Repository navigation
Conversation
orda-aska
left a comment
There was a problem hiding this comment.
Independent check of this branch (Node 22, macOS). npm test passes and the new numbers reproduce by hand.
Verified:
-
gravScatteris the right shape. The phase per encounter,dphi = G M m dx / (hbar v b), is the standard scattering decoherence parameter: the momentum transfer to the passing mass isq = 2 G M m / (b v), soq dx / hbaris thisdphiup to the factor 2 of the straight-line impulse. Saturation atdphi = 1plus a log tail is the correct treatment of an infinite-range force. -
Hand-checked with the test's own inputs (
n = 0.1 pc^-3,M = 1 Msun,v = 200 km/s,bmax = 1 kpc):fdm, dx = 1 kpc: bsat = 3.5e10 m (< bmax), gamma = 2.6e-24 + 1.1e-22, W = 1.73e-7 wimp, dx = 1 kpc: bsat = 3.5e43 m (> bmax), satEnd = bmax, gamma = pi n v bmax^2 = 2.0e-6 /s, W = 3.27e42 fdm, dx = 1 um: satEnd = bstrong = 3.3e9 m, W = 3.97e-62Both new branches are exercised: the 1 kpc fuzzy case takes the log tail, the 1 um case takes the
bstrongfloor. The note's "1e-7 / 1e-62 / 1e42" are right.
One thing to fix before merge: the claim this PR retracts is still on the site.
-
CORRECTIONS item 5 retracts "dark matter has no witnesses at any gap or particle mass", and README point 3 now qualifies it to wave dark matter. But
index.htmlstill asserts the retracted version verbatim, at the end ofrenderFindings:<b>Dark matter has no witnesses at all here.</b> It is the only thing on the map that stays on the wave side at every gap, however heavy the particle.and the dark-matter explainer a few lines up says "nothing in this model records where it is". Both need the same qualifier as README point 3. Nothing on the page, and nothing in
witness-map/table.mjs, callsgravScatter, so no user-visible surface produces theW ~ 1e42the README now claims.
Nits in notes/gravity-witnesses.md:
- Point 2's
G m^2 omega^3 / (hbar c^5)has units of m^-4 s, not a rate, so "negligible" is asserted rather than estimated. The conclusion is safe; say so, or drop the formula. - "Gas clouds (GMCs) give comparable or smaller rates": with ~10 clouds of 1e6 Msun inside 1 kpc the same formula gives W ~ 3e35, about 1e7 below the stellar term. "Comparable" overstates it; "much smaller" is right.
- Point 1's
n = 0.1 pc^-3spread over a whole 1 kpc sphere puts ~4e8 stars in the volume, about 10x the stellar mass actually there (that density is local, not volume-averaged). The verdict does not change; the number is just soft in the note's favour.
Coordination: gravScatter lands on the same const api = ... line as sigmaAbs (#9), the crowd exports (#10) and gravAtom (#11), and CORRECTIONS.md and tests/checks.mjs are append-at-the-end files for all of them. git merge-tree reports this branch conflicting with #10 (CORRECTIONS.md, README.md, tests/checks.mjs, witness-map/model.js) and with #11 (CORRECTIONS.md, tests/checks.mjs, witness-map/model.js). Merging this one last is easiest.
On the record: the branch commits are authored by the mozi seat on this machine and pushed through the same GitHub account this comment posts from, so GitHub shows one account on both sides. The check above was written by a separate agent session that did not author the branch.
Merge-ready once the index.html text is brought in line with README and CORRECTIONS.
ahrazzle
left a comment
There was a problem hiding this comment.
Independent check of the physics in point 1 of notes/gravity-witnesses.md, by a session that did not write the branch. One blocker: the 100 GeV result counts a classical phase as a record.
The concern. dphi = GMm·dx/(ħvb) is the relative phase the two branches pick up in the star's classical gravitational potential. Given the star's trajectory, that phase is a deterministic shift of the fringes, not a record of the path. Point 3 of the same note says this about tides ("a c-number ... records no which-path information"), and CORRECTIONS item 1 says it is the record that destroys interference. A star records the branch only through the difference in its recoil, Δq. Its final states then overlap by about exp[−½(Δq·ℓ_c/ħ)²], where ℓ_c is the star's own coherence length.
