{
 "about": "Hand-kept: the stages of paper 1's production, the checklist before prose, and the journal. The numbers live in ledger.json (generated); this file says what has been done and what hasn't. The process page (four/papers/) renders both.",
 "paper": {
  "id": "p1",
  "title": "Choosing a charge law for four dimensions: what 1/r does to atoms and the periodic table",
  "status": "in preparation: restructured around the charge law; seven figures and the methods drafted; the counterfactual table queued",
  "started": "2026-11-04",
  "bar": "arXiv's: checkable claims, verified references, reproducible numbers. Every number in the text comes from a baked file and is checked against it."
 },
 "authors": [
  {
   "name": "Claude",
   "affiliation": "Anthropic",
   "role": "computation, code, analysis, drafting"
  },
  {
   "name": "Astra",
   "role": "independent audit: the failure contracts behind selftest 58, and the audit rounds for this paper",
   "note": "author, second (PI, 2026-11-11)"
  },
  {
   "name": "minor mobius",
   "note": "pseudonym",
   "role": "direction, judgement of results, PI (last author)"
  }
 ],
 "authorship_note": "Settled (PI, 2026-11-11): the paper is published in four.mino.mobi's own journal, byline Claude, Astra, minor mobius (last, PI), with an author-contributions statement saying what each did and that Claude and Astra are AI systems. arXiv doesn't accept AI models as authors; if the paper ever goes there, the byline is minor mobius and the contributions statement carries the rest.",
 "pipeline": [
  {
   "id": "solver",
   "name": "The solver",
   "what": "atom.js: the radial equation with l + (D − 3)/2, Numerov on a log grid, a self-consistent field with Hartree and local exchange, the ground-configuration search",
   "path": "atom/js/atom.js",
   "status": "built"
  },
  {
   "id": "bake",
   "name": "The bake",
   "what": "every element in both worlds, solved once (on the runner for Z past ~80) and frozen: configurations, energies, orbitals, densities",
   "path": "atom/tools/bake.mjs → atom/js/table.js",
   "status": "built"
  },
  {
   "id": "selftest",
   "name": "The gates",
   "what": "the selftest's blocks recompute a sample of the bake and check each method against closed forms and published values; block 67 checks the paper's own data and every claim in its ledger",
   "path": "atom/test/atom4.selftest.mjs",
   "status": "built"
  },
  {
   "id": "laws",
   "name": "The law calculations",
   "what": "one electron under Gauss's 1/r² (no atom: nothing binds, or it falls to the centre) and under 1/r^p (only p = 1 keeps a shell together)",
   "path": "papers/p1/laws.js",
   "status": "built"
  },
  {
   "id": "audit",
   "name": "The solver audit",
   "what": "the bake's stop against one a hundred-thousandfold tighter, on 19 atoms across every row of both worlds: what in the table is converged and what isn't",
   "path": "papers/p1/audit-scf.mjs → data/audit-scf.json",
   "status": "built"
  },
  {
   "id": "data",
   "name": "Data and ledger",
   "what": "the paper's tables as JSON and CSV, each figure's data, and the ledger of claims, all generated from the bake",
   "path": "papers/p1/build.mjs → data/, figdata/, ledger.json",
   "status": "built"
  },
  {
   "id": "refs",
   "name": "References",
   "what": "one entry per source, marked read, index, cited or memory; checked against Crossref; only read and index may be cited",
   "path": "papers/p1/refs.json",
   "status": "built"
  },
  {
   "id": "reproduce",
   "name": "Reproduction",
   "what": "re-solves a sample of the paper's atoms from the solver and matches the data to the bit; ships in the open mirror",
   "path": "papers/p1/reproduce.mjs",
   "status": "built"
  },
  {
   "id": "counterfactual",
   "name": "The counterfactual table",
   "what": "every element solved again with the shell theorem forced on, everything else held: what the table owes to the law rather than to the geometry",
   "path": "atom/tools/lawtable.mjs → data/shelltable.json",
   "status": "partly",
   "note": "112 of 140 settled after lawtable-repair; row 6's f block (89–116) has no settled ground state under the shell theorem"
  },
  {
   "id": "widgets",
   "name": "Widgets",
   "what": "screening on and off, solved live; a neutral atom's far pull",
   "path": "papers/p1/widgets/",
   "status": "built"
  },
  {
   "id": "figures",
   "name": "Figures",
   "what": "SVG drawn from the bake and the law code by a script on every build, never by hand (Okabe–Ito palette, the shared library's theme); captions beside them in the same file",
   "path": "papers/p1/figures/make.mjs → figures/*.svg",
   "status": "partly",
   "note": "pruned to six for the main text and two for the appendix; 6 waits on the counterfactual run"
  },
  {
   "id": "text",
   "name": "The text",
   "what": "each section's source with every number a token the build fills from the code or the bake; an unknown token stops the build",
   "path": "papers/p1/text/*.md → draft/*.md",
   "status": "partly",
   "note": "methods and part I (the law, sections 2–4) drafted and rendered"
  },
  {
   "id": "page",
   "name": "The paper page",
   "what": "each draft section rendered by pandoc with its maths as MathML (browsers draw it natively, no maths library downloaded) on the paper stylesheet forked from wormhole's; a page carries its source's hash, so the selftest catches a stale one",
   "path": "papers/p1/page.mjs → draft/*.html",
   "status": "partly",
   "note": "methods rendered; the full assembled paper is to come"
  },
  {
   "id": "pdf",
   "name": "PDF",
   "what": "the page printed headlessly",
   "path": "papers/p1/p1.pdf",
   "status": "to do"
  },
  {
   "id": "mirror",
   "name": "Open code",
   "what": "the solvers, the reproduction and the data, MIT, on tangled; staged and tested on every push",
   "path": "tangled.org/morphyxmino.bsky.social/four",
   "status": "waiting",
   "note": "the repo has to be created on tangled by its owner"
  },
  {
   "id": "audit",
   "name": "Audit",
   "what": "Astra's rounds on the paper's claims and code",
   "path": "",
   "status": "to do"
  }
 ],
 "checklist": [
  {
   "item": "Before the paper is assembled: every author–year in the text must be a read or index reference. Part I cites Ehrenfest 1917, still only cited",
   "status": "done",
   "note": "Ehrenfest 1917 confirmed via indexes (read 26 May 1917, Proc. Amsterdam 20, 1918); 1920 via a scan of the original"
  },
  {
   "item": "The counterfactual table: run lawtable-shell, bake it, add its claims, and say whether the generalized Madelung rows come from the law or from the geometry alone",
   "status": "partly",
   "note": "112 of 140 settled: rows the geometry's through row 5, the d-and-f order and the binding the law's; 89–116 don't settle under the shell theorem (next: a homotopy from the law's converged field)"
  },
  {
   "item": "Element 128 out-holds its row's noble gas: is it partial screening? Read it off the counterfactual table",
   "status": "done",
   "note": "yes, the law: 3.1 eV with the shell theorem against 29.6 under the law"
  },
  {
   "item": "Choose the title: the table first, or the law first",
   "status": "open",
   "note": "left to the end, with the abstract (the PI may write it by hand)"
  },
  {
   "item": "Find citable sources for Hardy's inequality in D dimensions and the fall to the centre (Landau–Lifshitz §35, to check)",
   "status": "done",
   "note": "Landau–Lifshitz §35 'Fall of a particle to the centre' (3rd ed. contents); Hardy's constant via Barbatis, arXiv:2311.08017"
  },
  {
   "item": "Forward citation search on Pyykkö & Zhao (1991) and Négadi & Kibler (1996) in a citation index, for any later self-consistent table at D = 4",
   "status": "to do",
   "note": "a web search found none; the paper says \"to our knowledge\" until an index search is run"
  },
  {
   "item": "Turn every cited reference into read or index",
   "status": "partly",
   "note": "28 of 30: madelung1936 (the rule's page) and nist-asd (the promotion energies) remain"
  },
  {
   "item": "Solver audit: atom.js's self-consistent field stops on the total energy. Check whether that leaves the far density unconverged, and what it does to the table's energies and ionizations",
   "status": "done",
   "note": "19 atoms, both worlds, stop 10⁻⁷ → 10⁻¹² hartree: energies and ionizations move ≤ 3 × 10⁻⁷ hartree, Σ⟨r²⟩ ≤ 2 × 10⁻⁵ relative. The far tail is unconverged only where it is noise (10⁻¹³ electrons per bohr and below). Nothing quoted depends on it."
