] BioMed Research International, article-number 3453007, issued 2021 - CONFIRMED | are redistributed under Creative Commons Attribution 4.0 International | https://physionet.org/content/butppg/view-license/2.0.0/ | [licence-ccby] "Creative Commons Attribution 4.0 International"; also the licence field of the shipped butppg-sample.json - CONFIRMED | the licence text ships beside them in ./specimens/LICENCE | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [licence-ships] section 12b "the licence file accounts for" all 15 shipped files; specimens/LICENCE served live (HTTP 200, same bytes as the repo) - CONFIRMED | Verkruysse W, Svaasand LO, Nelson JS, "Remote plethysmographic imaging using ambient light", Optics Express 2008;16(26):21434-21445. | https://api.crossref.org/works/10.1364/OE.16.021434 | [cite-verkruysse] Crossref: Verkruysse, Svaasand, Nelson, "Remote plethysmographic imaging using ambient light", Optics Express 16(26) p 21434, 2008; PMC record fpage 21434 lpage 21445 - CONFIRMED | Kamshilin AA, Nippolainen E, Sidorov IS, Vasilev PV, Erofeev NP, Podolian NP, Romashko RV, "A new look at the essence of the imaging photoplethysmography", Scientific Reports 2015;5:10494 | https://api.crossref.org/works/10.1038/srep10494 | [cite-kamshilin] Crossref: the seven authors as printed, "A new look at the essence of the imaging photoplethysmography", Scientific Reports 5, article 10494, 2015 - CONFIRMED | "the intensity of the back-reflected light is maximal in the end-diastole phase, which is well-established fact" | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=4440202 | [kam-quote] verbatim in the Discussion - CONFIRMED | this particular paper disputes the usual blood volume explanation of the effect and proposes a mechanical one | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=4440202 | [kam-disputes] "Here we report experimental observations, which are not consistent with this model and demonstrate the importance of elastic deformations of the capillary bed"; it keeps the direction ("maximal in the end-diastole phase") - CONFIRMED | WebKit commit 9c0b105d80212675839a00325b58df55f408f1c0, "[MediaStream] Add support for torch", dated 2023-10-04 | https://github.com/WebKit/WebKit/commit/9c0b105d80212675839a00325b58df55f408f1c0.patch | [webkit-commit] "Date: Wed, 4 Oct 2023 10:39:08 -0700 Subject: [PATCH] [MediaStream] Add support for torch" - CONFIRMED | "[MediaStream] Add support for torch" | https://github.com/WebKit/WebKit/commit/9c0b105d80212675839a00325b58df55f408f1c0.patch | [webkit-quote] commit subject verbatim - CONFIRMED | it adds torch to a video track's capabilities | https://github.com/WebKit/WebKit/commit/9c0b105d80212675839a00325b58df55f408f1c0.patch | [webkit-adds-torch] "Add support for `torch` to MediaTrackCapabilities and MediaTrackSettings" - CONFIRMED | its own layout test switches the light on with track.applyConstraints({torch: true}) after the stream is open | https://github.com/WebKit/WebKit/commit/9c0b105d80212675839a00325b58df55f408f1c0.patch | [webkit-test-apply] the test asserts getSettings().torch true right after "applyConstraints({ torch : true })" - CONFIRMED | Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology, "Heart rate variability: standards of measurement, physiological interpretation, and clinical use", Circulation 1996;93(5):1043-1065 | https://api.crossref.org/works/10.1161/01.CIR.93.5.1043 | [cite-taskforce] Crossref title "Heart Rate Variability", subtitle "Standards of Measurement, Physiological Interpretation, and Clinical Use", Circulation 93(5) 1043-1065, 1996; author "Task Force of the European Society of Cardiology the North A[merican Society of Pacing and] Electrophysiology". N1 said Crossref omits the subtitle; it carries it in the subtitle field - CONFIRMED | Shaffer F, Ginsberg JP, "An Overview of Heart Rate Variability Metrics and Norms", Frontiers in Public Health 2017;5:258 | https://api.crossref.org/works/10.3389/fpubh.2017.00258 | [cite-shaffer] Crossref: Shaffer, Ginsberg, "An Overview of Heart Rate Variability Metrics and Norms", Frontiers in Public Health 5, article 258, 2017 - CONFIRMED | "HRV measure, Mean (SD), Range, Studies" | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [shaffer-table6-columns] Table 6 header cells "HRV measure", "Mean (SD)", "Range", "Studies": the page joins the cells with commas inside its quotation marks; no word