5
[2026-08-24] · src C. Y. Tsang et al. (STAR Collaboration); Science (2026); arXiv:2408.15441
Since the 1970s, a proton's baryon number - the quantum charge that makes matter stable rather than able to decay to nothing - has been assigned one-third each to its three valence quarks. An alternative, proposed in 1996 by Dmitri Kharzeev, is that it instead rides on a non-perturbative, Y-shaped "junction" where the gluons connecting the quarks meet. STAR now reports mid-rapidity measurements of baryon number over net electric charge (B/ΔQ) in isobar collisions (Ru+Ru versus Zr+Zr) and of net-proton yield along rapidity in photonuclear collisions. Both show a larger B/ΔQ ratio and a less rapidity-asymmetric net-proton yield than valence-quark-only models predict - consistent instead with baryon number being carried by something that is stopped, and re-emerges perpendicular to the beamline, more readily than a charged quark is.
If gluons rather than quarks carry a baryon's defining quantum number, that is a genuine revision to the textbook picture of nucleon structure, with consequences for how baryon number could have separated from antibaryon number in the early universe. Two caveats keep this at incremental rather than new: the analysis behind it has been on arXiv since August 2024, so Science's publication this month is the field catching up rather than a fresh measurement, and no experiment outside STAR has yet reproduced the signal.
[2026-08-24] · src Nesseris, Akrami, Starkman; Phys. Rev. D 114, L041305 (2026); arXiv:2503.22529
DESI's baryon acoustic oscillation measurements have been read, since 2024, as a mild-to-moderate preference for a dark energy equation of state that evolves with redshift, w(a) = w0 + wa(1-a) - the Chevallier-Polarski-Linder (CPL) form - over a static cosmological constant. This paper shows that CPL is being extrapolated into a regime where it is already known to fail, and that using it there rather than marginalizing over the higher-order terms CPL drops injects information the data do not actually contain. Once those terms are marginalized over, the authors find current data do not constrain either the present-day equation of state or its rate of change tightly enough to make a confident claim of evolution away from w = -1.
This does not rule out dynamical dark energy - it says the DESI hint's significance is more parametrization-dependent than the "crack in ΛCDM" framing it has gotten. A correction to how confidently a result should be read is exactly the kind of update this panel should not let slide past unremarked, even when - especially when - the original claim was the more exciting one to report.
[2026-08-24] · src Kuridze, Wöger, van Noort et al.; Nature 656, 595-601 (2026)
Using the 4-metre Daniel K. Inouye Solar Telescope, which resolves the photosphere down to roughly 19-20 km, this team finds spinning plasma vortices as small as 20 km across at the edges of magnetic flux concentrations, wherever oppositely-moving plasma streams meet - the same shear-driven Kelvin-Helmholtz instability that curls a breaking ocean wave or bands Jupiter's clouds. The instability had been predicted to occur on the Sun but was below the resolution of every previous instrument; this is the first observation showing it is ubiquitous rather than occasional.
These vortices are a plausible new channel for twisting magnetic field lines and feeding the small impulsive heating events ("nanoflares") proposed to explain why the corona runs roughly a million degrees hotter than the surface beneath it - a mismatch with no settled mechanism. This paper establishes that the vortices are common, not that they carry enough energy to matter for coronal heating; quantifying that share is the next measurement, not this one.
[2026-08-24] · src Caltech (Endres, Alicea groups) et al.; Nature (2026); arXiv:2601.16275
Conformal field theory predicts that at a continuous quantum phase transition, the excited-state energies above the ground state fall into fixed ratios set by the theory's universality class - a prediction used for decades to classify critical phenomena but never directly measured. This group trapped chains of up to 35 strontium atoms in optical tweezers, excited them into Rydberg states to get strong nearest-neighbour interactions, and tuned the chain through phase transitions in the Ising and tricritical Ising universality classes. A new technique they call many-body modulation spectroscopy resolves the finite-size excitation spectrum well enough to recover the universal energy ratios CFT predicts for both classes, matching them directly.
This is a general-purpose readout, not a one-off confirmation: the same modulation technique should let a tunable quantum simulator diagnose which universality class an unfamiliar phase transition belongs to, rather than requiring the class to be guessed and checked against theory first. That turns CFT from a framework physicists calculate with into one they can measure against directly.
[2026-08-24] · src Bandopadhyay, Amend, Coughlin, Nixon, Pasham, Wevers; Astrophysical Journal (2026)
In a repeating partial tidal disruption event, a star survives repeated close passages by a supermassive black hole, shedding mass each time without being fully destroyed. In some observed systems the resulting flares fade steadily from pass to pass; in others they stay roughly the same brightness, and nothing in the standard picture explained the split. This group's models show the difference comes down to the star's spin before its first encounter: a star already rotating rapidly - as expected if it reached its tight orbit via Hills breakup of a tidally locked binary
gets far less additional spin-up from each tidal passage than an
initially slow-rotating star would, which caps how much extra mass is stripped per pass and lets the peak fallback rate, and so the flare, decline pass over pass instead of holding steady.