  },
  {
   "item": "Convergence runs for appendix C: grid (N, r₀, r_max), SCF tolerance, a few elements per row",
   "status": "done",
   "note": "11 atoms, both worlds, six grids (N 500/1000, r₀ 10⁻⁴/10⁻⁶, r_max 100/250): ionizations move ≤ 1.4 × 10⁻⁴ hartree, total energies ≤ 2.4 × 10⁻⁵ of themselves (worst: our krypton's deep 1s)"
  },
  {
   "item": "Re-derive the 3D validation from the current bake",
   "status": "done",
   "note": "32 of 36; median ionization error 4.3 %"
  },
  {
   "item": "Decide the Z range in the main text: to 78 with 79–140 marked first-pass, or all 140 with the relativistic caveat",
   "status": "open"
  },
  {
   "item": "The equation pipeline: KaTeX pre-rendering in the sandbox, or hand-written MathML",
   "status": "done",
   "note": "pandoc → MathML (page.mjs); methods rendered"
  },
  {
   "item": "Tag the mirror version the paper cites",
   "status": "to do"
  },
  {
   "item": "The term-corrected table (termtable.mjs, runner run hund-f-and-terms): which elements change configuration, how far the ionizations move; make it the main table (data, figures, section 8 tokens) and move the Hund-spin table to supplement S1",
   "status": "done",
   "note": "no configuration changes from the terms; 8 ionizations ±0.30–0.67 eV; element 82 corrected by the wider search; main table now, S1 = Hund-spin"
  },
  {
   "item": "Hund's rule over the rest of the 4D f block (4f⁵–4f³¹, elements 85–110, runner); fold into data/hund.json (hund-collect.mjs) and section 7.2",
   "status": "done",
   "note": "4f²⁹ (108) breaks it too (0.26 eV); every other filling holds (bounds for f⁶–f²⁵); 102 and 106 unfinished"
  },
  {
   "item": "Prior art on Hund's rules in D dimensions: three web searches found none; run a citation-index search before the paper calls section 7 new",
   "status": "to do"
  },
  {
   "item": "Section 8.2 quotes palladium 8.3 eV and xenon 12.1 eV from memory: cite a measured source (NIST ASD)",
   "status": "done",
   "note": "NIST Handbook: Pd 8.3369 eV, Xe 12.1298 eV"
  },
  {
   "item": "Introduction and abstract, written last against the finished Parts I–III",
   "status": "done",
   "note": "text/intro.md; the whole draft assembled as draft/paper.{md,html,pdf}"
  },
  {
   "item": "The site's own table chapter and js/table.js still show the bake (element 82 as 4f² 6s², the Hund-spin energies): rebake with the term-corrected search, or overlay data/termtable.json on the site",
   "status": "done",
   "note": "atom/tools/terms.mjs bakes js/terms.js; js/tablec.js lays it over the bake; every page reads the table from tablec.js; selftest 69 checks it against data/termtable.json"
  },
  {
   "item": "The counterfactual table compares against the bake's configurations; element 82 now differs in the main table: re-check its counterfactual entry against 4f¹ 5d¹ 6s²",
   "status": "done",
   "note": "82 under the shell theorem is 6s² 6p², different from both its old and new configurations; 'changed' and 'the law's own configuration is lower' stand"
  },
  {
   "item": "Figures 2, 3 and 6 (numbered 1, 4 and 6 before the renumbering) draw from atom/js/table.js (the bake), not the corrected table: point them at build.mjs's rows (differences ≤ 0.67 eV in eight ionizations, and element 82's configuration)",
   "status": "done",
   "note": "figures/make.mjs lays data/termtable.json over the bake, as build.mjs does"
  },
  {
   "item": "Check the measured 4s→3d promotion energies (section 6.3) and Pd/Xe ionizations (8.2) against NIST ASD and cite with the access date",
   "status": "to do"
  },
  {
   "item": "Appendices A–E (the 4D kernel and shell expansion, the electron-gas constants, convergence, the law calculations, reproducing every number)",
   "status": "done",
   "note": "text/appendix.md; A carries the kernel's closed form (kernelclosed)"
  }
 ],
 "journal": [
  {
   "date": "2026-10-25",
   "what": "The write-up pinned: a kit survey of what the repo already had for a paper, and the decision that writing waits while the science moves. The literature register opened, each source tagged by how far it had been checked."
  },
  {
   "date": "2026-11-04",
   "what": "A lesson from elsewhere in the project: the plane-wave code's pseudopotentials were wrong because their all-electron atom stopped iterating on its eigenvalues while its far density was still unsettled. The atom solver behind this paper stops on its total energy. That goes on the checklist as an audit, not an assumption."
  },
  {
   "date": "2026-11-04",
   "what": "Writing started. The outline maps every claim to the baked file and the selftest block that hold it up; the figure plan and the companion objects are listed with it."
  },
  {
   "date": "2026-11-04",
   "what": "The data release and its gate: build.mjs writes the tables, the figure data and a ledger of claims from the bake, and selftest 67 fails if any of it goes stale or any claim stops holding. Re-deriving the 3D check gave 32 of 36 configurations (Ti, V, Co and Ni each put one electron too many in 3d) and a 4.3 % median ionization error."
