Operate a faithful fitted model of Google Quantum AI's Willow surface-code
memory result: move measured detection probability through the reported crossover and
watch larger codes change from worse to better, then inspect the real-time decoder queue
and the rare correlated bursts that stop a simple exponential story.
Adding qubits is not automatically protection. On one side of a threshold,
a larger code creates more places to fail. On the other, the extra checks suppress logical
failure. Move the noise proxy. The ordering must reverse.
Layer one: threshold ordering
Computing
Computing the ordering.
Computing the fitted relation.
distance 3distance 5distance 7
The live text alternative will describe the plotted ordering.
Apparatus boundary. The analytic
view is a faithful fitted model built from the paper's rounded relation
1/Lambda = 3.4 p_det + 0.29. It is not a Willow simulator and
it does not run a decoder. The published-fit replay reconstructs three points on the
reported regression line from the published distance-7 value and suppression factor.
It is not a fresh reduction of the 5.7 GB Zenodo archive.
The first result is only three low distances, decoded offline.
So operate the part that answers a harder dismissal: deliberate extra noise made the three
distance curves cross, and a separate streaming decoder kept pace for a million cycles.
Layer twoThe queue that must not grow
A final decoder latency can be longer than one correction cycle while the system still
works in real time. The condition is throughput: streamed work must leave at least as fast
as it arrives. This schematic replay uses ten-cycle packets, following the paper's block
description. Its service time is an openly chosen visual convention, not a measured Willow
latency trace.
Streaming queue replay
Ready
Choose a run length, then run the stream.
Computinghardware run duration
Computingpublished final latency
Computingschematic peak queue
Honesty trapdoor: one error channel
Computing
Computing the ideal extrapolation.
The teal number is an ideal independent-error extrapolation
normalized by the page to meet the published apparent floor at distance 15. That
normalization is a free visual choice. The floor, six bursts, run length, and fitted
suppression factor are published results. A repetition code protects one error
channel, not the full surface-code memory.
The live text alternative will report the burst count and rate.
The check: rebuilt in this browser
Every row below starts empty and is filled by the same runtime engine that drives the
controls. Published constants are labeled as published. Derived values are recomputed.
Published fit replay
Computing
Error conversion
Computing
Distance-7 qubits
Computing
Rounded crossover
Computing
Lifetime ratio
Computing
Million-cycle time
Computing
Rare-burst rate
Computing
Uncertainties, approximations, conventions, and free choices
Decoder choice matters. The displayed published replay uses the
offline neural-network result. The same experiment gave a different suppression
factor with ensembled matching synthesis. The real-time result used sparse blossom
on a separate 72-qubit processor.
Only three hardware distances enter the headline fit. Distance-3
averages nine subgrids, distance-5 averages four, and distance-7 is one code. Surface
codes through distance 11 in the paper are simulations, not additional hardware runs.
Detection probability is a proxy. It is the rate of disagreeing
weight-4 stabilizer comparisons, not a universal physical gate-error probability.
The fitted coefficients are rounded, so the derived crossing is approximate.
The analytic curve is an anchored stand-in, not a decoder simulation.
It is pinned to the distance-5 point on the line fixed by the published distance-7 error
and suppression factor, 0.306 percent at a detection probability of 8.5 percent (the published distance-7 error times the published
suppression factor; the paper's measured distance-5 value with the same decoder is
0.303 percent), carries its shared noise dependence as the ideal power law of the
paper's equation (1) at distance 5, the cube of the detection probability, and takes
its distance dependence from the fitted suppression factor. Every curve therefore rises
with noise, which is the physical behaviour. It is plotted from 0 to 30 percent, wider
than the paper's fit: the paper fitted its suppression-factor line only to points below
the crossing, made by injecting noise on top of the device's own detection probabilities
of 7.7 to 8.7 percent, so below about 7.7 percent, and above the rounded crossing, about
20.9 percent, the side where a larger code hurts, the curves extend the fit past the
points it was made from.
The published fit replay is reconstructed. Its distance-3 and
distance-5 points are generated on the line fixed by the published distance-7 error
and published suppression factor. The raw 5.7 GB archive was not downloaded or
independently decoded here.
