4
[2026-08-18] · src Nature
The BMW collaboration's hybrid lattice-QCD-plus-data calculation of the leading-order hadronic vacuum polarization contribution to the muon's anomalous magnetic moment, a_μ^HVP,LO = 715.1(3.4)×10⁻¹⁰ at 0.48% precision, pushes the full Standard Model prediction to a_μ = 11,659,205.2(3.6)×10⁻¹⁰ — within roughly 0.5σ of Fermilab's final 127-ppb measurement, a_μ = 11,659,207.05(1.47)×10⁻¹⁰ (arXiv:2606.17323). The twenty-five-year, ~4σ anomaly is gone. What isn't gone: the lattice number disagrees with the traditional data-driven prediction built from four decades of e⁺e⁻→hadrons cross-section measurements — the very calculation the anomaly was originally measured against.
The muon g-2 anomaly was read for a generation as a hint of undiscovered particles. It now looks like the theory side, not nature, was off — but the fix opens a fresh problem: lattice QCD and the e⁺e⁻ data-driven dispersive method now disagree with each other by several sigma on the same quantity, and only one can be right. Working out which of the collider cross-section measurements (KLOE, BaBar, CMD-3) carries the systematic error is the open question this leaves behind.
[2026-08-18] · src arXiv
A Comment on the widely cited NQR measurement of an "imaginary" charge density wave in the kagome metal CsV₃Sb₅ (responding to Suetsugu et al., Nature Physics, DOI 10.1038/s41567-026-03339-8) argues the reported signature of spontaneous loop currents and time-reversal symmetry breaking at ~120 K does not survive quantitative scrutiny: the NQR linewidth follows an ordinary Curie-Weiss law with no anomaly there, the asymmetric lineshapes taken as evidence of an internal field are reproduced by plain crystalline mosaic spread, and the claimed "emergent field" turns out to be an artifact of the fitting procedure's constraints.
If the comment holds up, one of the cleaner claimed signatures of exotic loop-current order in a kagome superconductor was crystal disorder read as physics — a caution in a subfield that has produced conflicting loop-current claims in CsV₃Sb₅ for several years running, and worth weighing against any future "time-reversal symmetry breaking" headline from this material.
[2026-08-18] · src arXiv / Phys. Rev. Lett.
Basson et al. (Carleton College) report the first bounds on all 43 previously unconstrained Lorentz-violating coefficients in the nonrelativistic limit of the Standard-Model Extension's minimal matter sector — the standard parameterization for direction- or velocity-dependent violations of special relativity — completing coverage of all 132 such coefficients and tightening 13 existing limits (Phys. Rev. Lett. 137, 081601, 2026). No violation was seen; every coefficient is consistent with zero.
This is bookkeeping rather than discovery: no anomaly, and the new limits come from reanalyzing published clock-comparison, spin-precession and matter-antimatter data rather than from new experiments. What it does is close off, for the first time completely, the space where a Lorentz-violating coefficient in ordinary matter could still be hiding unconstrained — relevant to anyone using these limits to bound a quantum-gravity or CPT-violating model, less so to anyone waiting for a violation to actually show up.
[2026-08-18] · src arXiv
A reanalysis of DESI DR2 baryon acoustic oscillation data combined with Pantheon+ supernovae and Planck CMB distance priors, fit to the w₀wₐCDM model, finds no statistically robust evidence for a time-varying dark-energy equation of state once redshift subsamples are compared self-consistently — the 2.8–4.2σ preference reported elsewhere is consistent with a statistical fluctuation in the current data.
The evolving-dark-energy hint from DESI has been one of cosmology's most discussed results since 2024, but which supernova compilation it's combined with changes the answer: this analysis, run with Pantheon+, finds nothing significant, while analyses built on the DES-SN5YR compilation still report a preference. That the result depends on which supernova sample you pick, rather than surviving across all of them, is itself the finding — ΛCDM is not dead yet.
1
[2026-08-18] · src build/orrery-test.el
Every gate in build/orrery-test.el is one function invoked the same way:
Makefile:27,31,35 run emacs -Q --batch -l orrery-test.el -f
orrery-verify-schema (and -golden, -docs) — --batch suppresses the
GUI, the minibuffer and any interactive prompt, then runs exactly the
function named after -f and exits. On failure, none of the three gates
signal a Lisp error. They all route through orrery-test--fail
(orrery-test.el:45-48):
(defun orrery-test--fail (format &rest args)
"Report FORMAT with ARGS on stderr and exit non-zero."