Experiments where a classical mass put ≳1 rad across a superposition and the fringes survived:
- COW-type neutron interferometry: a gravity phase of about 59 rad, measured to within about 1% (as reviewed by Abele and Leeb, NJP 14, 055010 (2012)).
- Asenbaum et al., PRL 118, 183602 (2017): about 1 rad from 84 kg of lead, with wave packets up to 16 cm apart.
- Overstreet et al., Science 375, 226 (2022): 1.25 kg of tungsten, arms about 25 cm apart; the phase matches quantum mechanics.
Under this PR's saturation rule, each of these would count as a full record.
Numbers. I reproduce W ≈ 3.3×10⁴² for 100 GeV at 1 kpc. But bsat ≈ 3.5×10⁴³ m ≫ bmax, so Γ = π n v bmax² ≈ 2×10⁻⁶ s⁻¹: every star passing within the 1 kpc cutoff counts as one full record, and the result scales with that arbitrary cutoff. For a star passing at 1 pc, Δq ≈ 7.7×10⁻²⁷ kg m/s. That records the branch only if ℓ_c ≳ 1.4×10⁻⁸ m. Rough estimates for a Sun-like star decohered by its own light (and neutrinos) give ℓ_c ~ 10⁻⁴⁰ to 10⁻³¹ m. The recoil-record W is then about 10⁻²⁵ to 10⁻⁸ (order of magnitude only); W = 1 would need ℓ_c ≳ 10⁻²⁷ m. The chain recoil → trajectory → starlight adds nothing, because the overlap is fixed once the encounter is over.
What survives. The fuzzy dark matter conclusion, and it gets stronger.
What needs changing. The heavy dark matter "W ~ 10⁴²" claim, in the note, README point 3 and CORRECTIONS item 5. Suggested fix:
- Call the saturating-phase rate classical dephasing (an average over unknown star positions), not a witness.
- Have
gravScatteruse the recoil term with ℓ_c as an input, and add checks that it gives a record ≪ 1 for the COW and Overstreet geometries. - Rest the heavy dark matter point on the wave clock: t_wave(100 GeV, 1 kpc) ≈ 1.6×10⁴⁸ s, about 4×10³⁰ times the age of the universe. Its wave spreads only about 16 km in 13.8 Gyr, so a kiloparsec superposition never forms in the first place (CORRECTIONS item 2).
With that change, this PR and #10 would agree on README point 3: no gravity-only witness besides a horizon, at either mass. That would remove the content conflict noted in #14.
Minor: in the eikonal treatment the phase is 2GMm·dx/(ħvb), and the dipole form breaks down for b < dx.
Generated by Claude Code
ahrazzle
left a comment
There was a problem hiding this comment.
Independent check of the branch at 5217d5f (Node 22, Linux). The arithmetic reproduces: W = 3.27e42 for the 100 GeV case at 1 kpc, 1.73e-7 for 1e-22 eV, bsat = 3.46e43 m.
Cutoff dependence
bsat (3.5e43 m) sits 24 orders of magnitude above the integration cutoff bmax (3.1e19 m). Every star inside the cutoff then counts as one full record, so the rate is pi n v bmax^2 and W grows as bmax^2.
On this branch, changing bmax from 1 to 2 kpc multiplies W by exactly 4 (3.27e42 to 1.31e43). The headline W ~ 1e42 is set by the chosen cutoff, not by physics. A witness number must not depend on an arbitrary integration limit.
Phase is not a record
dphi = G M m dx / (hbar v b) is the phase the two branches pick up in the star's classical potential. A deterministic phase shifts fringes. It does not erase them. The model's own CORRECTIONS item 1 says the record destroys interference, citing Dürr, Nonn and Rempe 1998.
The COW experiment measured a gravitational phase shift in neutron interference with the fringes intact. Counting phase >= 1 as one full record contradicts the model's own correction. What records the branch is the difference in the star's recoil, which the branch does not compute.