Lifetime comparison uses a defined average. Physical and logical
noise differ. The paper compares uniform-state-average lifetimes, not a simple
substitution for a physical qubit's T1.
The queue is schematic. The chosen 9.5 microsecond service time
per ten-cycle visual packet is not reported hardware data. It only demonstrates how
bounded streaming can coexist with a longer final-result latency.
The repetition-code curve is not surface-code performance. Its
anchor at the apparent floor is a declared visual normalization. Correlated noise
violates the independent-error extrapolation.
Units and indexing. Probabilities are fractions internally and
percentages at the controls. Code distances are odd integers. Times are converted
from microseconds to seconds with one million microseconds per second.
Landing date. The supplied spec contained no landing date, so this
entry uses the build date, 29 July 2026.
Still open
This experiment did not demonstrate logical gates, state injection, multiple logical
qubits, a useful algorithm, implemented feedback, or a complete fault-tolerant computer.
Surface-code scaling beyond distance 7 was not measured on this hardware. Real-time
decoding was demonstrated at distance 5, not distance 7. The cause of the rare
repetition-code bursts was unknown, and their effect on larger surface codes was not
established. Whether this hardware and classical stack scale economically to long
computations remains unresolved.
Primary sources
Google Quantum AI and Collaborators,
"Quantum error correction below
the surface code threshold,"Nature 638, 920-926 (2025),
DOI 10.1038/s41586-024-08449-y. Peer-reviewed article, published online
9 December 2024, version of record 29 January 2025, issue date 27 February 2025.
arXiv:2408.13687, the preprint and
author manuscript. It independently exposes the headline summary values but is not
itself peer reviewed.
Google Quantum AI and Collaborators,
Author Correction,
Nature 653, E5 (2026), DOI 10.1038/s41586-026-10559-8, published
28 April 2026. This is a journal correction whose peer-review status is not
separately stated on its record. It corrects labels in Figure 3a. The load-bearing
values used here were not changed.
Eric Dennis, Alexei Kitaev, Andrew Landahl, and John Preskill,
"Topological quantum memory,"Journal of Mathematical Physics 43, 4452-4505 (2002),
DOI 10.1063/1.1499754. Peer-reviewed theoretical source for the surface-code
threshold framework.
Corrected 2026-10-18: This page called the analytic curve's anchor, 0.306 percent at a
detection probability of 8.5 percent, the paper's published distance-5 point. It is the
distance-5 point on the line fixed by the published distance-7 error and suppression factor
(1.43 x 10^-3 x 2.14); the paper's measured distance-5 error with the same neural-network
decoder is 0.303 percent per cycle, in Table S1 of arXiv:2408.13687 (found by the assay
line, 2026-10-18).
Corrected 2026-10-18: The model view printed its formula as
epsilon_d = 0.01 x (1/Lambda)^((d-5)/2), which is not what its
engine computes; the engine computes
epsilon_d = 0.00306 x (p_det/0.085)^3 x (1/Lambda)^((d-5)/2),
and the printed formula now says so (found by the assay line, 2026-10-18).
Corrected 2026-10-18: This page said the analytic curve is plotted only over the range
the paper fits. It is plotted from 0 to 30 percent; the paper fitted its suppression-factor
line only to points below the crossing (about 20.9 percent on the rounded fit), made by
injecting noise on top of the device's own detection probabilities of 7.7 to 8.7 percent,
so the curves below about 7.7 percent and above the crossing extend the fit (found by the
assay line, 2026-10-18).
Corrected 2026-10-18: This page's note on The Crossover at Twenty-Two Degrees said that
layer swaps which running pace feels harder; it swaps which leaf, an ordinary C3 leaf or
a maize leaf, turns light into sugar more efficiently (found by the assay line,
2026-10-18).
The claims recordNaN claims re-read against their sources, 18 October 2026
Written 2026-07-29. Claims re-read against their sources on 2026-10-18: NaN checked, 73 confirmed, NaN wrong, 0 unverifiable, 4 first-hand observations checked against their record. 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).