It ties three previously separate observables - short orbital period, high stellar spin, and declining flare brightness - into one mechanism with a falsifiable prediction: systems with fading flares should show measurably higher stellar spin than systems with flat ones. That is checkable against the growing catalog of repeating partial disruption events without needing new instruments, only a spin measurement on the systems already found.
2
[2026-08-24] · src —
Today is piano, alone: trumpet and alto each had a turn in the last two days and neither wants to be stacked again so soon, while piano hasn't been touched since Thursday and is the one instrument that actually rewards sitting with one thing for a full block. The block is guide tones — the single-note line that traces the 3rd and 7th through a ii–V–I — which nothing in the last five piano sessions has covered: voicings, a reharm, stride, a walking bass and a scale, but never the bare melodic skeleton underneath all of them. Key is D♭ (Ebm7–Ab7–Dbmaj7), the one remote key this rotation hasn't visited yet and a genuinely awkward read for a single treble-clef line, unlike D, which is where strict cycling by fourths would otherwise land next.
Rootless shells for Ebm7–Ab7–Dbmaj7, both hands, quarter = 70, four times through the changes — not to warm up generally, but so the hands already know where D♭ lives before the line work asks them to hear it.
Play both guide tone lines from the exercise in D♭: b7-of-ii → 3-of-V → 7-of-I (D♭ → C, then held) and 3-of-ii → b7-of-V → 3-of-I (G♭ held, then → F). Hands separate first, then add the left-hand shells under the line. Success criterion: the half-step motion lands exactly on the downbeat of the chord change — not anticipated, not late — three times through the progression in a row without a stop or a correction.
Sing the canonical line before playing it, then play it as a phrase with a shape: a lean into the half-step, a settle on the arrival, rather than four equal eighth notes. The goal is a line that sounds like the first four notes of a solo, not a fingering pattern that happens to be correct.
Stop the moment you've played it clean three times in a row, even if that's ten minutes in. A fourth repetition chasing a cleaner feel usually just erodes the third one — banking the clean take is the win, not proving you can do it twice more.
[2026-08-24] · src —
hearMiles Davis — Tune Up (Cookin' with the Miles Davis Quintet) · Tidal · Spotify
"Tune Up" is built almost entirely out of consecutive ii–V–I's stepping down by whole tones, so it's the clearest possible demonstration of a guide tone line doing real work: Miles never runs the changes as an exercise, but the half-step motion under every turnaround is audibly the same shape as today's assignment, just played by a horn instead of a right hand.
3
[2026-08-24] · src Ninth Circuit (No. 24-2701)
A three-judge Ninth Circuit panel, in an opinion by Judge Consuelo Callahan joined by Judges Holly Thomas and Anthony Johnstone, vacated a district court order that had struck down "Example 4" of ATF's 2022 frame-and-receiver rule — the provision treating a partially completed handgun or long-gun receiver as a firearm only when sold with the jig or tools needed to finish it. California and the Giffords Law Center, which had sued arguing Example 4 was too permissive, lacked Article III standing on both causation and redressability: the panel found the state's evidence of harm mostly predated the rule (2016–2022) and that the share of unserialized "ghost guns" recovered in Los Angeles County had fallen, not risen, since Example 4 took effect. The case returns to the district court with instructions to dismiss it without prejudice.
The panel did not rule on whether Example 4 is a lawful reading of the Gun Control Act, whether the 2022 rule is constitutional, or whether a plaintiff with better-tailored evidence of injury could establish standing to bring the same challenge again. The dismissal is without prejudice for exactly that reason.
Example 4 remains in force, but on a standing ruling rather than a merits one — nothing about ATF's authority to write it, or a future administration's authority to rewrite it, has been settled. California or Giffords could refile with different evidence or a different plaintiff; nothing in this opinion forecloses that.
[2026-08-24] · src Federal Register, 2026-16313
DHS's interim final rule on mandatory electronic filing took effect August 11,
It does not itself make any form online-only; it authorizes USCIS to
require e-filing of a given benefit form once that form has been available for online submission for at least 180 consecutive days, provided USCIS posts notice on its website at least 60 days before enforcing the requirement for that specific form. Filers who cannot file electronically may request a hardship waiver at USCIS's discretion. Because it issued as an interim final rule rather than a proposed one, it is already in effect; the public comment period runs through October 13, 2026, and DHS can still amend the rule in response to comments received.