  },
  {
   "date": "2026-11-04",
   "what": "The references checked against Crossref. It corrected four entries the register had from memory or from other papers' citations: Al-Jaber's volume and pages, Loos & Gill's, Kventsel & Katriel's title (it's the Thomas–Fermi atom in N dimensions), and Loeser & Herschbach 1986, which is two papers: the exact energies at 84, 3882 and Hartree–Fock at 84, 3893."
  },
  {
   "date": "2026-11-04",
   "what": "The reproduction: a script that re-solves a sample of the paper's atoms with the bake's own function. 14 of 14 matched the data to the bit, in 18 s on four cores. It ships in the open mirror, whose workflow runs it on every push."
  },
  {
   "date": "2026-11-04",
   "what": "Two widgets. Screening on and off solves any 4D atom live, with and without the shell theorem, and gives back the bake's numbers (element 12 holds its outer electron at 18.39 eV against 10.30; element 138 at 26.71 against 9.22). The far pull walks an electron away from a neutral atom; the full cloud meets −S/8s³ to within a fraction of a percent past 20 bohr."
  },
  {
   "date": "2026-11-04",
   "what": "This page: the paper's production made public as it happens, its claims recomputed from the bake on every build."
  },
  {
   "date": "2026-11-04",
   "what": "Restructured around the charge law, at the PI's direction: the choice of law, why it has to be made and what it costs are now the first part of the paper, and the table is measured against a counterfactual. New ledger claims, computed: Gauss's 1/r² in 4D binds nothing below α = ½ and falls to the centre above it (the ground level scales as 1/r₀² with the grid's inner end, constant to four figures over three decades); among power laws only 1/r keeps 2s with 2p (10⁻¹¹; −6 % at p = 0.8, +15 % at p = 1.2); ∇²(1/r) = −1/r³; k(1) = 8/3π. Five references for the law confirmed on Crossref."
  },
  {
   "date": "2026-11-04",
   "what": "A first look at the counterfactual, in the sandbox: with the shell theorem forced on, element 26 settles in 4s² 4p² instead of 3d² 4s² (even when seeded with the real configuration) and its ionization halves; element 100 fills its f shell to 4f³⁰ before 5p. The law doesn't just tighten the binding: it decides where the blocks begin. The full table is queued on the runner."
  },
  {
   "date": "2026-11-04",
   "what": "Seven figures, drawn from the bake and the law code by a script that runs with every build (so a stale figure fails the selftest like a stale number): Gauss's law's fall to the centre, the shell splitting against the exponent, the 3-sphere kernel, ionization against Z in both worlds, the 3D validation, partial screening, and the far pull."
  },
  {
   "date": "2026-11-04",
   "what": "The first figure of ionization against Z showed what the outline had wrong: the hardest-held electron in the 4D table isn't a noble gas's. Element 128, a full 18-electron d shell with nothing outside it (our palladium's kind), holds its outermost electron at 29.56 eV against its row's noble gas 138 at 26.71; element 68, the same kind a row up, comes within a volt of its row's. In ours palladium holds far less hard than xenon. A first wording of the claim said 68 also beat its row's noble gas; the ledger's own value showed it doesn't (22.32 against 78's 23.10), and the claim was corrected before it went in. Whether partial screening causes this is a question for the counterfactual."
  },
  {
   "date": "2026-11-04",
   "what": "The methods section drafted. Its numbers (grid, tolerances, mixing, the kernel's quadrature, the exchange constants, the search's rounds) are tokens the build reads from the solver's own source, so the text says what the code does."
  },
  {
   "date": "2026-11-04",
   "what": "The figures pruned, at the PI's direction: half of the first seven only illustrated. The main text now has six: the table itself; the choice of law (Gauss's fall to the centre beside the shell splitting against the exponent); orbitals only four dimensions has, cut through the nucleus by a plane on solved atoms' radial functions (the fourth p orbital along w, and a d orbital positive all round one plane and negative all round its completely perpendicular one); ionization against Z; six elements in profile; and the counterfactual (a stand-in until its run). The kernel and our-world validation moved to the appendix; the far-pull curve, the α scan and the shell's potential were cut. The six profiled atoms are the bake's own, re-solved with their radial functions kept; the build stops if one drifts from the bake."
  },
  {
   "date": "2026-11-04",
   "what": "The solver audit, the methods section's one open item: 19 atoms across every row of both worlds re-solved and driven from the bake's stop (energy change under 10⁻⁷ hartree) to one under 10⁻¹². Total and ionization energies moved by at most 2.8 × 10⁻⁷ hartree (8 μeV), Σ⟨r²⟩ by 2 × 10⁻⁵ of itself. The far density is not converged relatively past about 15 bohr, but only where it is noise, 10⁻¹³ to 10⁻¹⁷ electrons per bohr; atoms with a loose s electron, which do put charge out there, are converged to 10⁻⁶. So the plane-wave bug's lesson doesn't touch the table: it bites what builds a potential from the far density through exchange, which the pseudopotentials now converge pointwise. A new ledger claim holds it."
  },
  {
   "date": "2026-11-04",
   "what": "The kit, re-surveyed: pandoc 3.1.3 is now in the sandbox (the survey found none), which turns the text's maths into MathML that browsers render natively and writes the LaTeX source arXiv compiles itself, so the paper needs neither a downloaded maths library nor a local TeX. Still no TeX engine; KaTeX is fetchable but not needed."
  },
  {
   "date": "2026-11-04",
   "what": "The counterfactual answered part III's question. With the shell theorem forced on, the first 24 elements and all six noble gases come out the same, so the rows the generalized Madelung rule predicts are the geometry's (four-dimensional shells), not the law's. What the law decides is everything inside the rows from the first d element on: 52 of the 107 atoms that settled change configuration, every one from element 25, because without the law p beats d; outer electrons are held a median 2.5 times harder under the law; and element 128's record 29.6 eV falls to 3.1, the noble gases holding hardest as ours do. In every changed atom the law's own configuration still comes out lower under the law, so the comparison holds both ways."
  },
  {
   "date": "2026-11-04",
   "what": "And what the counterfactual didn't settle, caught before it went in: for 33 atoms (38, 39, 41, 88–117) no candidate's field converged, and the search reports an unsettled filling with no energy. A first count had all 84 differing configurations as results; 33 of them were those fillings. Counted honestly it is 52 of 107. One unsettled filling, element 110 as 4f³² 5s² 5p⁸, is a closed shell, which would split row 6 if it held, so 'the rows are the geometry's' waits on a wider re-search with gentler mixing (lawtable-repair, queued; element 38 settled in two minutes in the sandbox)."
  },
  {
   "date": "2026-11-04",
   "what": "pw-fixed came in (the 3D check of the plane-wave code, for the molecules paper): corundum lowest and the order right, θ within 0.11 eV of same-functional PySCF, but the impostor 0.26 and poly6 0.40 off, against the 0.2 eV bar agreed. The stop-loss applies there; it doesn't touch paper 1."