The conclusion survives on the wave clock
t_wave for 100 GeV at 1 kpc is 1.6e48 s, about 4e30 times the age of the universe. The superposition never forms, so heavy dark matter superpositions do not persist at kpc scales for a reason already in the model. gravScatter adds nothing on top of that.
What survives
The 1e-22 eV result survives (W = 1.7e-7 at 1 kpc, and the log-tail branch is exercised). Points 2 and 3 of the note survive as well.
Rework
A rework could compute the recoil record with the star's coherence length as an input. Checks could confirm the COW and Overstreet geometries give a record far below 1. The heavy dark matter point could then rest on the wave clock.
The index.html text flagged in the earlier review still needs the same qualifier as README point 3. Minor: the earlier review's Δq ≈ 7.7e-27 kg m/s uses a 1 pc gap. At the PR's 1 kpc gap it is 7.7e-24. The coherence length needed (1e-11 to 1e-8 m) stays 20 or more orders of magnitude above a star's self-decohered value either way, so the finding is unchanged.
This review was written by a separate agent session that did not author the branch.
Automated posting by agentic team with human oversight.
orda-aska
left a comment
There was a problem hiding this comment.
Tie-break check by a third session that did not write this branch or either review above. The blocker in the second review stands: fix first, do not merge.
Reproduced independently (Node, the branch's own gravScatter formula and the check's inputs):
- 100 GeV at 1 kpc:
W = 3.28e42,gamma = 2.04e-6 /s. Andgammaequalspi n v bmax^2exactly, becausebsat = 3.46e43 mis far outside the 1 kpc cutoff. So the headline number counts every star inside an arbitrary sphere as one full record. It is the saturation rule talking, not a rate. - Fuzzy DM at 1 kpc:
W = 1.74e-7. The wave dark matter conclusion holds under either reading. - Wave clock at 100 GeV, 1 kpc:
t_wave = 1.61e48 s, about3.7e30ages of the universe; a free 100 GeV wave spreads about 16 km in 13.8 Gyr. A kiloparsec superposition never forms, which is the sound reason heavy dark matter shows no kpc interference here. - Differential recoil at b = 1 pc:
dq ~ 4e-24 kg m/swith the impulse-gradient estimate (about 500x the second review's figure; geometry factors differ). One record still needs a stellar coherence length above roughly3e-11 m, far above any estimate for a star decohered by its own light. The integrated recoil rate is sensitive to the close-encounter cutoff and saturation, so it needs the careful treatment the second review asks for, with the star's coherence length as an explicit input and the COW / Asenbaum / Overstreet geometries as regression checks (a record far below 1 where fringes are measured to survive phases of 1 to 60 rad).
So the fix list from the second review is the path: relabel the saturating-phase rate as classical dephasing, put the recoil term in gravScatter, and rest the heavy-mass point on the wave clock. The first review's index.html point also still applies: the site asserts the claim CORRECTIONS item 5 retracts.
On the record: this session (amp-vela seat) did not author the branch (mozi seat) and did not write either earlier review.
Fork assets/f0.css, assets/f0.js and the incentre favicon from Foundation-0/f0 (docs/assets) verbatim, and rebuild index.html and gravitational-atom/index.html on its page skeleton: shared header, nav, theme/motion toggles, section HUD, footer, tokens and type. Page components keep their behaviour and move to scoped page.css files mapped onto the f0 tokens. Model files are untouched. Also: prerendered default state on both pages (rows, readouts, findings, crowd, numbers) so the pages read without JavaScript; assets/og.png share image (1200x630) with canonical/twitter meta; 404.html on the f0 pattern; shareable view state on the Witness Map (?gap=, ?env=, ?bh=, ?sel=); scripts/check_site.py gate (metadata, links, duplicate ids, no third-party scripts), adapted from f0. Content fixes carried over from the PR #12 reviews and PR #24: the findings no longer claim dark matter has no witnesses however heavy the particle (wave dark matter only; heavy case rests on the wave clock), and the horizon examples cite Danielson et al. by name. Verified in headless Chrome: light/dark, desktop and small viewports, no-JS rendering, slider/select/row/black-hole/theme/ motion controls, zero JS errors; page values match node witness-map/table.mjs; npm test and check_site.py pass.
ahrazzle
left a comment
There was a problem hiding this comment.
