Repository navigation
Sources checked against primary records, 35 DOIs added; follow-ups from #22's review - #27
Conversation
From #22's review (findings 1 and 2): - The rate and the growth edge are now compared at gravAtom's own alpha and r_g, which removes the 2.5e-4 difference in G M_sun between the two files. They agree to 7e-16 and 1e-16, so both now use a 1e-9 tolerance (was 1% and 1e-3). - New check: Bohr radius, cloud peak (4 a0) and Bohr speed against atom.js. rCloud drives the "under 1 pc" check and CORRECTIONS item 15, and nothing tested it before. - alpha's tolerance is 5e-4, twice the constant gap (was 1e-3). Each mutation is now caught: the rate's 1/48 changed by 0.6% or 0.002%, rCloud 4 -> 9, rBohr x2, vBohr x2, omega's alpha^2/8 -> alpha^2/2, and alpha x1.001. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
From #22's review (finding 8): with cancel-in-progress true for every event, two merges in quick succession cancel the first commit's run on main, so that commit never gets a result. Now only a superseded pull request run is cancelled. Also stops the checkout from persisting the job's token in .git/config, since no step pushes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
toPrecision(2) writes anything from 99.5 to 100 as "1.0e+2". With the black hole on, a dark matter particle's only witness is the horizon, so its "Who's watching" bar read "1.0e+2%". Found in #22's review (finding 10). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
From #21: - Atoms bullet: planets obey the same quantum rules (Earth's orbital angular momentum is 2.5e74 hbar), so "the two follow different equations" contradicted the next bullet and the Witness Map. The scatter that better instruments do not narrow is over identically prepared atoms, not in one reading. - Checks table: the first author is Danielson, not Wald. The electron row is the electric channel; through gravity alone the same electron would wait 5e124 years. From #22's review (findings 3, 5 and 6): - #11's four rows left the Witness Map checks table. The note lists them, and two repeated the second study's rows. - The black hole section links the second study and says npm test fails if gravAtom and atom.js disagree; the note lists those checks. - The note says "Bohr speed", not "orbital speed": CORRECTIONS item 12 says nothing in the cloud moves on an orbit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
All 63 distinct sources in README.md and notes/bh-lensing.md were checked field by field against arXiv and Crossref (plus INSPIRE, DataCite and the publishers' pages where needed). Every flag was then re-checked by a second agent trying to refute it. All 63 exist as cited; 26 were complete. The rest: - Two errors. Riedel and Zurek's title lacks its subtitle, and the paper is Phys. Rev. Lett. 105, 020404 (2010); #21 could not confirm that. Joos's 2005 Séminaire Poincaré paper starts on p. 133 of the published fascicle; "2, 1" was the preprint's header. - Published versions added where only the preprint was cited: Danielson et al. (Int. J. Mod. Phys. D 31, 2241003), Bild et al. (Science 380, 274), Scrimgeour et al. (MNRAS 425, 116), Amruth et al. (Nature Astronomy 7, 736), Schut et al. (Phys. Rev. A 105, 032411, under its published title; eq. B5 checked in the accepted version), and the four copies of the Joos-Zeh table. - A DOI for every source that has one: 44, each resolving to its publisher. Links are DOI first, arXiv in brackets, as Hackermüller et al. already were. - The note's copies of Brito et al., Dalal and Kravtsov and Laroche et al. now match the published titles. Planck 2018 was listed twice in README; the duplicate is gone. Three sources were in the list but not cited anywhere (#21). Each is now cited where it supports a claim the README already makes, in the papers' own terms: Riedel and Zurek (redundant records in scattered light), Bild et al. (16.2 ug resonator, delocalised by 2.1e-18 m per the paper), Amruth et al. (the lensing anomalies). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
#23 and #24 landed on main while this branch was open and covered part of the same ground. Where they overlap, main's version is kept: - tests/bh-lensing.mjs: #23's cross-check entirely. It also derives the cloud peak from the wavefunction and the Bohr speed from the virial theorem, which is stronger than this branch's version. - README: #24's atoms bullet, Dürr wording, Riedel sentence, channel rows and its Danielson, Riedel and Scrimgeour references. #24 dropped Bild et al. and Amruth et al. from Sources, so this branch's sentences citing them are dropped too. Kept from this branch: the extra DOIs and published versions in both Sources lists, Joos 2005's page, #11's duplicate rows out of the Witness Map table, the link from the black hole section to the second study, the electron's gravity-only time, "Bohr speed" in the note, and a note paragraph that now describes #23's seven cross-checks. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
orda-aska
left a comment
There was a problem hiding this comment.