  },
  {
   "date": "2026-11-04",
   "what": "The counterfactual's repair run: of the 33 atoms that hadn't settled, a wider search (every configuration in reach, mixing 0.15, 1500 iterations) settled 38, 39, 41, 88 and 117 (38–41 again put an electron in 4p rather than 3d). Three more, 114–116, 'settled' on configurations an electron would rather leave (negative ionization energies): the search missed lower ones, so they don't count. The other 25, 89–113, the heart of row 6's f block, don't converge at all under the shell theorem. Every atom up to 88 is now settled, so the rows are the geometry's through row 5; whether the shell theorem would close a shell at 110 and split row 6 stays open. Next: follow each atom from the law's converged field to the shell theorem's in small steps of the kernel, instead of searching."
  },
  {
   "date": "2026-11-04",
   "what": "Methods finished as a draft. Grid convergence, appendix C's promise: 11 atoms from every row of both worlds re-solved on six other grids (half and twice the points, the inner end ten times out or in, the outer end at 100 or 250 bohr). No ionization moved by more than 1.4 × 10⁻⁴ hartree (4 meV); total energies by at most a few parts per million, the worst our own krypton, whose 1s is deeper than any 4D atom's. A new ledger claim holds it. The counterfactual's paragraph now says how atoms that wouldn't settle were searched again and what counts as settled. And the methods section is the first rendered page of the paper: pandoc turns its maths into MathML on a stylesheet forked from wormhole's journal pages."
  },
  {
   "date": "2026-11-04",
   "what": "Part I drafted: why a law has to be chosen (Gauss's law in four dimensions has no atom: Hardy's floor, and the computed fall to the centre), the ways out and what each costs, why 1/r exactly (the only power law that keeps a shell together), and what it costs in four dimensions (not harmonic, no shell theorem or Faraday cage, the five routes to a field theory and the lens vacuum this universe takes, what is declared alongside, and the evidence that matter is stable). Every number in it is read from the ledger or the code. One new claim backs the stability section: the solved table's total energies follow each world's Thomas–Fermi law, a fitted exponent of 2.53 in four dimensions against 5/2, and 2.39 in ours against 7/3."
  },
  {
   "date": "2026-11-06",
   "what": "Hund's first rule doesn't hold in four dimensions, and the paper now says so. A crystal-field side project turned up the free 4D d² ion with a spin-singlet ground term. A survey of every open subshell in the table (tools/hund.mjs: degree-l harmonics on the 3-sphere, the law projected onto them, every M_S block by Lanczos) finds that it fails for d² and d¹⁶ in all three d rows (elements 26, 39, 54, 66, 114, 126, by 0.30–0.77 eV) and for 4f³ (element 83). It holds for every p filling, every other d filling, 4f² and 4f⁴, and for every open subshell of our own table. It isn't the law: Gauss's 1/r² gives the same order on the same orbital. The PI's call: the table corrected for it (tools/termtable.mjs, now on the runner) becomes the main table, and the Hund-spin table moves to a supplement. Parts II and III drafted around it: checks against what is known (6), Hund's rule in four dimensions (7), the table (8), the counterfactual (9), partial screening (10), neutral atoms (11), limitations (12) and the conclusion (13). Four new ledger claims back section 7 (hund3, hund4, hund4f3, hundlaw), 23 in all, every one holding."
  },
  {
   "date": "2026-11-06",
   "what": "Why Hund's rule breaks, made exact, and an explainer for it. For two d electrons every term's energy is F⁰ + a₂F² + a₄F⁴, where a₂ measures how much the term lets the two electrons line up. In 4D the ten-state singlet keeps them nearest right angles (⟨cos²γ⟩ 1/8, half of random; the best triplet 0.19). It lies below every triplet whenever F⁴/F² < 5/3; element 26 sits at 0.39 under the law and 0.65 under Gauss's. In ours ³F wins for every value. Section 7.3 rewritten around it, with a 24th ledger claim (hundcrit). The interactive version is four.mino.mobi/atom/hund.html: orbitals as boxes, the d orbitals on the sphere and on slices of the 3-sphere, the pair's separation-angle distribution for every term, the level ladder against F⁴/F² with real atoms marked, and the survey across the table."
  },
  {
   "date": "2026-11-06",
   "what": "The f block surveyed (runner run hund-f). 4f³'s mirror 4f²⁹ (element 108) breaks Hund's rule too, by 0.26 eV, spin ½. Every other f filling keeps it: completely for f², f⁴, f⁵, f²⁸ and f³⁰, and for f⁶–f²⁵ over the highest two or three spins, the lower-spin blocks running to 10⁶–10⁸ arrangements, so a bound. Elements 102 and 106 didn't finish. So in four dimensions the rule breaks two from either end of the d shell and three from either end of the f shell, and nowhere else surveyed. Ledger claim hund4f29 makes 25, all holding."
  },
  {
   "date": "2026-11-06",
   "what": "The term-corrected table is in (runner run hund-f-and-terms, 92 d- and f-block elements searched again with neighbour seeds, every candidate corrected for its open subshells' Hund gaps) and is now the paper's main table. The Hund-spin bake is supplement S1 (data/table4-hundspin.*). The terms change no ground configuration. They lower the eight Hund-breaking atoms' energies and move eight first ionizations, in pairs, by 0.30–0.67 eV: up for 54, 66, 114, 126, down for the next element, whose ion lands on the broken filling (55, 67, 115, 127). The wider search found one configuration the bake had missed: element 82 is 4f¹ 5d¹ 6s² (our cerium's kind), 0.36 eV below 4f² 6s². Rows, noble gases, element 128's lead and the 2.54 hold ratio are untouched. Ledger claim termtable makes 26, all holding; §7.4 and a methods paragraph written."
  },
  {
   "date": "2026-11-06",
   "what": "The full term correction, tested and rejected. Beyond Hund's first rule, the solver also averages every open subshell over orbitals, so we tried correcting each candidate by its whole term structure: lowest term less the average of the maximum-spin arrangements. That average has a closed form, checked against brute force. Tested on ours first (TERM_MODE=full, elements 5–36): it fixes none of the four misses and breaks two that were right (Sc, Fe), 26 of 32 against 28. The 4s→3d promotion energies against measured ones show why. The solver comes out too cheap for d in every 3d atom, by 0.35–1.70 eV, a mean 1.06 (local exchange without correlation), and the full correction leaves that mean where it is (1.07). So the paper corrects only Hund's first rule, the one thing a spin-polarized averaged field cannot represent, and states the bias. It also gives a robustness measure: 11 of the 92 4D d- and f-block elements have a fewer-d competitor within a volt (element 53's is 0.01 eV). New data/promotion3.json (promotion-collect.mjs), ledger claim promotion (27, all holding), §6.3, §7.4 and §12 updated, reference nist-asd added (from memory, to verify)."