changed. The offline tool flags it as not word for word because of those commas - CONFIRMED | "42 (15), 19-75, 15" | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [shaffer-table6-row] row cells "42 (15)", "19-75" (en dash), "15"; same comma joining, hyphen for the en dash - CONFIRMED | Choi A, Shin H, "Photoplethysmography sampling frequency: pilot assessment of how low can we go to analyze pulse rate variability with reliability?", Physiological Measurement 2017;38(3):586-600 | https://api.crossref.org/works/10.1088/1361-6579/aa5efa | [cite-choi] Crossref: Choi A, Shin H, title as printed, Physiological Measurement 38(3) 586-600, 2017 - CONFIRMED | Schäfer A, Vagedes J, "How accurate is pulse rate variability as an estimate of heart rate variability? A review on studies comparing photoplethysmographic technology with an electrocardiogram", International Journal of Cardiology 2013;166(1):15-29 | https://api.crossref.org/works/10.1016/j.ijcard.2012.03.119 | [cite-schafer] Crossref: Axel Schafer, Jan Vagedes, International Journal of Cardiology 166(1) 15-29, 2013; full title from Europe PMC - CONFIRMED | Mather JD, Hayes LD, Mair JL, Sculthorpe NF, "Validity of resting heart rate derived from contact-based smartphone photoplethysmography compared with electrocardiography: a scoping review and checklist for optimal acquisition and reporting", Frontiers in Digital Health 2024;6:1326511 | https://api.crossref.org/works/10.3389/fdgth.2024.1326511 | [cite-mather] Crossref: Mather, Hayes, Mair, Sculthorpe, title as printed, Frontiers in Digital Health 6, article 1326511, 2024 - OBSERVED | read at source 2026-08-17 via PMC4440202 | research/ppg-pulse/README.md | [obs-kam-read] Sources table, line 38, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17") - OBSERVED | read at source 2026-08-17: it adds torch to a video track's capabilities | research/ppg-pulse/README.md | [obs-webkit-read] Sources table, line 40, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17") - OBSERVED | Read at source 2026-08-17. Its columns are | research/ppg-pulse/README.md | [obs-shaffer-read] Sources table, line 41, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17") - OBSERVED | whose abstract was read at source on 2026-08-17 through Europe PMC's API | research/ppg-pulse/README.md | [obs-choi-read] Sources table, line 44, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17"); the same row says "The full text is paywalled", which is false (entry choi-paywall) - OBSERVED | abstract likewise read at source 2026-08-17 through Europe PMC | research/ppg-pulse/README.md | [obs-schafer-read] Sources table, line 45, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17") - OBSERVED | read at source 2026-08-17, where the widest limits tabulated are Nam et al. | research/ppg-pulse/README.md | [obs-mather-read] Sources table, line 43, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17"); I confirm Nam is the widest of the ten LoA in Table 8 - OBSERVED | verified at source on 2026-08-17 | research/ppg-pulse/README.md | [obs-licence-verified] Sources table, line 35, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17") - OBSERVED | read at source on 2026-08-17. The beat marks | research/ppg-pulse/README.md | [obs-quality-def-read] Sources table, line 39, records the read on 2026-08-17 ("read at source" or "fetched 2026-08-17") - CONFIRMED | which is why the technique has a surveillance literature as well as a clinical one | https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=(%22remote%20photoplethysmography%22%20OR%20rPPG)%20AND%20(surveillance%20OR%20%22lie%20detection%22%20OR%20deception%20OR%20%22privacy%22)&resultType=core&format=json&pageSize=25 | [rppg-surveillance] Elgendi et al., "Balancing cardiac privacy with quality in video recordings", Communications Medicine 2025 (doi 10.1038/s43856-025-01175-0): rPPG "exposes individuals to potential misuse, as sensitive information can be inferred without consent"; several 2025 to 2026 papers on facial privacy protection against rPPG. A misuse and privacy literature exists; "surveillance" is the page's word for it - CONFIRMED | the frames never leave the tab, and nothing is stored | public/strata/ppg-pulse/instrument.js | [frames-stay] captureFrames reduces each frame with P.reduceFrame to channel means; verifier section 12 "no absolute URL to another origin in the page's inline script content", "nor in its sibling modules" - CONFIRMED | Which way does it dip? Record 133007, further down this page, ships with an electrocardiogram recorded at the same time, so that is checkable here rather than assertable. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [built-dip-311] section 10 "the red channel bottoms out a median of 311 ms after the R wave", "less than half a beat (552 ms)" - CONFIRMED | The database's own description says its PPG signal was inverted. | https://physionet.org/content/butppg/2.0.0/; https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=8440059 | [built-inverted] "Finally, the PPG signal was inverted." The BMRI 2021 paper describes that step for the first release ("The red video channel was selected ... The calculated PPG signal was inverted"); headers fetched 2026-10-19: 100001_PPG.hea and 111004_PPG.hea gains -128.9 and -247.8 (negative), 112001 to 133007 positive with PPG_R/G/B, as the page says - CONFIRMED | This is a real person, measured at 1000 Hz by an instrument that is not a camera. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | [built-ecg-17] section 10: 17.48 ms over eight intervals; "the dataset's own marks give this person 16.90 ms"; over 2000 phases 0.00 to 47.1, median 30.9, high on 89 per cent - MINOR | those are ordinary healthy people measured for ten seconds. | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive); https://physionet.org/content/butppg/2.0.0/ | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Unsupported adjective: BUT PPG describes 50 subjects aged 19 to 76 and does not call them healthy; the figures beside it are confirmed. Loose wording, listed. (checker: [built-short-window] Numbers confirmed (section 8b: 64 usable finger recordings, median 30.3 ms, 28 and 41 per cent). But "Those are ordinary healthy people measured for ten seconds" (instrument.js line 858) is unsupported: the dataset describes "50 subjects (25 female, 25 male) aged between 19 to 76 years" and the BMRI paper "volunteers"; neither says healthy (positive control: the same searches find "health monitoring" in both), and the dataset carries blood pressure, glycaemia and SpO2 columns. The comparison is to norms for healthy adults, so the word matters. Suggest "ordinary volunteers") - CONFIRMED | Your own count has a grid too: a whole number of beats in 30 seconds can only land on even multiples | public/strata/ppg-pulse/instrument.js | [wrist-grid] arithmetic: n beats in 30 s is 2n bpm, so steps of 2 bpm; verifier section 12b - MINOR | which is the single // commonest bug in code that reads a pulse off a camera | public/strata/ppg-pulse/ppg.js | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. A code comment in ppg.js repeats an unmeasured generalisation the page itself retracted; not reader-facing prose on the page. (checker: [commonest-bug] ppg.js lines 288 to 290, shipped and linked from the page ("The estimator is /strata/ppg-pulse/ppg.js"): "it reports twice the rate whenever the dicrotic shoulder makes the second harmonic the taller line, which is the single commonest bug in code that reads a pulse off a camera." The layer retracted both halves from the page as unmeasured (README; verifier section 8 "NEGATIVE: and so is the unfalsifiable claim about what the commonest bug in other people's code is"), and its own count finds the second harmonic taller on 20 of 432 channels and on no accepted record, so "wheneve) - CONFIRMED | (added) Brno records "reduced by its authors to three per frame colour channel means at 30 Hz" | https://physionet.org/content/butppg/2.0.0/; https://physionet.org/files/butppg/2.0.0/133007/133007_PPG.hea | Added: description "the PPG signal was created using averaging of each frame", 30 Hz; the v2 headers carry three channels PPG_R, PPG_G, PPG_B ("133007_PPG 3 30 300"), and they decode to frame means of 0 to 255 (verifier section 3) - CONFIRMED | (added) "The beat marks in ./specimens/butppg-qrs.bin are that database's own per-record .qrs annotation files, unaltered and simply concatenated" | https://physionet.org/files/butppg/2.0.0/133007/133007.qrs | Added: I cut seven records out of the shipped bin by the offsets in butppg-qrs.json (130078, 134082, 130005, 131053, 140088, 140067, 133007) and each is byte-identical to PhysioNet's .qrs; two different records differ (control). The shipped PPG and ECG .dat files and both CSVs are byte-identical to PhysioNet