(message "FAIL %s" (apply #'format format args))
(kill-emacs 1))Two mechanisms, each doing a specific job. message in batch mode has
nowhere to echo to — there is no minibuffer — so Emacs writes it straight
to stderr instead; this is true of every message call in the gates,
success or failure alike (orrery-verify-schema's " schema ok" at line 64
goes to the same stream). kill-emacs is the separate half: called with an
integer argument, it sets that number as the process's exit status before
terminating. make (Makefile:27) runs each gate as an ordinary shell
command and fails the recipe the instant that command's exit status is
non-zero — make never looks at what a command printed, only at what it
returned.
Emacs batch mode already turns an uncaught error into a non-zero exit and a
printed backtrace on its own — orrery-test--fail is not the only way a
gate could fail loudly. Calling it explicitly is a choice: a one-line FAIL
... message naming exactly what broke, at an exit status the author
picked (1), instead of a raw Lisp backtrace from wherever the failure was
first noticed. orrery-verify-golden (orrery-test.el:112-121) is the case
that makes the difference concrete — its failure message
(orrery-test.el:106-110) reports the byte lengths of both documents being
compared, something a generic "wrong-type-argument" or "assertion failed"
backtrace would never surface, because the comparison itself did not fail;
the values did not match. Catching the mismatch explicitly and reporting
through orrery-test--fail is what makes a red CI job legible from its log
alone, without anyone opening test/golden/ to work out which of the two
gates even ran.
Exercise (verifiable): run, from the repo root:
emacs -Q --batch --eval '(message "hi")' 1>/tmp/out.txt 2>/tmp/err.txt
cat /tmp/out.txt # empty
cat /tmp/err.txt # "hi"Confirm message's output landed in err.txt, not out.txt. Then run
make schema against a checkout with no dist/ir.json (rm -f dist/ir.json
first) and check echo $? afterwards — it should be 1, matching the
literal argument orrery-test--fail passes to kill-emacs at line 58, not
whatever exit status an uncaught error would have produced instead.
2
[2026-08-18] · src twelve-keys/src/exercises/blues_walking_bass/exercise.py
Key: E major — four sharps, the next stop on the fourths cycle after 2026-08-15's B major. First blues-family exercise in the rotation; the last three piano sessions worked upper-structure triads, the Coltrane cycle and stride's LH leap, so today swaps all of those for a walking quarter-note bass line — the pianist standing in for the bass player, which is a longer and more sustained physical task than anything piano has drilled yet, and exactly what the longer block today is for.
5 minutes, hands separately, in E: LH alone playing the ascending 1-3-5-♭7 arpeggio over just the I7 (E7 — E, G♯, B, D) and IV7 (A7 — A, C♯, E, G) bars of the form, quarter notes, quarter=72, looped until the hand finds each shape without looking. Then RH alone on the shell voicings for those same two chords, placed on beats 2 and 4 only, to get the backbeat placement independent of the bass line before asking the hands to run together.
Hands together through the full 12-bar form in E (E7 – A7 – E7 – E7 – A7 – A7 – E7 – E7 – B7 – A7 – E7 – B7) at quarter=72, LH walking the 1-3-5-♭7 arpeggio ascending through each bar while RH drops the shell on beats 2 and
E major's sharps put several of the walking-line targets (G♯, C♯, D♯) in
territory the naturals-heavy keys don't ask for. Success criterion: two clean choruses in a row with the bass line landing on the new chord's root exactly on beat 1 of each new bar — the tell that it's still guessing is a bass note that arrives a fraction late or gets adjusted after landing.
Once the straight ascending line is solid, swap in a chromatic approach into each new root — arrive at the last beat of the bar a half-step below (or above) the coming root rather than jumping straight there — for the two turnaround bars (V7–IV7–I7–V7) at the end of the form. That's the exercise notes' own "ornament as your ear suggests" move, and it turns the walking line from a pattern into a line that's actually going somewhere.
Stop the moment the LH starts needing a glance down to find the next root, or the RH shells start landing early or late against the backbeat instead of locking to it — both are the hands losing the independence this exercise is built to train, and reps past that point rehearse the disconnect instead of the coordination.
[2026-08-18] · src —
hearOscar Peterson — Blues Etude (Live in Baalbek, Lebanon / 1972) (Solo) · Tidal · Spotify
Peterson's solo blues playing is the reference point for exactly today's move — a genuinely walking LH bass line under RH shells and fills, with no bass player to lean on. Listen for how steady the quarter-note pulse stays under the LH alone through the turnarounds, which is precisely what today's chromatic-approach variation is reaching for at a much slower tempo.