  },
  {
   "date": "2026-11-06",
   "what": "The first whole draft. The abstract and introduction are written (text/intro.md), last, against Parts I–III as they stand. The build now assembles the whole paper as draft/paper.md: abstract and introduction, Part I, the methods as section 5 opening Part II, Parts II and III, the figures with their captions, what is still to write, and the references. It renders to draft/paper.html and prints to draft/paper.pdf (pdf.cjs, headless Chromium, A4). Four references were checked against indexes and sources: Ehrenfest 1917 and 1920 (a scan of the 1920 original), Landau–Lifshitz §35, and Hardy's constant (Barbatis's review). Palladium's and xenon's ionization energies match NIST's tables (nist-handbook), so 28 of 30 references are now citable. Still from memory: the 4s→3d promotion energies (NIST ASD), and the page of Madelung's 1936 book that carries the rule. Working title, law first, pending the PI's call."
  },
  {
   "date": "2026-11-06",
   "what": "The PI's call: no title and no abstract until the end, and the abstract may be hand-written. Both are now placeholders in the whole draft. The papers page leads with a 'read the current draft' link (draft/paper.html, and the PDF). The appendices are written (text/appendix.md): A, the 4D kernel; B, the 4D electron gas; C, numerical convergence (a table of the six grids); D, the law calculations; E, reproducing every number (every data file, what made it and where). A turned up a closed form, k(t) = 4[(1 + t²)E(t) − (1 − t²)K(t)]/3πt² in complete elliptic integrals, equal to the solver's quadrature to 3 × 10⁻¹⁵; ledger claim kernelclosed makes 28, all holding."
  },
  {
   "date": "2026-11-07",
   "what": "Ions (runner run ions-1: every cation and anion of both worlds searched for its own ground state, 350 searches, all exit 0). In ours the cation search changes no ionization. The anion search gets all seven unbound anions unbound and every p-block acceptor bound (C 1.32 against 1.26, Ge 1.25 against 1.23, F 2.80 against 3.40). It fails where correlation binds (H⁻, the alkalis, P, As), so in 4D an affinity above about 0.7 eV is believed. In 4D, element 128's 29.56 eV survives. Seven ions rearrange (36, 93–95, 103–105), lowering those ionizations by up to 1.29 eV, and the main table carries them. A new result: one electron short of a full 4D p shell, atoms hold an extra electron harder and harder down the group (3.4, 5.7, 7.8, 10.3, 12.9 eV) where ours stay level near 3.4, partial screening again (§10.3). 80 of 140 bind an extra electron by more than 0.7 eV; the light noble gases don't, the heavy ones come out barely bound. New: data/ions.json (ions-collect.mjs), an electron_affinity_eV column, three claims (ions3, ions4, halogens4), 31 in all, holding."
  },
  {
   "date": "2026-11-07",
   "what": "Editorial pass on Part I after the PI's read of sections 1–4. Explanations replace the oracles: §2.1 now derives the collapse (the Hamiltonian's scale symmetry; r^σ near the origin, σ = −1 ± √(1 − 2α), complex past α = ½, so infinitely many nodes and no lowest level) and names the solver (laws.js: Numerov on a log grid, node-count bisection); §3.1 explains the hidden symmetry (Bertrand's closed orbits, the Runge–Lenz vector, SO(D + 1) from Fock's momentum-space sphere, why only 1/r: its Fourier transform 1/k^(D−1) is Gauss's kernel one dimension up) and answers 'why not 1.5' (it binds with a size, but 2p sits 64% of the shell's binding above 2s, so no shells); §4.1–4.2 give the flux and the mean-value reasons; §4.3 explains each route and says the paper uses only the lens vacuum's statics, its dynamics being the later paper; the □ (which rendered as a white box) is now the wave operator written out. References are pandoc citations through citeproc, author–date, each linked to its entry and each entry to its DOI; metadata from Crossref (refs-crossref.mjs → data/crossref.json), with new DOIs for Ehrenfest 1920, Herschbach et al. (1993), Barbatis (2024), Landau–Lifshitz, Sansonetti–Martin and Rienstra-Kiracofe (now index), and Ehrenfest 1917 read from the KNAW scan (Proc. 20, 1918, 200–209). A notation table (text/notation.md) follows the introduction, and the symbols are unified: the trial exponent is η (was p, which is momentum), the shell kernel κ (was k, which is wavenumber), the atom's second moment Q (was S, which is spin), the exchange constant A_D (was C_D, beside the Gegenbauer C_n), fields bold, electron mass m_e, the nonlinear exponent β, elliptic integrals upright. A section 5 subsection, 'The code', says what computes everything."
  },
  {
   "date": "2026-11-07",
   "what": "Figures placed in context at the PI's request: build.mjs puts each figure after the paragraph that first cites it (an appendix figure after its first citation in the appendices), in the whole draft and in each section page, each with an anchor #fig-N; the separate Figures section is gone. The main-text figures are now met in the order 2, 1, 4, 5, 3, 6, so their numbers no longer follow first citation."
  },
  {
   "date": "2026-11-07",
   "what": "Figures renumbered in order of first citation (old 2, 1, 4, 5, 3, 6 → 1–6; make.mjs's figure functions renamed by subject so the numbers live only in CAPTIONS and the text). The paper's figures 4 and 5 (elements in profile, orbitals on the 3-sphere) now exist for every element: the big table's new 'Inside the atom' (table/inside.js, from table/profiles.json, baked by atom/tools/profiles.mjs: all 140 re-solved in the table's configurations, each matching the table's energy, selftest 70). /table/#Z opens on element Z; §8.3 and both captions link it."
  },
  {
   "date": "2026-11-07",
   "what": "§4.3 reframed at the PI's direction: not 'a sixth route' but a downselection. Five routes to 1/k³; four fail outright (none exists, negative probabilities, fields that don't add, three-dimensional atoms); route 3 is taken, and within it the permeability, which the statics leave free, is chosen as μ₀/ℓk: that choice is the lens vacuum. Every choice of permeability gives the same statics, which is why the paper depends on none of it. The site's force chapter now says the same (js/main.js routes, genpages.py), and its □^(3/2) is written out too."
  },
  {
   "date": "2026-11-07",
   "what": "Methods put in the paper's own format: numbered run-in paragraphs 5.1–5.12 under '## 5 · Methods' (it had ### subheadings and unnumbered run-ins), its plain-text maths in TeX. The PI asked whether the table is exactly Négadi and Kibler's: the rows and noble gases are; the filling isn't. 103 of 140 configurations are exactly the strict n + l rule's; 34 move one or two electrons from the outer s into the d or f below (10 of them two), 81 and 82 move 4f → 5d, 113 moves 5d → 6p. The solver over-counts such departures in ours (6 in Z ≤ 36 against the measured 2, Cr and Cu), so the 4D count is likely high. §8.1 says so, from tokens; claim madelungfill."