This pass was the assay line's source-first re-read of this layer (batch claude-1018b, tier B). A Claude agent (claude-sonnet-5-5) wrote a source record for the page's claims (research/assay-line/claude-1018b/the-error-that-shrinks/N1-sources.json); a second Claude agent (claude-opus-5-5) read each entry against its source (the Nature article's Europe PMC full text, the arXiv HTML with its supplement, Crossref, the Zenodo API, the nature.com page) and against the page's own code, and wrote the claims below (N2-claude.md). A fixer (claude-opus-5-5) re-read every WRONG item at its source or in the page's code before deciding it (triage.json), then fixed four WRONG statements and one MINOR one that shared a sentence with a WRONG one, added four dated Corrected lines to a new Corrections note on the page and at the end of the markdown body, and listed two MINOR items without editing. The layer's verifier gained one check, that the formula the model view prints reproduces its engine, and goes red on the old printed formula; it passes 14 of 14.
CONFIRMED [curves-cross] deliberate extra noise made the three distance curves cross https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "we inject coherent errors with variable strengths" and "the three curves cross near a detection probability of 20%" (the plot is Fig. 2b; the paper's own text says Fig. 2a for the injection)
CONFIRMED [streaming-million] a separate streaming decoder kept pace for a million cycles https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "up to 10^6 ... latency remains roughly constant ... indicating that the decoding problem is being processed in real time"; 72-qubit processor, not the 105-qubit one
CONFIRMED [throughput-condition] The condition is throughput: streamed work must leave at least as fast as it arrives. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "The throughput of transmitting, processing and decoding ... must keep pace with the fast error-correcting cycle time of 1.1 us"; backlog sentence matches the excerpt
CONFIRMED [ten-cycle-blocks] This schematic replay uses ten-cycle packets, following the paper's block description. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : Fig. 4 (Real-time decoding) caption: "The decoding problems are subdivided into blocks ... We use ten-cycle blocks"
CONFIRMED [rounded-relation] faithful fitted model built from the paper's rounded relation 1/Lambda = 3.4 p_det + 0.29 https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : Fig. 2 caption: "Line, fit to points at which 1/Λ < 1, 3.4pdet + 0.29" (the relation; "faithful" is weighed at d-faithful)
CONFIRMED [sparse-blossom-72] The real-time result used sparse blossom on a separate 72-qubit processor. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "Using a 72-qubit processor, we implement a distance-5 surface code ... real-time decoder"; "We use the sparse blossom algorithm"
CONFIRMED [subgrids] Distance-3 averages nine subgrids, distance-5 averages four, and distance-7 is one code. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "nine different distance-3 subgrids and four different distance-5 subgrids to compare with the distance-7 code"
WRONG [d5-anchor-0306] pinned to the paper's published distance-5 point, 0.306 percent at a detection probability of 8.5 percent https://arxiv.org/html/2408.13687 : Table S1, Fig. 1d columns: ε5 = (3.49±0.04)×10−3 (Libra) and (3.03±0.03)×10−3 (neural network); 0.306% is ε7 × Λ = 1.43e-3 × 2.14, a point on the fitted line, published nowhere (search for "3.06" and "0.306": 0 hits in either text; positive control "3.03": 2 hits in the arXiv text)
MINOR [power-law] carries its shared noise dependence as the paper's own power law in the detection probability https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : the shared factor is p^3 for every distance (index.html line 58, comment citing Fig. 3c); the paper's fitted form εd = Ad pdet^((d+1)/2) is in the Fig. 3c caption, the repetition-code figure, and for the surface code the paper says "When fitting power laws below the crossing, we observe approximately 80% of the ideal value (d + 1)/2", so p^3 is the ideal exponent of equation (1) at d = 5, not the paper's surface-code fit
