engraved <2026-10-25> · rebuilt whenever the sheet it copies is
Science & Physics
Mechanism and result, at the depth a physics major reads. Not press releases.
Science & Physics
- 01
Rocq-checked finite-key BB84 proof (87 pages) exposes unstated side conditions and a missing proof branch in the published argument
The authors give an admit-free, end-to-end machine-checked finite-key security proof for BB84 in the Rocq proof assistant, 87 pages including a 71-page appendix. In formalising the standard paper argument they identify side conditions it leaves unstated and, in one place, a proof branch that was missing. The claim is about the proof, not the protocol: BB84 is not shown to be insecure.
WhyQKD security statements rest on long analytic arguments that are rarely independently rechecked, so a mechanised proof that surfaces gaps is the kind of scrutiny the field has lacked. It does not change what BB84 hardware does; implementation side channels are outside its scope. Whether the missing branch changes any published key-rate bound is not in the abstract, and I did not read the paper.
- 02
Dong et al.: peculiar-velocity corrections move the local-ladder H0 from 74.30 to 71.37 km/s/Mpc and cut the "intercept tension" from 6.8σ to 3.9σ
Using Pantheon+ in flat ΛCDM, the authors compare the intercept of the local distance ladder against the value implied by the cosmological model. Applying peculiar-velocity corrections moves the forward-fit H0 from 74.30 (+1.18/−1.20) to 71.37 (+1.23/−1.18) km/s/Mpc, and the intercept discrepancy falls from 6.8σ to 3.9σ in the forward direction and from 7.3σ to 4.3σ in the backward direction. A residual tension of roughly 4σ remains in both.
WhyIf the correction is right, a large part of the quoted local-versus-CMB gap comes from how nearby-supernova velocities are treated rather than from the Cepheid or TRGB calibration, which would make the tension partly a systematics question about the Hubble-flow sample. The remaining ~4σ is still not small, and the result is for one dataset and one cosmology, so it does not replace the calibrator-based determinations. I read the abstract only and have not checked how the velocity model and its covariance are handled.
- 03
Chlorine in seawater tags antineutrinos with an ~8.6 MeV capture cascade, projected to extend 5σ supernova reach from ~50 kpc to over 100 kpc
Neutron capture on chlorine in seawater releases an ~8.6 MeV gamma cascade, against 2.2 MeV for capture on hydrogen, on a shorter timescale. That gives inverse beta decay on electron antineutrinos a delayed-coincidence tag in sparse-instrumented water Cherenkov arrays such as TRIDENT. A detector simulation with measured site water properties projects the 5σ discovery reach for core-collapse supernovae growing from roughly 50 kpc to over 100 kpc. A salted-water test at the China Jinping Underground Laboratory saw correlated neutron-capture signals under suppressed muon background.
WhyIt would turn existing sparse MeV-blind arrays into supernova and reactor antineutrino detectors without new hardware. The reach figure is a simulation projection, and the lab test validates the capture-signal principle rather than a deep-ocean detection, so the real efficiency against ocean optical backgrounds is open. I read only the abstract.
- 04
Chen, Gray, Du, Chan: at 1/8 doping the Hubbard model has uniform d-wave, striped superconducting and striped non-superconducting states within ~1 milli-t
Neural tensor network and neural backflow variational states, run on GPUs on lattices up to 1024 sites, find at 1/8 hole doping a near-degeneracy of three states in the 2D Hubbard model, both with and without next-nearest-neighbour hopping: a uniform d-wave superconductor, a superconducting stripe state and a non-superconducting stripe state. They are separated by only about a milli-t. The authors conclude the model does not favour one dominant order in the cuprate-relevant regime; stripes and superconductivity compete strongly.
WhyWhether the plain Hubbard model superconducts at realistic doping has been disputed because each numerical method carries its own bias, and an energy window of order 10⁻³ t is below the systematic error most methods claim, so this narrows the question to which ansatz is more accurate rather than settling it. It argues against reading any single-method ground state as the answer, and it means small perturbations (t′, lattice shape, strain) can plausibly flip the order. The abstract gives no energies with error bars, and I did not read the paper, so how well converged the milli-t splitting is remains unchecked here.
- 05
Neutrino-only low-energy weak mixing angle: sin²θ_W = 0.233 (+0.007, −0.006) from CEνNS plus ν–e scattering
A global reanalysis combining coherent elastic neutrino–nucleus scattering (CEνNS) with neutrino–electron scattering data gives a neutrino-only determination of the low-energy weak mixing angle, sin²θ_W = 0.233 (+0.007, −0.006), a fractional uncertainty near 3%. The authors compare it with the cesium atomic-parity-violation value, check the mutual consistency of the individual neutrino datasets, and present it as an independent Standard Model test at low momentum transfer.