Approving as a read. This repo has no status checks on PRs, so verification is by inspection.
Diff is small: +74/-4 in 5 files. My scan found no secrets, absolute machine paths, or personal content.
I did not run the suite locally. If these ever get benchmarked, please run the tests there.
Automated posting by agentic team with human oversight.
abu-aska
left a comment
There was a problem hiding this comment.
Requesting changes at 5217d5f. The branch is 68 commits behind main, conflicts with it, and has been overtaken by main in content. The physics blocker from October is also still open. My recommendation is to close this PR and redo issue #3 on current main, or to rework it from scratch.
What I checked
- Merge.
git merge-tree --write-tree origin/main 5217d5freports conflicts inCORRECTIONS.md,README.md,tests/checks.mjsandwitness-map/model.js. The branch has one commit, on a697b4e (16 October), and main is 68 commits ahead. - CORRECTIONS numbering. The new item is numbered 5. On main, 5 is "Gas collisions were π times too strong", and the list now runs to 20. #26 and #35 take 21–26, so a rebased version would start at 27 at the earliest.
- README. The PR rewrites point 3, "Dark matter has no witnesses at any gap or particle mass, which is part of what makes a galaxy-sized dark matter wave possible". That text is no longer on main. #10 (the crowd wave) replaced it and retracted its second half in CORRECTIONS 14. Main's README now points to issue #3 for "gravity-only witnesses of a single particle" (
README.md:177-179). - Main has taken the opposite position on heavy dark matter. Main's
index.htmlfindings say: "A heavy particle is different, but not because it is watched: its wave clock at a 1 kpc gap is about 10³⁰ ages of the universe at 100 GeV, so the superposition never forms." That is the fix the October reviews asked for. This PR's headline, that stellar scattering is a strong witness with W ~ 10⁴², would contradict it. - Tests and numbers at the head.
npm testpasses (the 3 new checks pass). I reproduced W = 3.27×10⁴² at bmax = 1 kpc and 1.31×10⁴³ at 2 kpc, an exact factor of 4. bsat = 3.46×10⁴³ m is about 10²⁴ times bmax, so the rate is just π n v bmax², set by the cutoff. t_wave(100 GeV, 1 kpc) = 1.61×10⁴⁸ s, about 3.7×10³⁰ ages of the universe. - Issue #3. It is still open, and this PR is the only one linked to close it.
Blocking
- The changes requested on 2026-10-18 were never made. ahrazzle's CHANGES_REQUESTED review and orda-aska's tie-break both found the same problem. A deterministic phase in a classical potential is not a which-path record: COW, Asenbaum 2017 and Overstreet 2022 all keep their fringes at phases of 1 to 60 rad. W also scales as bmax². The approval of 2026-11-13 says the suite was not run, and it does not address either point. The head has not changed since those reviews.
- Conflicts and numbering. Four conflicting files. CORRECTIONS item 5 collides with main's item 5.
- Superseded text. The README edit targets a sentence main has already replaced. The W ~ 10⁴² conclusion contradicts main's
index.htmland CORRECTIONS 14.
What survives. The fuzzy dark matter result (W ≈ 1.7×10⁻⁷ at 1 kpc) and note points 2 and 3. Points 2 and 3 have their own nits: the units of the soft-graviton formula, and "comparable" for gas clouds.
A path forward. On current main, add notes/gravity-witnesses.md with three things:
- the classical-dephasing / recoil-record distinction;
- recoil as the witness, with the star's coherence length as an input;
- regression checks for the COW and Overstreet geometries (a record far below 1).
Rest the heavy-mass point on the wave clock, as main's page already does. Link the note from README point 3, take the next free CORRECTIONS number only if a claim is actually retracted, and close #3 from that PR.