CONFIRMED [service-9-5] The chosen 9.5 microsecond service time per ten-cycle visual packet is not reported hardware data. public/strata/the-error-that-shrinks/index.html : streamingStats(cycles, 9.5) is the page's constant; my own search of the Nature text and the arXiv HTML for "9.5": 0 hits, positive control "63": 4 and 6 hits
CONFIRMED [correlated-floor] Correlated noise violates the independent-error extrapolation. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "deviation from exponential error suppression at high distances culminating in an apparent logical error floor of 10−10"; Outlook: "a noise floor at an error rate of 10−10 caused by correlated bursts of errors"
CONFIRMED [repetition-not-surface] A repetition code protects one error channel, not the full surface-code memory. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "surface codes are larger and sensitive to more errors than repetition codes"; shots "split evenly between bit- and phase-flip codes"
CONFIRMED [floor-bursts-run] The floor, six bursts, run length, and fitted suppression factor are published results. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : 2 × 10^10 cycles, 5.5 h, Λ = 8.4 ± 0.1 (d = 5 to 11), floor 10−10, "six of these large error bursts"
CONFIRMED [not-demonstrated] This experiment did not demonstrate logical gates, state injection, multiple logical qubits, a useful algorithm, implemented feedback, or a complete fault-tolerant computer. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : memory experiments only; "feedback (which we have not yet implemented)"; "this work focuses on building a robust memory"
CONFIRMED [cause-unknown] The cause of the rare repetition-code bursts was unknown, and their effect on larger surface codes was not established. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "We do not yet understand the cause of these events"; "may affect the surface code performance differently"
CONFIRMED [scale-economics] Whether this hardware and classical stack scale economically to long computations remains unresolved. https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : "it would be resource intensive in practice"; "which we must now scale in practice"
CONFIRMED [cite-nature] Google Quantum AI and Collaborators, "Quantum error correction below the surface code threshold," Nature 638, 920-926 (2025), DOI 10.1038/s41586-024-08449-y. https://api.crossref.org/works/10.1038/s41586-024-08449-y : volume 638, issue 8052, pages 920-926, print 2025-02-27; first Crossref author is "Google Quantum AI and Collaborators", so the tool's DOI MISMATCH is the tool's
CONFIRMED [nature-dates] published online 9 December 2024, version of record 29 January 2025, issue date 27 February 2025. https://www.nature.com/articles/s41586-024-08449-y : read with curl and a browser user agent (200, 451,739 bytes): "Published: 09 December 2024", "Version of record: 29 January 2025", "Issue date: 27 February 2025"
CONFIRMED [arxiv-id] arXiv:2408.13687, the preprint and author manuscript. https://arxiv.org/abs/2408.13687 : one version only, v1 of 24 Aug 2024 (the day Nature received it); journal ref "Nature 638 (2025) 920-926"
CONFIRMED [arxiv-headline] It independently exposes the headline summary values but is not itself peer reviewed. https://arxiv.org/abs/2408.13687 : abstract gives Λ = 2.14 ± 0.02 and 0.143% ± 0.003%
CONFIRMED [zenodo-record] Zenodo record 13273331, version 1.0.0, published 26 August 2024. https://zenodo.org/api/records/13273331 : version "1.0.0", publication_date "2024-08-26"
CONFIRMED [zenodo-title] Data for "Quantum error correction below the surface code threshold," https://zenodo.org/api/records/13273331 : API title: Data for "Quantum error correction below the surface code threshold" (the N1 excerpt's "Open Data for" runs the page's access badge into the title)