too - MINOR | (added) "from all 2094 Brno records above ID 130000, which is the range that uses the three-channel layout" (instrument.js line 1262, built at load) | https://physionet.org/files/butppg/2.0.0/112001/112001_PPG.hea | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Loose range wording: headers fetched by the checker show the three-channel layout from 112001; the pool above 130000 is three-channel throughout but is a subset of that range, not the range. The page's "The 200 record sample covers only the three channel layout" stays true. (checker: Added: the three-channel layout begins at 112001, not 130000. Headers fetched 2026-10-19: 112001, 112002, 121076, 125001 and 129108 all read "3 30 300" with PPG_R, PPG_G, PPG_B; 111004 and 100001 are the older layout (positive control). Records 112001 to 129108 (1,746 of them) are three-channel and outside the pool, so the pool is a subset of that range, not the range. Same in README line 95 and fetch-butppg.mjs line 39 ("the v2 three-channel layout begins here"). The page's "The 200 record sample covers only the three channel layout" stays true) - MINOR | (added) "And ten seconds is not five minutes, which is what those studies recorded." (instrument.js line 853, about the Nunan norms) | https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pmc&id=5624990 | [declined] Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Unsupported precision: Shaffer and Ginsberg Table 5 lists Nunan et al. sample as "Varied"; 'short-term' is in Nunan's title but five minutes is not stated in the page's source. Loose wording, listed. (checker: Added: Shaffer and Ginsberg Table 5 lists Nunan et al. (17) with "Sample: Varied" and "Position: Varied" (positive control: the same table gives "5 min" for Abhishekh et al.). "Short-term" is in Nunan's title, but the page's source does not say those studies recorded five minutes. Suggest "short-term recordings, conventionally five minutes") - CONFIRMED | (added) "what a camera would have said about this person ... is a draw from a range that covers most of the healthy adult scale" (instrument.js line 1080) | research/ppg-pulse/verify-ppg-pulse.mjs (my run on a scratch copy at 9ddc1e9045, 2026-10-19: 425 passed, 1 failed, the failure being the og image missing from my partial archive) | Added: section 10 range 0.00 to 47.1 ms against the 19 to 75 ms span: overlap 19 to 47.1 is 28.1 of 56 ms, 50.2 per cent. Just most - CONFIRMED | (added) the HR column is "the column the dataset's own quality label is defined against" | https://physionet.org/content/butppg/2.0.0/ | Added: "The third column contains the reference HR" and quality 1 means most annotators were within 5 bpm of "the reference HR" ## Findings - [fixed] WRONG, contact-geometry-transmission (K4): public/strata/ppg-pulse/index.html Green, red, blue section said "the light goes through the finger rather than bouncing off it"; a phone's lamp and camera sit side by side under the finger, which the page's own source Kamshilin (PMC4440202) and Matsumura (PMC3949790) call reflection mode. Now: "the phone's lamp sits beside its camera under the fingertip, so what comes back is light that has wandered into the flesh and out again". Copies fixed: instrument.js ("as light through a finger must be" to "as light that has been through flesh must be"; dropped "in a geometry where the light goes straight through"), README and verifier ("transmission through a finger" to "light that has been through flesh"). Corrected line added on the page and in the markdown. - [fixed] WRONG, choi-paywall (K6): index.html Sources said Choi and Shin 2017's "full text sits behind a paywall"; Crossref gives a CC BY 3.0 VOR licence from 2017-02-23 and Europe PMC lists "Open access". Now: "full text is open access under a CC BY 3.0 licence, but this page read and quotes only the abstract"; README line fixed likewise. Corrected line added on the page and in the markdown. - [declined] MINOR, plain-hides-at-some-rates (K1): "a 30 frames a second grid usually invents about 23.6 ms of it, and at some rates hides the real thing". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. The checker's note holds on its face: the page shows the camera hiding real variability by frame phase (section 10), and at a whole-frame rate its panel inflates it; the plain card's 'at some rates' is loose wording, not a false fact about the world. - [declined] MINOR, four-need-more (K2): "Each of those is something readers reasonably expect from a fingertip and each of them needs either a second wavelength, a calibration again". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Loose grouping: three of the four expectations plausibly need a second wavelength, a cuff calibration or a rhythm standard; the fourth (a stress or recovery score) is refused for the frame-grid reason the page gives elsewhere. Not false about any source. - [declined] MINOR, pulse-tenths (K3): "a real fingertip pulse moves the light by tenths of a per cent and anything larger is your finger moving". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Unscoped generalisation: Lima et al. 2002 report fingertip pulse-oximeter perfusion index 0.3 to 10.0 per cent (a different measure, peak-based, on a different sensor); for this camera instrument the page's own accepted means are under one per cent, so the gate stands and the wording is loose. - [declined] MINOR, comb-round-numbers (K5): "the exceptions are not rare curiosities: they are the round numbers". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Overstatement: only round numbers that divide 1800 at 30 fps read zero (the checker's recreation gives 70 and 80 bpm at 25.18 and 33.33 ms). The sentence's point, that exceptions are common rates and not curiosities, stands; the wording is loose. - [declined] MINOR, capture-20s (K7): "a twenty second fingertip capture gives neither". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Loose wording: a twenty second capture never gives 60 seconds and gives 30 beats only at 90 bpm or more, so 'neither' holds at resting rates; the page's refusal stands either way. - [declined] MINOR, check-stored (K8): "Nothing on this page is a stored result". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Overclaim about the page's own checking (the brief's MINOR class): a few prose figures (57.8, 0.267, 0.737, 15.7) are typed, though true by the offline verifier. - [declined] MINOR, check-spectral-bin (K9): "A ten second record has a raw spectral bin of 6.0 bpm". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Attribution slip inside the page's own method note: ppg.js zero-pads to CFG.fftPoints before bandPeak's parabola, so 'only because of parabolic interpolation' names one of two interpolations; the caution (no added resolution) stands. - [declined] MINOR, verifier-claims (K10): "reruns every number above from the shipped files, regenerates the synthesised specimens and compares their hashes". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Overstatement of the verifier's reach: the checker's mutation shows one figure (57.8 in one sentence) is pinned by a token another sentence also satisfies; every other rerun held. - [declined] MINOR, built-short-window-healthy (K11): "those are ordinary healthy people measured for ten seconds.". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Unsupported adjective: BUT PPG describes 50 subjects aged 19 to 76 and does not call them healthy; the figures beside it are confirmed. Loose wording, listed. - [declined] MINOR, commonest-bug-comment (K12): "which is the single // commonest bug in code that reads a pulse off a camera". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. A code comment in ppg.js repeats an unmeasured generalisation the page itself retracted; not reader-facing prose on the page. - [declined] MINOR, three-channel-range (K13): "(added) "from all 2094 Brno records above ID 130000, which is the range that uses the three-channel layout" (instrument.js line 1262, built ". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Loose range wording: headers fetched by the checker show the three-channel layout from 112001; the pool above 130000 is three-channel throughout but is a subset of that range, not the range. The page's "The 200 record sample covers only the three channel layout" stays true. - [declined] MINOR, nunan-five-minutes (K14): "(added) "And ten seconds is not five minutes, which is what those studies recorded." (instrument.js line 853, about the Nunan norms)". Tier B MINOR not in a sentence being fixed for a WRONG item: listed, page left as is. Unsupported precision: Shaffer and Ginsberg Table 5 lists Nunan et al. sample as "Varied"; 'short-term' is in Nunan's title but five minutes is not stated in the page's source. Loose wording, listed.