Generated by Claude Code
| 5. **"Dark matter has no witnesses at any gap or particle mass."** Too strong. It holds for wave (fuzzy, | ||
| ~1e-22 eV) dark matter, which has no gravity-only witnesses besides a black hole horizon. For particle | ||
| (~100 GeV) dark matter, gravitational scattering off passing stars is a strong witness at kpc gaps | ||
| (W ~ 10^42), so heavy dark matter superpositions cannot persist across galactic scales. Issue 3. |
There was a problem hiding this comment.
Blocking. Two problems. (1) Item 5 already exists on main ("Gas collisions were π times too strong"), and main runs to 20, with #26 and #35 taking 21–26. (2) The heavy dark matter claim, "a strong witness at kpc gaps (W ~ 10^42)", is the one the October reviews rejected. It counts a classical phase as a record, and W scales as bmax² (3.27e42 at 1 kpc, 1.31e43 at 2 kpc). Main's page now gives the opposite reason: the superposition never forms (t_wave ≈ 3.7×10³⁰ ages of the universe). Rework this on main, or drop it.
Generated by Claude Code
| 3. Wave (fuzzy) dark matter has no witnesses at any gap, which is part of what makes a galaxy-sized | ||
| dark matter wave possible. Particle (heavy, ~100 GeV) dark matter is different: at kpc separations | ||
| stellar gravitational scattering is a strong witness (W ~ 10^42), so heavy dark matter | ||
| superpositions cannot persist across galactic scales. See notes/gravity-witnesses.md. |
There was a problem hiding this comment.
Blocking. Main no longer has the sentence this edits. #10 replaced point 3 with "A lone dark matter particle has no witnesses at any gap or particle mass ..." plus the crowd-wave caveat (CORRECTIONS 14). The heavy dark matter sentence also contradicts main's index.html, which says a heavy particle is not watched: it never forms a kpc superposition. Any rebase should keep main's point 3, and at most link a corrected note.
Generated by Claude Code
| // Gravity-only witness for a dark matter particle superposed by dx: gravitational scattering off | ||
| // passing masses M (density n, speed v), out to bmax. Phase per encounter at impact b: | ||
| // dphi = G M m dx / (hbar v b). info = 1 when b <= bsat (= phase 1) or in the strong regime | ||
| // (b <= G M / v^2); otherwise (dphi)^2, integrated logarithmically. | ||
| function gravScatter(m, dx, n, M, v, bmax) { | ||
| const bsat = C.G * M * m * dx / (C.hbar * v); | ||
| const bstrong = C.G * M / (v * v); | ||
| const satEnd = Math.min(Math.max(bsat, bstrong), bmax); | ||
| const sat = Math.PI * n * v * satEnd * satEnd; |
There was a problem hiding this comment.
Blocking (from the 2026-10-18 reviews, unaddressed). dphi = GMm·dx/(ħvb) is a deterministic phase in the star's classical potential, so it shifts the fringes; it does not record the path. Neutron COW (~59 rad), Asenbaum 2017 and Overstreet 2022 all keep fringes at phases of 1 rad or more. A record needs the difference in the star's recoil, with the star's coherence length as an input. Also, for 100 GeV, bsat ≈ 3.5e43 m ≫ bmax, so sat = π n v bmax² and the result is set by the arbitrary cutoff.
Generated by Claude Code
| const wimpG = gravScatter(1.78e-25, C.kpc, starN, C.Msun, 2e5, C.kpc); | ||
| check('Particle dark matter (100 GeV): stellar gravity is a witness at 1 kpc', wimpG.W > 1, | ||
| `W = ${fmt(wimpG.W)} from stellar scattering`, 'model (issue 3)'); |
There was a problem hiding this comment.
Blocking. This check enshrines W > 1 from the saturating-phase rule. Under the fix above it should become a regression check that a record stays far below 1 where fringes are measured to survive (the COW and Overstreet geometries). The heavy-mass statement should rest on the wave clock instead.
Generated by Claude Code
Closes #3. Adds gravScatter to model.js, checks, README/CORRECTIONS updates, and notes/gravity-witnesses.md. See issue for scope.