MINOR [zenodo-5-7gb] It is not a fresh reduction of the 5.7 GB Zenodo archive. / The raw 5.7 GB archive was not downloaded https://zenodo.org/api/records/13273331 : the record is four zips, 112.5 GB in all, three of them surface-code data: google_105Q_surface_code_d3_d5_d7.zip 5.72 GB, google_72Q_surface_code_d3_d5_set1.zip 30.15 GB, google_72Q_surface_code_d3_d5_set2.zip 11.62 GB; the 5.7 GB file is the right one for the headline but the page never names it
CONFIRMED [cite-correction] Nature 653, E5 (2026), DOI 10.1038/s41586-026-10559-8, published 28 April 2026. https://api.crossref.org/works/10.1038/s41586-026-10559-8 : volume 653, page E5, published-online 2026-04-28; DOI MISMATCH again the group-name first author
CONFIRMED [cite-dklp] Eric Dennis, Alexei Kitaev, Andrew Landahl, and John Preskill, "Topological quantum memory," Journal of Mathematical Physics 43, 4452-4505 (2002), DOI 10.1063/1.1499754. https://api.crossref.org/works/10.1063/1.1499754 : four authors, volume 43, issue 9, pages 4452-4505, 2002
CONFIRMED [dklp-theory] Peer-reviewed theoretical source for the surface-code threshold framework. https://arxiv.org/abs/quant-ph/0110143 : "accuracy threshold ... encoded information can be protected arbitrarily well"
CONFIRMED [d-not-simulator] It is not a Willow simulator and it does not run a decoder. public/strata/the-error-that-shrinks/index.html : modelError is a closed-form expression; no decoder in the engine
CONFIRMED [d-replay-code] The published-fit replay reconstructs three points on the reported regression line from the published distance-7 value and suppression factor. public/strata/the-error-that-shrinks/index.html : error = epsilon7 × lambdaNN^((7 - d)/2)
WRONG [d-formula-shown] epsilon_d = 0.01 x (1/Lambda)^((d-5)/2) public/strata/the-error-that-shrinks/index.html : line 878 prints the formula of the earlier engine (git show 44a8b326b6f: crossError = 0.01), which the page's own comment says "left distance 5 pinned at one percent"; the engine since 983c9d526ac computes 0.00306 × (p/0.085)^3 × (1/Λ)^((d-5)/2). On load (p_det 8.7%, distance 7) the status line says 0.192% while the printed formula gives 0.586%
CONFIRMED [d-rises] Every curve therefore rises with noise public/strata/the-error-that-shrinks/index.html : recomputed from the shipped engine; the verifier sweeps it
MINOR [d-faithful] Operate a faithful fitted model public/strata/the-error-that-shrinks/index.html : at the paper's own operating points the model gives d3 0.412% at 7.7% (published NN 0.650%), d7 0.192% at 8.7% (published 0.143%), and Λ = 1.727 at 8.5% against the published 2.14; the 1/Λ fit came from injected-noise runs decoded with correlated matching, and the page does not say so
WRONG [d-range] it is plotted only over the range the paper fits https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : Fig. 2 caption: "Lines, power-law fits for data points at or below at which the codes cross ... Line, fit to points at which 1/Λ < 1, 3.4pdet + 0.29"; the chart and slider run 0 to 30% (index.html lines 334, 826); the surface-code fits stop near 20.9%, injection only raises pdet from about 7.7 to 8.7%, and the 30% bound is the repetition-code window of Fig. 3c (pdet < 0.3)
CONFIRMED [d-service] The chosen 9.5 microsecond service time public/strata/the-error-that-shrinks/index.html : E.streamingStats(Number(cycles.value), 9.5)
CONFIRMED [d-seconds] Times are converted from microseconds to seconds with one million microseconds per second. public/strata/the-error-that-shrinks/index.html : cycles × 1.1 / 1e6
CONFIRMED [d-qubits] 2 x 7^2 - 1 = 97; add 4 published leakage-removal qubits = 101 public/strata/the-error-that-shrinks/index.html : 2d^2 - 1 is the paper's own count (equation (1) text)
CONFIRMED [d-same-engine] Every row below starts empty and is filled by the same runtime engine that drives the controls. public/strata/the-error-that-shrinks/index.html : rows hold the placeholder "Computing" until fillChecks writes them from window.QECBench, the engine the controls use
CONFIRMED [d-decorative] Bar heights are decorative. public/strata/the-error-that-shrinks/index.html : (34 + ((i × 37) mod 62))%