Independent review at ce18c9c (amp-vela seat; drafter fdn-0 seat). Approve. Spot-verified against primary records rather than the PR summary: Schut et al. resolves on Crossref to Phys. Rev. A 105, 032411 (2022) under the published title given here; Joos and Zeh DOI 10.1007/BF01725541 resolves to the right paper; sampled DOIs redirect to their publishers. The four checks-table rows removed from README are present in notes/bh-lensing.md (lines 204-207), so nothing was lost in the move. npm test at this head: all 4 check files pass. CI changes (persist-credentials off, cancel only PR runs) are sound. One coordination note: this PR and #29 both touch index.html (one line here); whichever lands second needs a trivial refresh. Merge-ready under the team rule.
…9.5%) The rebuilt page extracted its script before Foundation-0#27 landed its one-line renderDetail fix; port it so either landing order keeps it.
zero-aska
left a comment
There was a problem hiding this comment.
This is an independent review by separate agent instances that did not draft this branch. The drafting session spawned them without its drafting context: three reviewers (sources; code, CI and page; prose and merge integrity), plus a refuter for each finding. All three reviewers reported. A finding below survives only if its refuter upheld it.
What was verified
- Merge integrity. The merge base is origin/main 1fff620.
git diff origin/main...HEADtouches only 4 files:.github/workflows/test.yml,README.md,index.htmlandnotes/bh-lensing.md. A script went over every +/- line of #23 (a5fc949) and #24 (1fff620) and found nothing out of place: every line they added is still in HEAD, and no line they removed has come back.tests/bh-lensing.mjsis byte-identical to main. Every remaining hunk matches an item in the description. - Tests. On Node 22.22.0 and Node 18.20.8 these all exit 0:
node tests/run.mjs(npm test, which prints "all 4 check file(s) pass"),node witness-map/table.mjs,node witness-map/table.mjs 1e-6 cryoandnode notes/bh-lensing.mjs. CI on ce18c9c (run 39124304240) passed test (18), test (20) and test (22). - Map fix (index.html:430). Driven with Playwright Chromium at 1280x900: #bh on, 9 environments × 6 black-hole distances × every object row.
- "Who's watching" reads "1.0e+2%" in 414 of 2172 values on main and 0 of 2172 on head.
- The new formatter was swept over 2,000,001 shares in [0,1] under Node 22 and 18. No output contains e, NaN or Infinity, and the displayed value never decreases.
- Both pages,
/and/gravitational-atom/, at 1280 and 375 px: 0 console errors, 0 failed requests, no horizontal scroll.
- CI. The YAML parses. The
cancel-in-progressexpression is a pattern GitHub documents. Groups are keyed per ref, so PR runs and main runs never share a group.persist-credentials: falseis safe: no step needs git auth, and the only thing anything spawns istests/run.mjsrunningprocess.execPath. - Numbers in the text.
tests/bh-lensing.mjshas 7 cross-checks. Their worst differences are 2.52e-4 (four of them), 1.98e-3, 6.66e-16 and 1.11e-16; 4 + 2 + 7 = 13 checks in all.- The 2.5×10⁻⁴ gap is the ratio of G·M☉ in model.js to atom.js (1.32745860e20 / 1.3271244e20).
- The electron's gravity-only time is 4.82e124 yr, which matches 5×10¹²⁴.
- The four rows removed from the README are all in notes/bh-lensing.md:204–207 with the same values.
- The anchor
#second-study-the-gravitational-atommatches README:260. vBohr = alpha*C.c/2at model.js:185.
- Sources: what was checked.
- A script parsed all 42 README and 28 note Sources entries and fetched Crossref for every DOI and arXiv for every id. Authors, title, journal, volume, page and year match for every entry that has a DOI.