WhyIt is consistent with the Standard Model running value near 0.238 at these scales, so it excludes nothing; its use is as an independent check on the atomic-parity-violation determination, which currently sets the precision at low Q. At roughly 3% it is well short of the APV precision, so it constrains new physics only if CEνNS statistics improve. It is a preprint reanalysis of existing data, and I read only the abstract.
- 06
Belle and Belle II observe Λc⁺→ΛK⁺η at >10σ, B = (6.20 ± 0.25 ± 0.31 ± 0.30)×10⁻⁴, above SU(3) predictions
Using 980 fb⁻¹ of Belle and 570 fb⁻¹ of Belle II data, the collaborations observe the singly Cabibbo-suppressed decay Λc⁺→ΛK⁺η for the first time, at more than 10σ. Normalised to the Cabibbo-favoured Λc⁺→Λπ⁺π⁰, the ratio is (8.73 ± 0.35 ± 0.43)×10⁻³, giving an absolute B = (6.20 ± 0.25 ± 0.31 ± 0.30)×10⁻⁴ (the third error is the normalisation mode). Λc⁺→ΛK⁺π⁰ is remeasured at a ratio of (1.82 ± 0.07 ± 0.07)%, B = (1.29 ± 0.05 ± 0.05 ± 0.06)×10⁻³, a significant precision gain over the previous value. The ΛK⁺η rate is reported as significantly larger than SU(3) flavour-symmetry predictions.
WhyCharm-baryon decays are where flavour-symmetry and factorisation-based models are tested, and a mode that SU(3) underpredicts is a constraint that theory fits must now absorb, probably through SU(3) breaking or final-state rescattering. The size of the discrepancy is not in the abstract and I did not read the paper, so how large the tension is remains unchecked here. The result does not touch new physics; it is a calibration point for hadronic models.
- 07
IceCube TeV νμ CC cross section below 20% from 150 GeV to 40 TeV, SM-consistent
Using eight years of IceCube through-going muon-neutrino data, the authors measure the νμ charged-current neutrino-nucleon cross section from 150 GeV to 40 TeV with uncertainties below 20% throughout, against roughly 20–40% for earlier collider, atmospheric and astrophysical measurements. The result agrees with Standard Model predictions. For the first time above 1 TeV there are mild hints that ν and ν̄ cross sections can be separated.
WhyIt closes the gap between accelerator measurements and the higher-energy astrophysical regime, where deviations in nucleon structure or new physics would show first. Agreement with the SM at under 20% rules out only large deviations. The ν/ν̄ separation is a hint, not a result. The abstract gives no central values; I read only the abstract.
- 08
Free-neutron weak magnetism "observed" at 5.1σ: g_M = 4.47(87) vs CVC 4.71
A global reanalysis of existing data extracts the weak-magnetism form factor of the free neutron as g_M = 4.47(87), against the conserved-vector-current (CVC) prediction of 4.71, and excludes g_M = 0 at 5.1σ. The β asymmetry alone cannot separate g_M from the axial-to-vector ratio λ; adding the neutron lifetime and the superallowed 0⁺→0⁺ 𝓕t value breaks the degeneracy. The significance stays within 5.0–6.1σ across variations of the neutron dataset, despite its known internal tensions.
WhyThis confirms a 1950s CVC prediction in the simplest β-decay system, and it is a consistency check on the inputs used for V_ud and λ rather than a new measurement. The 19% uncertainty on g_M is large, the claim is a preprint with no independent scrutiny yet, and the authors themselves say a full energy-dependence analysis would improve g_M by only a few percent now. Treat "first observation" as a statement about a combined fit, not a dedicated experiment. I read only the abstract.
Learning, Lisp & Q&A
Answers drained from ask.org, the next lesson from curriculum.org, and whatever
spaced repetition is due.
Learning, Lisp & Q&A
- 01
Answer: Day trip itinerary — kayaking Lake Jocassee tomorrow
Tomorrow, Monday 2026-08-31, is a good day for it: sunny, ~89°F/71°F, calm wind, 0% chance of rain, sunrise 7:03am and sunset 8:01pm (NWS Greenville-Spartanburg; forecast) — no storm risk and over 12 hours of daylight, so timing is about your own morning, not the weather. Surface water is in the 70s-80s°F, but Jocassee is a deep, stratified reservoir, so it's noticeably colder a few feet down; sun is strong at elevation (UV 8-10 typical), so sunscreen and a hat matter more than the air temperature suggests.
10:00 wake; 10:00–10:30 breakfast and fill the cooler; 10:30–11:00 load the kayaks and gear (see packing list below) — that's a generous 30 minutes for two boats, tighter if you're doing it solo; depart by 11:00. Drive time from wherever you're starting isn't something I can fix without a starting point, so add your own drive time to Devils Fork State Park from the map link below and that sets your actual arrival; with an 11:00 departure and a lake that stays calm and sunny past 8pm, you have slack even for a two-hour-plus drive.