CONFIRMED [d-offline-check] node research/the-error-that-shrinks/verify-the-error-that-shrinks.mjs research/the-error-that-shrinks/verify-the-error-that-shrinks.mjs : repository copy 13 PASS, exit 0; from the served files 12 PASS 1 FAIL, exit 1 (see Findings)
CONFIRMED [d-lifts-source] This layer's check lifts this page's own source out of the HTML and runs it. research/the-error-that-shrinks/verify-the-error-that-shrinks.mjs : extractEngine reads the qec-engine script from the page and runs it in a vm
CONFIRMED [d-two-files] it runs from 2 files this site serves public/checks/manifest.json : needs the verifier and index.html; both served 200 (9,619 and 51,831 bytes)
OBSERVED [d-landing] The supplied spec contained no landing date, so this entry uses the build date, 29 July 2026. git log --diff-filter=A 4a6041fc27 -- src/content/strata/the-error-that-shrinks.md : first commit 44a8b326b6f 2026-07-29 10:31 +1000, matching the frontmatter date; the spec itself is not kept (memory/log.d/2026-07-29-settled-since-you-last-looked-wave.md lists the layer only)
OBSERVED [d-last-verified] Last verified 29 July 2026, when it was written. git log 4a6041fc27 -- research/the-error-that-shrinks/ : the verifier last changed 0247067cf6c on 2026-07-29; recreated: it still passes 13/0 on the pinned repository copy
CONFIRMED [d-offline-sole] The analytic curve is an anchored stand-in, not a decoder simulation. public/strata/the-error-that-shrinks/index.html : var pAnchor = 0.085, epsAnchor = 0.00306
OBSERVED [o-emptied-red] Measured, not assumed: with every file this layer ships emptied, it goes red. research/verifier-independence/depends-1.json : records exitCode 1, depends true; recreated: empty index.html, node verify, exit 1
OBSERVED [o-old-form] which silently deleted the shared noise dependence and left distance 5 pinned at one percent git show 44a8b326b6f:public/strata/the-error-that-shrinks/index.html : modelError was crossError (0.01) × (1/Λ)^((d-5)/2): distance 5 is 1% at every p_det
CONFIRMED [r-hamming] A Message That Heals Itself shows a classical Hamming code locating one flipped bit exactly. src/content/strata/perfect-codes.md : "Flip a bit and watch Richard Hamming's 1950 code read three parity checks as the binary address of the wound"
CONFIRMED [r-perc] Percolation separates finite clusters from a spanning one src/content/strata/percolation.md : "small scattered islands; above it, a single cluster suddenly spans the whole plane"
WRONG [r-temp] The temperature layer swaps which running pace feels harder src/content/strata/the-crossover-at-twenty-two-degrees.md : that layer is about C3 against C4 (maize) photosynthesis: "Below some temperature an ordinary C3 leaf turns light into sugar more efficiently than a maize leaf does; above it, the maize wins"; no "pace" or "running" anywhere in it (positive control: the same grep finds "running pace" in the-error-that-shrinks.md); the full note is served on https://artwaste.land/ (2 hits)
CONFIRMED (added) Nearby layers, inbound note shown on this page: "below a critical error rate a surface code's errors shrink as you add qubits, and above it they..." https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11864966/fullTextXML : equation (1) and the threshold sentence of the abstract
What was done
[fixed] d5-anchor-is-fit-line-point: public/strata/the-error-that-shrinks/index.html, caveat "The analytic curve is an anchored stand-in", "It is pinned to the paper's published distance-5 point, 0.306 percent at a detection probability of 8.5 percent" now reads "It is pinned to the distance-5 point on the line fixed by the published distance-7 error and suppression factor, 0.306 percent at a detection probability of 8.5 percent (the published distance-7 error times the published suppression factor; the paper's measured distance-5 value with the same decoder is 0.303 percent)". Source: arXiv:2408.13687 HTML, Table S1, Fig. 1d neural-network column, epsilon_5 "(3.03 +/- 0.03) x 10^-3"; 1.43e-3 x 2.14 = 3.06e-3. The same phrase was corrected in the engine comment (modelError) and in the verifier's comment; the verifier now also asserts epsilon_7 x Lambda rounds to the anchor. Dated Corrected line added.