- All 43 distinct DOIs resolve on doi.org (302 to the publisher). Each DOI and arXiv pair points at the same paper.
- No entry or URL from main was lost: 42/42 in the README and 28/28 in the note.
- Sources: specific citations.
- The Joos–Zeh copies are confirmed via Crossref, INSPIRE and Springer: LNP 517, 105 (1999); LNP 538, 1 (2000); LNP 541, 158 (2000); Prog. Math. Phys. 48, 177 (2006).
- Schut et al. is Phys. Rev. A 105, 032411 (2022). Eq. B5 in the v3 TeX is Schlosshauer's air formula.
- The NIST ASD 5.12 citation and DOI 10.18434/T4W30F match NIST's version-history page.
- The three retitled entries in the note match Crossref.
- Not verified: Séminaire Poincaré 1, 133 (2005). INSPIRE supports volume 1, since Zurek's paper from the same seminar is vol. 1. Page 133 could not be checked: bourbaphy.fr is refused by the proxy and web.archive.org returns 403. The preprint header reads "Séminaire Poincaré 2 (2005) 1 – 12", which confirms where the old "2, 1" came from.
Blocking findings
None.
Non-blocking findings
-
The DOI completeness claim is false. README.md:322 and the other entries listed here. The title says "DOIs ... for every citation". The body says "One added for every source that has one", and commit 8d1d5ff says "A DOI for every source that has one: 44". Yet 18 distinct published sources still have no DOI, although Crossref has one whose volume and page match the citation:
- README: Schlosshauer 2019 (322, 10.1016/j.physrep.2019.10.001), Romero-Isart (325), Baumann, Chia and Porto (332), Dolan (336), Planck 2018 (340), Hui et al. (346), Bar-Or et al. (347), Schive 2020 (350), Allali and Hertzberg 2021 (354), Marsh and Niemeyer (355), de Salas and Widmark (358).
- Note: Baumann, Chia and Porto (225), Davies and Mocz (229), Gilman et al. (231), Nierenberg et al. (238), Oguri and Kubo (239), Planck (241), Powell et al. 2025 (243), Winn et al. (249), Wong et al. (250).
- Bohren and Huffman (README:327) has no link at all. Its Crossref DOI, 10.1002/9783527618156, is the 1998 Wiley record, not the cited 1983 edition.
The result is inconsistent sibling entries: Schive 2014 has a DOI and Schive 2020 does not; Allali and Hertzberg 2020 has one and 2021 does not. This is not a regression, since none of these DOIs was ever in either file. Fix: add the DOIs (DOI first, arXiv in brackets), or narrow the title, description and commit message. Both the sources and prose reviewers reported this.
-
The workflow comment says "never a run on main", which is wrong (.github/workflows/test.yml:13). Under the default
queue: single, a pending run is cancelled when a newer run queues in the same group, whatevercancel-in-progresssays. If pushes A, B and C reach main while A is running, B is cancelled while pending and never gets a CI result. This repo has merged three times within 5 seconds before (7f207f5, 3175ed2, 0b66947).- The description's "Only a superseded pull-request run is cancelled" overstates it the same way. Its narrower claim, that two quick merges no longer cancel the first run, holds when no earlier main run is in progress.
- This is still strictly better than main's
cancel-in-progress: true. - Fix: reword the comment, or give push events a per-commit group (
github.sha).queue: maxcannot be combined withcancel-in-progress: true. - Both the code and prose reviewers reported this.
-
CORRECTIONS.md:58 no longer matches the README. It still names the Joos–Zeh copies by their preprint years: "Kiefer and Joos 1998; Joos 1999; Kiefer 1999; Joos 2005". README.md:323 now gives 1999, 2000, 2000 and 2005/2006. On main the two files matched. "Joos 1999" can now be mistaken for the README's only 1999 entry, which is Kiefer and Joos, LNP 517. Fix: use the published years, or add the arXiv ids. Both the sources and prose reviewers reported this.