  },
  {
   "date": "2026-11-07",
   "what": "Négadi and Kibler re-read (arXiv atom-ph/9512001): they state the generalized Madelung rule for any D and work it out for D = 2 (recovering D = 3); they publish no 4D table. The paper had said they 'predicted the rows of a four-dimensional table'; it now says their rule, applied to D = 4, gives those rows (intro, §3.2, §8.1, figure 2, claim rows, refs role, the site's table chapter). §8.1 also now says the rule is a filling order, like Madelung's in ours, which has exceptions (chromium and copper the first of about twenty), so the 37 departures aren't a disagreement with them."
  },
  {
   "date": "2026-11-08",
   "what": "ions-23 (runner, 158 commands all exit 0, 342 ions all settled): every +2 and +3 ion of both worlds searched for its own ground state. Against NIST (now fetched: data/nist-ie.json, H–Kr; it also confirms the paper's measured first ionizations to 0.01 eV), ours' second and third ionizations are off by a median 2.0 % and 1.4 % (worst V 7 %, C 4 %), better than the first's 4.3 %, and our ions lose 4s before 3d as measured. In 4D the 3d row's ions do the same; the first halves of the 4d and 5d blocks, and 81–82, give up a d electron at +1 or +2 while keeping the outer s, as our Y⁺ (5s²) and Lu⁺ (6s²) do (both checked against NIST). §6.4 says so (claim ions23). The nist-asd reference is now read for ionization energies; §6.3's promotion energies are still from memory and the text says so. The big table's 'Inside the atom' offers every element at −1 (where bound), 0, +1, +2 and +3."
  },
  {
   "date": "2026-11-08",
   "what": "Scoping a Hartree–Fock run of the table (hf-scope.py, PySCF in our world). High-spin ROHF with the configuration held gets 8 of 9 of our 3d configurations (misses V), against the local-exchange solver's 5 (misses Ti, V, Co, Ni). Where the two agree they're right 5 of 6; where they disagree ROHF is right 3 of 3. So the second 4D table should be Hartree–Fock, used both as a bracket (agreement = confident) and as the better-founded answer where they differ. Routes in 4D: gauss.js (Gaussian HF, any D, already checked against PySCF) with a held configuration covers Z ≤ 80, the 3d and 4d blocks; exact exchange in the radial solver would cover all 140. Lit hunt delegated to Astra (prior art, Hund in D dims, D = 4 benchmarks, promotion energies)."
  },
  {
   "date": "2026-11-08",
   "what": "Built the Hartree–Fock atom (atom/js/hf.js): numerical, one radial function per subshell, the energy the average over max-spin determinants, any D, the pinned 1/r. Exchange by the Gegenbauer expansion of 1/|x−y| (a_k(t), in 3D t^k), whose radial kernels are convolutions on the log grid (FFT), and angular weights B_k reduced to one zonal integral (in 4D every allowed k weighs (l+1)(l'+1)); Davidson on each subshell's Fock operator. Checks: our He, Be, C ³P, N ⁴S, Ne, Ar, Fe ⁵D, Cu ²S, Kr reproduce the numerical Hartree–Fock limits to 10⁻⁶ hartree or better (Richardson from h = 0.02, 0.01); 4D elements 2, 4, 12, 22 agree with gauss.js (an independent Gaussian code) to its basis limit; all 40 angular weights (s–f, both D) match brute-force harmonics to 10⁻¹⁴; the 4D kernels to 3 × 10⁻⁷. The bracket on our 3d row (atom/tools/hftable.mjs, with the full term correction on Hartree–Fock orbitals): Hartree–Fock 10 of 10 ground configurations (Sc–Zn), local exchange 6; where they agree both right, where they differ Hartree–Fock right. Queued hf-table-1: both worlds, every element. Selftest 71; hf.js mirrored."
  },
  {
   "date": "2026-11-08",
   "what": "hf-table-1 collected (all 84 commands exit 0): every element of both worlds ranked by Hartree–Fock and by local exchange over the same candidates. Gapped configurations (Na 4s¹ with 3s empty) are left out of the Hartree–Fock ranking, since the variational solve returns the compact configuration under another name. Ours: Hartree–Fock 36/36, local exchange 32/36; they agree on 32, all right; they differ on Ti, V, Co, Ni, where Hartree–Fock is right. 4D: agree on 126 of 140; open 30, 31, 37, 38, 40, 58, 61 (d row), 107–111 (f row), 53, 83. data/hftable.json, table/hfverdict.js (the table page's firm/open line), claim hfcheck, §5.12 and §6.6."
  },
  {
   "date": "2026-11-08",
   "what": "The independent literature audit (Astra) checked and imported: 66 references (24 read, 42 index, none unchecked), Madelung now the 1936 third edition, p. 359. Revisions it asked for: a bounded novelty sentence with the earlier D-dimensional atoms; §6.3's measured promotions from NIST's Handbook (lowest-term centroids; the remembered values matched the lowest levels to 0.01 eV; Ni changes sign between the two); the 3d promotion errors show local exchange and Hartree–Fock do not bracket the measured value (both too cheap Sc–Cr), so §6.6 calls their spread method sensitivity; §7.2/7.3 state the f search's real coverage, the fixed-orbital scope and the two-irrep singlet, 'whatever the law' becomes 'under both laws tested'; §6.4's 0.7 eV is a calibration in ours, not a 4D error bar; §4.5 limited to single atoms and separated atoms' tails; §4.3 cites Heydeman et al. for route 2 (the field-strength bound, now in fields.js and selftest 33), Tarasov for route 3's statics, Lindqvist for route 4, and states each exclusion's assumptions; correlation literature cited as omitted, not unknown; Alliluev, Fiorani and Bertrand's translation, Katsura–Aoki (Z ≈ 205 point; 230 is our dirac.js). §6.2 gains Hartree–Fock against Escobar-Ruiz et al.'s correlated 4D helium-like energies (0.012–0.014 Ha above, as it must)."
  },
  {
   "date": "2026-11-08",
   "what": "Queued hf-lanth-1: the lanthanide calibration, ours Cs–Hf by both methods, measured ground configurations from NIST's Handbook (papers/p1/nist-levels.py → data/nist-lanthanides.json), to test Hartree–Fock's large f-row margins on an f block that has been measured."