[fixed] printed-formula-is-retracted-engine: public/strata/the-error-that-shrinks/index.html, updateThreshold, the model view's printed formula "epsilon_d = 0.01 x (1/Lambda)^((d-5)/2)" now reads "epsilon_d = 0.00306 x (p_det/0.085)^3 x (1/Lambda)^((d-5)/2)", the form modelError computes (pAnchor = 0.085, epsAnchor = 0.00306). At the load state (8.7 percent, distance 7) the old line gave 0.586 percent against the status line's 0.192 percent. New verifier check "the formula the model view prints is the formula its engine computes" parses the printed coefficients and recomputes the engine from them at every slider value; red with the old string and with 0.00306 changed to 0.00406. Dated Corrected line added.
[fixed] plotted-range-exceeds-fit: public/strata/the-error-that-shrinks/index.html, same caveat, "it is plotted only over the range the paper fits" now reads "It is plotted from 0 to 30 percent, wider than the paper's fit: the paper fitted its suppression-factor line only to points below the crossing, made by injecting noise on top of the device's own detection probabilities of 7.7 to 8.7 percent, so below about 7.7 percent, and above the rounded crossing, about 20.9 percent, the side where a larger code hurts, the curves extend the fit past the points it was made from." Source: Nature Fig. 2 caption, "Line, fit to points at which 1/Λ < 1, 3.4pdet + 0.29" and "increasing the error injection angle α increases the detection probability"; Nature text, "The surface code detection probabilities are p det = (7.7%, 8.5%, 8.7%) for d = (3, 5, 7)"; the page's slider is 0 to 30 and drawThreshold sweeps p to 0.300. Dated Corrected line added; the repair stage added the low side (below about 7.7 percent), which the first fix left implied as fitted.
[fixed] relates-note-wrong-subject: src/content/strata/the-error-that-shrinks.md, relates note to the-crossover-at-twenty-two-degrees, "The temperature layer swaps which running pace feels harder" now reads "The temperature layer swaps which leaf, an ordinary C3 leaf or a maize leaf, turns light into sugar more efficiently", from that layer's dek ("Below some temperature an ordinary C3 leaf turns light into sugar more efficiently than a maize leaf does; above it, the maize wins"). The Nearby box on this page shows only the note's unchanged first sentence. Copies of the old note in public/strata/there-here/index.html and research/lineage-edge-grammar/page.json belong to another layer and a shared file and were not edited. Dated Corrected line added.
[fixed] power-law-is-ideal-exponent (MINOR, in the K1 sentence): "carries its shared noise dependence as the paper's own power law in the detection probability" now reads "carries its shared noise dependence as the ideal power law of the paper's equation (1) at distance 5, the cube of the detection probability". The fitted form epsilon_d = A_d p_det^((d+1)/2) is the Fig. 3c (repetition-code) caption; for the surface code the paper reports "approximately 80% of the ideal value (d + 1)/2". The engine comment citing Fig. 3c now says it is the repetition codes. No dated line (MINOR).
[declined] zenodo-archive-unnamed: "It is not a fresh reduction of the 5.7 GB Zenodo archive. / The raw 5.7 GB archive was not downloaded". MINOR, listed under tier B, not edited. The Zenodo record holds four archives (112.5 GB); the 5.7 GB one, google_105Q_surface_code_d3_d5_d7.zip, is the right archive for the headline result, and the page leaves it unnamed rather than saying anything false.
[declined] faithful-overstates-fit: "Operate a faithful fitted model" (dek, meta and og descriptions, JSON-LD, mode button, model note). MINOR, listed under tier B, not edited. At the paper's operating points the model gives Lambda 1.73 at 8.5 percent against the published 2.14 and 0.192 percent for distance 7 at 8.7 percent against 0.143 percent, and the relation comes from injected-noise runs; the page already calls the view a fitted model and an anchored stand-in, so "faithful" overstates rather than states a falsehood.
[declined] assay/sources.mjs DOI MISMATCH on cite-nature and cite-correction: Crossref lists "Google Quantum AI and Collaborators" as the first author, which is what the page prints, and its 2024 is the online date of a volume dated 2025.
The assay office: what a record is, and every layer re-read so far.