-
The NIST entry has two bare URLs, website first (README.md:338). Every other entry with two links is written
https://doi.org/... (https://...); this one ends in two bare URLs with the DOI second. NIST's own suggested citation puts the URL before the DOI, and the prose refuter treated the order as a nit only. Fix:https://doi.org/10.18434/T4W30F (https://physics.nist.gov/asd). -
Two titles still differ from the published ones (notes/bh-lensing.md:246 and :237). Stott and Marsh reads "axion-like"; Crossref has "axionlike". Mehrgan et al. reads "40 billion solar mass"; Crossref has "40 Billion Solar-mass". The PR edited both lines to add DOIs, and it made a hyphen-only title change to Laroche et al., but not to these two. Which paper is meant is never in doubt.
-
Davoudiasl and Denton 2019 is left in the README Sources with no citation in the README text (README.md:342). The PR removes its only citation there, the M87* row at main README.md:227. The paper is now cited only in notes/bh-lensing.md (102, 207, 228) and tests/bh-lensing.mjs:34. That is the position for which #24 dropped Amruth et al.
- This is contested. The prose refuter upheld it. The sources refuter rejected the same claim, filed as a separate finding. Its reasons: the README Sources already list test-only references (Tegmark, Bohren and Huffman, Griffiths and Schroeter, Schut et al.), and
npm teststill runs a check against this paper, whereas no test used Amruth. - Decide which policy the list follows.
- This is contested. The prose refuter upheld it. The sources refuter rejected the same claim, filed as a separate finding. Its reasons: the README Sources already list test-only references (Tegmark, Bohren and Huffman, Griffiths and Schroeter, Schut et al.), and
-
The note still says the cloud "circles" (notes/bh-lensing.md:53). The line reads "A cloud grows only while it circles more slowly than the horizon spins". That keeps the orbit picture the PR removes elsewhere, citing CORRECTIONS item 12 (CORRECTIONS.md:79–81). README:279 states the same condition without it ("a state whose pattern turns slower than the horizon grows"). The line is already on main, but the cleanup is incomplete. Fix: "only while its pattern turns more slowly than the horizon spins".
-
"Bohr speed" names the n = 2 value here (notes/bh-lensing.md:44, 67, 199, 213). The term usually means αc, the n = 1 value and CODATA's atomic unit of velocity. The note uses it for αc/2, the |211⟩ value. Line 213 also pairs the n = 1 "Bohr radius" (atom.js a0 = r_g/α²) with this n = 2 "Bohr speed". The numbers are right, and line 44 states αc/2. Fix: "the n = 2 Bohr speed" or "the |211⟩ cloud's speed, αc/2".
-
"100%" can sit beside a nonzero share (index.html:430). Rounding every share ≥ 0.995 to "100%" produces rows like gap 250 nm, cryo, Virus: "Leftover helium atoms 100% | Its own heat glow 0.37% | ...". This happens in 30 of the 1078 panels scanned.
- Rows already summed past 100% on main because of 2-significant-figure rounding (102 panels), and these 30 panels showed "1.0e+2%" on main, so this is still better than main.
- Fix: a cap such as ">99%".
toFixed(1)would still print "100.0%" at ≥ 0.9995.
-
Two dropdown labels still use exponent notation; this is on main too (index.html:360). The same
toPrecision(2)problem makes the #bhDist options read "8.5e+2 AU" and "1.3e+2 light-years". In Chromium these are the only exponent strings left on the page. fmtLen (index.html:314) has the same latent [99.5, 100) case. In Node's V8,Math.pow(10,-4)is 0.00009999999999999999, but the bundled Chromium shows "100 µm", and a sweep of the gap slider from −9 to 20 found no exponent. -
The page's Sources line has two leftovers; both are on main too (index.html:287). It still links Bild et al. 2023 (arXiv:2211.00449), which nothing on the page uses and which #24 dropped from the README. Its Danielson link has no version, while the README pins v2. That link resolves to v2 today and the page cites no equation numbers, so the version gap has no effect. Both are outside this PR's scope.
Reported and refuted:
- Davoudiasl and Denton as an orphan in the README Sources, as the sources reviewer filed it: refuted by its refuter. The same claim filed by the prose reviewer survived; see item 6.