  },
  {
   "date": "2026-11-08",
   "what": "hf-lanth-1 collected (18 of 18 exit 0): ours Cs–Hf, Hartree–Fock 15/18, local exchange 12/18; they agree on 14, but two agreements are wrong (Ce: both 4f² 6s², measured 4f¹5d¹6s²; Hf: both 5d³6s¹, measured 5d²6s²), so in the f block agreement is not proof. Against measured levels both make an extra f electron too cheap (local exchange by 4.3–4.7 eV for La, Ce, Gd; Hartree–Fock by 0.2–2.4). The 4 eV window, set in local-exchange energy, therefore left out the configurations with one f fewer: terbium's measured 4f⁸5d6s² (0.035 eV up) sat at +4.45 for local exchange and, with a 10 eV window, +0.92 for Hartree–Fock, against the 5.82 margin reported. The 4D f row (81–112) had the same blind spot: §6.6 now calls its configurations provisional, the table page marks them so, and hf-wide-1 reruns ours Cs–Hf and the 4D f row with a 10 eV window."
  },
  {
   "date": "2026-11-08",
   "what": "Paper 1 becomes the atomic reference (the PI: everything but the nucleus). First pieces: §11.1 and table 1, the atoms' response (response-table.mjs → data/response.json): α(0), α(iω), C6, Q, and in 4D C5 = 3Q²/64 and R* = C6/C5 for 27 closed-shell and one-s-electron atoms of both worlds, converged (quadrature 2 × 10⁻⁶, grids 0.5 %), ours against Derevianko–Porsev–Babb 2010 (read from arXiv; noble gases −8 to +4 %, alkalis and alkaline earths +7 to +25 % in α); in 4D R* is under 1.6 bohr for every closed shell from element 12 on, so the permanent pull leads outside the atom, and the trace of the second moment is a requirement on every later force field. §7.5, d² in a ligand field (atom/tools/ligandd2.mjs): on the axes (16-cell) the singlet holds at every field, its ground an orbital doublet; at the tesseract's corners a triplet takes over at Δ ≈ 2.4 eV and only by 0.07 eV; the 24-cell doesn't split d at all. Claims response and ligandd2. hftable.mjs gains a map mode (every one-electron move from the ground, both methods) for the valence energy map, running."
  },
  {
   "date": "2026-11-08",
   "what": "The atomic reference, continued. §8.4 and table 1, valence states (hftable.mjs map → valence-collect.mjs → data/valence.json): every configuration one electron from the ground by both methods, for 10 representative 4D atoms; ours calibrated against NIST's levels (nist-levels.py light and third, 33 states): Rydberg moves to a median 0.3–0.4 eV, but the s→p promotion of atoms with the p shell under half full is too cheap in both methods (by 0.6–2.3 eV), so 4D's second-row promotions (darbon's 2s→2p 3.4/4.0 eV) are probably a volt or two short. §11.2, joining and separating atoms (atom/tools/separate.mjs → separate-collect.mjs → data/separate.json): the light atoms by every method the molecular codes use, with the Gaussian basis's error against numerical Hartree–Fock; homonuclear pairs stretched to 16 bohr from three starts separate into neutral atoms with their own spins, but the bare-nuclei start misses by up to 16 hartree; in 4D two closed-shell atoms keep the permanent pull and Hartree–Fock reproduces −3Q²/64R⁵ (element 12: −0.0119 against −0.0116 mHa at 16 bohr). Claims valence and separate. Queued valence-1 (darbon's DH5: full curvature matrix, balanced alternatives) behind hf-climb-1."
  },
  {
   "date": "2026-11-09",
   "what": "PI's notes, first batch: the checks against our world (old §6) move to appendix E, as supporting evidence rather than part of the argument; sections 7–13 become 6–12, appendix E (reproducing) becomes F, and every cross-reference, the ledger's section fields, the table page's link and the notation table follow. Headings framed as questions or as dances round the section are rewritten to say what the section does (4.5 'Stability: shown for single atoms, open for matter in bulk', 5.6 'The stop, against one 10⁵ times tighter', 6.2 'd², d¹⁶, 4f³ and 4f²⁹ break the rule', 6.3 'The four-dimensional shell breaks it, not the law', …). The introduction's road map is rewritten for the new order."
  },
  {
   "date": "2026-11-09",
   "what": "PI approved both pending results. The f row (appendix E.6, claim frow): runs hf-wide-1 (10 eV window) and hf-climb-1 (Hartree–Fock's own one-electron neighbourhood) close the window gap; our lanthanides calibrate the rest: Hartree–Fock 15 of 18 grounds right, local exchange 12, and in all 17 measured states that change the f count Hartree–Fock makes f too stable, by 0.15–4.1 eV (B). In 4D moving an f electron out costs 5.8 eV (element 90), 12.6 (100), 19.9 (109) against 0.9 in our terbium, so 17 of the 32 f-row elements are firm (81, 88–100, 102–104); open though agreed 82, 84–87; leaning Hartree–Fock 83, 107–111; untested 101, 105, 106, 112. The table page shows each verdict. §7.5, darbon's DH5 (run valence-1, valencetest-collect.mjs → data/valencetest.json, claim valencetest): a true minimum (24 coordinates, 14 curvatures in groups 5 + 4 + 1 + 4, none negative) and below DH4 + H (1.55/2.32 eV, Hartree–Fock/MP2) and DH3 + H2 (2.13/3.43 eV); fragments optimized only in symmetric shapes, so the costs are upper bounds on that count. Appendix F now lists every data file."
  },
  {
   "date": "2026-11-10",
   "what": "PI: the paper had bloated; the main text goes to seven pages at most, with generous appendices. Measured first: pages.cjs lays the draft out as a two-column letter page at 10 pt (Physical Review's measure, roughly), and the main text was 13 pages, 10 600 words. Restructured to 6 pages, 4 300 words. The body is now §1 introduction, §2 the law (old §2–4 condensed: Gauss leaves no atom, the way out, only 1/r keeps a shell, the shell theorem's loss, the lens vacuum and what is declared), §3 a methods summary, §4 the table (rows, element 128, how firm each configuration is, profiles), §5 Hund's first rule, §6 what the law decides (old §8–9), §7 neutral atoms' pull, §8 limitations and conclusion. Moved intact to new appendices: F the law in full (the fall to the centre, the ways out, the symmetry behind the shells, the five field theories, the declarations, stability), G the methods in full (old 5.1–5.13), H Hund's rule in full (the method, the f survey's completeness, the representation theory and caveats, the corrected table, the ligand field), I starting data for chemistry (valence states with table 1, darbon's DH5), J response with table 2 and joining and separating atoms; reproducing is K, and the notation table opens the appendices. Kept in the body at the PI's choice: the law, the table, the counterfactual, Hund's rule breaking and the neutral atoms' pull. Every cross-reference and the ledger's section fields follow the new numbering."
  },
  {
   "date": "2026-11-10",
   "what": "PI's notes while reviewing. (1) The bracket of the pinned law: atom.js takes a law 1/r^η, and run eta-table-1 solves elements 1–100 under η = 0.5 and 1.5 (selftest 72; first look: η = 0.5 fills p before s, η = 1.5 sinks s below p). (2) The electron's two spin states were only declared; §2.5 and appendix F.6 now derive them from the smallest 4 + 1 Dirac electron (a slow one keeps two components of one chirality) and name what stays declared: that the electron is that smallest spinor, and Pauli's exclusion. A four-state electron would double every shell. (3) The solvers' MIT-licensed mirror, tangled.org/morphyxmino.bsky.social/four-atom, is cited in the introduction and appendix K. (4) The interactive table's tiles are links to #Z, and hover previews answer a mouse only, so a tap on a phone pins the element and Inside the atom follows."