No form has yet been designated mandatory-electronic under this authority — that requires a separate 60-day notice per form, none of which has posted yet. The rule also does not resolve what counts as sufficient hardship for a waiver; that is left to case-by-case USCIS discretion.
Nothing changes for a filer today. The clock that matters is the 60-day notice: once USCIS designates a specific form, anyone filing that form after the notice period runs must file electronically or obtain a hardship waiver, and paper filings of that form submitted after the deadline can be rejected on that basis alone.
[2026-08-24] · src N.D. Tex. (Nos. 6:25-cv-00056, 6:26-cv-00277)
The seven-day stay on Judge Hendrix's August 5 ruling — covered here last week — expired without the government seeking relief from the Fifth Circuit, and the permanent injunction barring ATF and DOJ from enforcing NFA registration requirements against Silencer Shop Foundation and Jensen plaintiffs, and their current and future members and customers, took effect August 13. The government has filed no notice of appeal and no stay application as of this writing. Representative Andrew Clyde has said the administration confirmed to him that DOJ does not intend to appeal at all; that is a lawmaker's account of an informal confirmation, not a DOJ filing or statement, and the government's own filing deadline under the appellate rules runs to October 5, 2026.
The injunction still runs only to the named plaintiffs and their members and customers — not nationwide — and 18 U.S.C. § 922(b)(4)'s unlicensed-transfer restriction, interstate transport rules, and state-level NFA requirements are untouched, exactly as when the ruling issued.
As of August 13, Silencer Shop Foundation and Jensen's named plaintiffs and their current and future members and customers can transfer and make the covered items — suppressors, short-barreled rifles and shotguns — without NFA registration or approval, without a stay in place to unwind that if the government changes course before October 5. Everyone else remains bound by NFA registration as before; nothing in this ruling changes that unless and until a broader order does.
2
[2026-08-24] · src Typst 0.15.0 release notes
Typst 0.15.0 (June 15, 2026) lists it as a headline change: "HTML export now
supports equations out of the box via MathML." Until this release, the only
way to get an equation into HTML output was a show rule that called
html.frame to rasterize it to an embedded SVG — a picture of the equation,
not text a browser or a screen reader can do anything with. The release notes
say so directly: "If you've previously relied on show rules that use
html.frame to render equations to SVG, try MathML output. Switching to it
will improve the accessibility of your document (though rendering will be
somewhat less consistent across browsers)." HTML export as a whole is still
behind the html feature flag; only the underlying target function was
freed from it this release.
LaTeX has no native path from a document to HTML at all — getting math onto a
web page means leaving the pdfTeX/LuaTeX pipeline entirely for tex4ht,
latexml, or a pandoc conversion, each translating TeX math into MathML or
images with its own fidelity gaps, and each a second toolchain to maintain
alongside the one that makes the PDF. Typst closes that specific gap: the same
source that makes the PDF now also makes real, selectable, screen-reader-
readable MathML, from one compiler, with one syntax. What it does not yet
close is cross-browser consistency — the release notes admit that up front —
so the honest use case today is course notes or documentation going straight
to a web page, not yet a camera-ready web edition of a paper where rendering
has to match everywhere.
[2026-08-24] · src The TeX FAQ, "Footnotes in tables"; Typst footnote reference
Set a small table — three or four columns of numbers, one column header or one cell carrying a caveat worth footnoting, the kind of thing a physics table actually has ("corrected for background", "this run only") — in both LaTeX and Typst, with a footnote attached to that cell.
In plain LaTeX, put \footnote{...} inside a tabular (in a table float)
and compile it as-is first. The TeX FAQ names exactly what happens: "the
tabular environment ... traps footnotes, and they can't escape to the
bottom of the page," so you get the footnote mark in the cell and no note
text anywhere on the page. Then fix it with threeparttable (or the
tablefootnote package) and get the note text to actually appear. In
Typst, put a footnote() call in the same kind of cell inside a table and
compile it with no extra import, since nothing in Typst's footnote reference
says a table traps one the way LaTeX's does.
Success criterion: for each engine, produce one output where the footnote's
text, not just its mark, is visible on the same page as the table, and
write one sentence naming the mechanism that got it there — threeparttable's
table note escaping the float, or Typst's ordinary footnote working
unmodified. Under an hour: the table is four rows, and the LaTeX half of this
has a known, named fix rather than one you have to invent.
This is the shape a real physics table takes constantly — a measured value with a footnoted caveat about how it was corrected or under what conditions it holds — so the hour buys knowing, before it is 11pm before a submission deadline, which engine traps a table footnote silently and what the two-line fix is. Finding out mid-paper that a footnote mark has no text under it, after the table is already built and captioned, costs far more than an hour; finding out now costs one.