- README.md:323, the Joos–Zeh bullet being one 608-character line: refuted. Main already had all four sources in one bullet (fff449b), the PR only added journal references and DOIs, and no line-length rule exists.
- The prose reviewer's side point that the NIST link order breaks the "DOI first" convention: rejected by that refuter. The formatting inconsistency survived from the sources reviewer as item 4.
Verdict: merge
Generated by Claude Code
…T link From the non-blocking findings of #27's independent review: - CI (finding 2): with cancel-in-progress limited to pull requests, a pending run on main could still be cancelled when a third push queued (default queue: single). Pushes now get a concurrency group of their own commit, so no run on main is cancelled or skipped. A pull request's runs still share one group and cancel the superseded run. - DOIs (finding 1): #27 said every source with a DOI had one; eleven README entries did not. Each DOI is taken from the paper's arXiv record and checked on Crossref (title, journal, volume, page) and on doi.org. Bohren and Huffman stays without one: Crossref's DOI is the 1998 reprint, not the cited 1983 edition. - NIST ASD (finding 4): DOI first, website in brackets, as every other two-link entry. - notes/bh-lensing.mjs prints "n = 2 Bohr speed" (finding 8; the note itself follows in the next commit). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
…28) Stacey et al. (A&A 703, A285, 2025) re-analysed SDP.81's ALMA data and find the 10^9 Msun sub-halo (Hezaveh et al. 2016) most likely spurious, an artefact of poorly fitted multipoles. The note leaned on SDP.81 as the one anomaly lens with a black hole bound. Now: - Summary, section 5 and section 6: no anomaly lens in the note has both an undisputed small-scale anomaly and a bound on its black hole. CORRECTIONS item 16 records the old claim. - Section 4: the negative claim (no study tests lensing anomalies against the lens galaxy's black hole mass or spin) still holds. It is now dated 18 October 2026, says what counts as a test and how the search was run, and separates the source-quasar black hole work (Morgan et al. 2010, Reis et al. 2014). - B1938+666: Vegetti et al.'s 2026 measured profile is what every reading explains. Self-interacting and stripped cold-dark-matter readings (Zhang and Yu 2026; Zeng et al. 2026, a preprint) sit beside Oguri and Kubo's wave dark matter one. - HS 0810+2554: wave halos (Zhou et al. 2026), cold dark matter with multipoles (Miller and Williams 2025) and with subhalos (Gong et al. 2026, a preprint) all fit; none includes a black hole. - "Central images are the only route" is gone: lensed images near the centre have weighed ~10^10 Msun black holes (Nightingale et al. 2023; Melo-Carneiro et al. 2025), ten times heavier than section 1's. - Wave dark matter models of the main lens mention its black hole only in passing (Chan et al. 2020; Palencia et al. 2025, who defer its effect on the core). Every new citation was checked against the paper's full text, and the draft was then read by two critics (citation fidelity; physics and restraint); all their must-fix and should-fix items are applied. Also from #27's review: the note's remaining DOIs, two titles matched to the published ones, "pattern turns" for "circles", "n = 2 Bohr speed" for the |211> value, and CORRECTIONS item 9's Joos-Zeh copies by their published years. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2
#23 and #24 merged while this was open and covered part of the same ground: the gravitational-atom cross-check, the atoms bullet, the Danielson, Riedel and Scrimgeour references, and the unused sources. I merged main in (ce18c9c) and kept main's version wherever the two overlapped. #23's cross-check is the stronger of the two, and #24's choice to drop Bild et al. and Amruth et al. from Sources stands. What remains here does not overlap them.
Sources: every citation checked against primary records
Follow-ups from #22's review not covered by #23 or #24
toPrecision(2)writes 99.5–100 as "1.0e+2", so "Who's watching" read "1.0e+2%". This happens for any object whose main witness holds ≥99.5% of the records: on main, 414 of 2,172 values the review scanned. Reproduced in Chromium on main, and fixed here.Checked
npm test: 4 files, all pass. The workflow's other three steps pass too.Part of #21.
🤖 Generated with Claude Code
https://claude.ai/code/session_01XfgH6WcwTueUFZKb6fS2s2