  },
  {
   "date": "2026-11-10",
   "what": "Astra's final audit, eight findings; the correction pass. F01: the reproducer compared the bake with the corrected table; it now re-solves each row by the pipeline that made it, and 26, 39, 54, 55 reproduce exactly. F02: 'firm' replaced by separate reports of agreement, coverage and margins; B is a sensitivity scale, not a bound. F03: the Hartree–Fock climb skipped term corrections more than 6 eV up although they reach −7.7 eV; hftable.mjs farcorr computes them (queued, 182 candidates) and every candidate and verdict carries its provenance; 16 f-row verdicts are provisional until they're in. F04: degeneracy, shells and rows kept apart; the rows survive the shell theorem, and eta-table-1 shows they change with the exponent (η = 0.5: peaks at 2, 12, 40, the hydrogen shells; η = 1.5 still repairing). F05–F08: scoped claims and presentation."
  },
  {
   "date": "2026-11-10",
   "what": "Run farcorr-eta-repair-1 collected. F03: all 182 skipped term corrections computed (54 complete, 128 partial); no ranking and no verdict changed; the f-row deciding gaps fell by the competitors' own corrections, up to 7.8 eV, and 13 verdicts rest on a partial one (closest: element 99, 7.0 eV, 2.9 above B). The bracket of the law, finishing F04: the table solved under 1/r^0.5 (all 100 settle) and 1/r^1.5 (82 settle) puts its noble gases at 2, 12, 40, 100 (hydrogen's shells) and at 2, 14, 24, 34, 72; only 1/r gives the generalized Madelung rule's rows. §6.5, appendix F.8, figure 1c, claim etarows; the abstract and conclusion say the rows need the exponent 1/r."
  },
  {
   "date": "2026-11-10",
   "what": "The paper and its supplement split (PI): draft/paper.html is abstract to conclusion with the six main figures and the references; draft/supplement.html holds the notation, appendices A–K, figures A1–A2 and tables 1–2, with its own PDF. Every reference to an appendix, section, figure or table on either page links to its anchor (build.mjs linkRefs; 231 internal links, none broken), and the paper links its own work: the interactive table (and #128), kata.mino.mobi/orbital, the Hund explainer, the MIT-licensed mirror on tangled, the ledger and supplement S1. The main text stays at 6 two-column pages."
  },
  {
   "date": "2026-11-10",
   "what": "Astra's follow-up audit of the F01–F08 pass (reviewed at 9ac683ed). F01 independently reproduced (11/11 sample, Z = 82, S1). F02: Methods' last 'firm' sentence replaced (agreement, coverage and margins reported separately). F03: hf-collect.mjs defaulted unrecorded term provenance to 'full'; rank now records it on every candidate, and for older runs it is reconstructed from the configuration (blocks are skipped only for their size), agreeing with all 376 recorded cases; unreadable is 'unknown'. Every far record accounted for (182: 54 complete, 128 partial, raw energies matched exactly), no corrected candidate below its ground; claim frow gates it. 20 f-row verdicts now rest on a partial competitor (was 13: Z = 90's 4f9 5d1 6s2 among the new), no verdict changed. F04 was already fixed in f67debed9."
  },
  {
   "date": "2026-11-11",
   "what": "Pauli's exclusion derived, not declared (PI). The half of spin-statistics that makes half-integer spin a fermion needs only a Dirac equation and an energy bounded below, which the lens vacuum keeps; it gives up locality only for the force. atom/js/pauli.js and selftest 73: the 4 + 1 Dirac levels ±E, and commuting electrons in one negative-energy packet of width b = N², whose pinned 1/r repulsion is at most √π/(4bν²), have energy at most −Nmc² + √π/2: no ground state. §2.5 and F.6 rewritten, three references (Pauli 1940, Duck–Sudarshan 1998, Leinaas–Myrheim 1977), claim pauli; the site's force and motion chapters and /start/ step 6 changed to match. What stays declared: that the electron is the smallest spinor."
  },
  {
   "date": "2026-11-11",
   "what": "Prior-art search (PI): about 25 web searches plus arXiv, OpenAlex and Crossref (papers/p1/PRIOR-ART.md). No self-consistent 4D table and no Hund's-rule result in any dimension were found, so §4.1's and §5's claims stand; §4.1 now says Négadi and Kibler tabulate only two and three dimensions. Narrowed: the intro now says nobody has asked what 1/r does to many-electron atoms (Burgbacher et al. 1999 asked it for hydrogen), and F.6 credits the no-locality half of spin-statistics to Sudarshan 1968 and Lian 2012, claiming only that it survives the pinned law with an explicit bound (plus Brown–Ravenhall 1951 and the Coulomb-gauge remark). F.5 names each half of the lens vacuum's precedent (Rytova–Keldysh screening; εμ at the vacuum's value) and keeps 'not found' for the pair. §2.4 cites the shell theorem's uniqueness (Tse et al. 2026, Landkof 1972); §5.2 points to Morgan & Kutzelnigg 1993. Ten references added, each checked against Crossref or arXiv."
  },
  {
   "date": "2026-11-11",
   "what": "Title and byline (PI). Title, law first: 'Choosing a charge law for four dimensions: what 1/r does to atoms and the periodic table'. Byline Claude · Astra · minor mobius (last, PI), on the paper and the supplement, with an author-contributions statement before the supplementary-material note. The journal layout sets the title block across both columns and leaves the back matter out of the count: the main text is seven pages."
  },
  {
   "date": "2026-11-11",
   "what": "Version 1 frozen (PI): papers/p1/v1/, first from commit e3c84d792 and, after the figure 3 change below, again from fcebc6b16, by the new freeze.cjs. Its own copies of the paper and supplement, the figures, the stylesheet, the ledger and S1's data, both PDFs and the link card; relative links rewritten to the version's copies, the card tags and the PDFs' links pointing at v1, and a line under the byline naming the commit and linking the living draft. v1 is the copy to share; the draft keeps moving with stars and nuclei."
  },
  {
   "date": "2026-11-11",
   "what": "Figure 3 (PI): our own world's first ionizations from the same solver, Z = 1–36, drawn over the four-dimensional curve for scale, with its noble gases marked and the caption giving them against measured (23.4, 21.3, 15.2, 13.5 against 24.6, 21.6, 15.8, 14.0 eV); the full-d-shell points move to purple so ours can be orange. Version 1 had been frozen minutes before and not yet shared, so it was cut again from this draft rather than numbered 2."
  }
 ]
}