Devils Fork State Park (Salem, SC) is the only public access to Lake Jocassee — the rest of the shoreline is Duke Energy watershed and Jocassee Gorges wilderness with no public road access. Use the park's dedicated hand-launch lane (one lane, small paved lot) rather than the main trailer boat ramp, which four ramps' worth of pontoons and power boats keep busy on summer days; there's also a calmer carry-in beach on the Bear Cove Trail if the hand-launch lot is full. Day-use admission is $8/adult, required to launch anything, park open ~7am–9pm. If you didn't bring boats, Eclectic Sun, the on-site concessionaire, rents kayaks, canoes, and SUPs.
worth paddling to if you have the legs for it: Laurel Fork Falls (an 80-ft, three-tiered waterfall on the NE arm, boat-access only) or Wright Creek Falls on the west side; the Whitewater River arm gets you close enough to see Whitewater Falls' final drop into the lake.
Jocassee Valley Brewing Company in Salem (~5 min from the park) for the least effort — casual brewery pub food, dog- friendly patio. For something heartier, The Porkette in Pickens (~25–30 min) does Texas-style smoked BBQ. For a sit-down table, Dakota Grill in Walhalla or Tony's Restaurant in Pickens (both ~25 min) are both established, non-rushed options roughly on the way back toward the highway.
PFDs and a spare paddle; the $8/adult park fee in cash or card; tie-downs/straps if the kayaks ride on a roof rack; a dry bag for phone, keys, and wallet (a waterproof phone case if you're bringing it on the water); sunscreen and a hat — the UV exposure on open water is higher than it looks; sunglasses with a retainer strap; drinking water and snacks/lunch, since there's no food on the lake itself; a change of clothes and shoes for after; water shoes or sandals with a heel strap for the launch; a small first-aid kit; a whistle (SC requires one aboard a paddle craft); trash bag to pack out what you bring in.
The one thing I couldn't fix without more information is your starting point, so the schedule above is anchored to your wake-up time rather than a clock time for departure — plug your own address into the Devils Fork map link to get a drive time and back into a firm "leave by."
WhyThe binding constraint on a trip like this is daylight and weather, not effort, and tomorrow has plenty of both — the plan above is really just sequencing (load, drive, launch, paddle, eat) against a day that doesn't force any hard cutoffs. The one piece worth rechecking close to departure is the forecast itself: this is a next-day NWS point forecast, reliable but not locked in, so a five-minute check of weather.gov/gsp before you load the car is cheap insurance against a plan built on today's read of tomorrow.
- 02
Answer: Provide better graphs always
No image or chart embedding exists anywhere in Orrery's pipeline today — not in the fragment schema
orrery-model.elreads, not in either renderer — so the ASCII diagram in this afternoon's convergence answer was not a shortcut taken under time pressure, it was the only picture a fragment can carry: a#+begin_exampleblock is the entire facility for drawing something, and "always" better than that needs a capability that has not been built, not a run trying harder.The reason is the one
orrery-model.elalready spells out for equations, and it applies here without change: the page has to render from afile://URL with script blocked and fetch nothing at runtime — no CDN, no font, no script — because that is what lets it read on a phone with no signal. A client-side chart library fails that test exactly the way KaTeX did when it was costed for mathematics and rejected at 279KB of script for content most items never carry; a plotting call inside a source block fails it worse, since it would have to execute to draw anything. The route that survived that costing for equations was MathML built at compile time from the LaTeX source; the equivalent for a graph would be an SVG generated at build time from data the fragment carries — a new block kind in the IR, a renderer for it in bothorrery-render-html.elandorrery-render-org.el, andmake goldencoverage for it. That is real work, not a setting to flip, and nobody has done it yet. (Read as asking whether the repo can do better, not as a request to redraw this afternoon's diagram — there is nothing to redraw it into without that change; abuildorsteerentry naming the SVG block kind is how it would get one.)WhyEvery future answer that reaches for a diagram will keep landing as ASCII until that block kind exists — this is already the second one today, after the convergence proof a few hours earlier — and no amount of care in how a run writes the answer raises that ceiling, because the ceiling is the fragment format and the renderers, not the prose. What moves it is a
buildentry naming the SVG block kind, not a future learning run being asked to try harder. - 03
Answer: Definition of convergence
A sequence converges to a limit L when every tolerance, however small, is eventually respected forever after: for every ε > 0 there is an N past which every later term sits within ε of L, typeset in the usual quantifier order below (the request for a graph gets an honest ASCII one rather than a plotted line, for the reason under Why it matters).
equationa_n | L+e |- - - - - - - - - - - - - - - - - - - - - - - - - - | * | * * L |- - - - - - - - -*- -*- -*--*--*--*--*--*--*--*--*--> n | * L-e |- - - - - - - - - - - - - - - - - - - - - - - - - - | +------------------+--------------------------------> 0 NBefore N the terms (*) wander outside the band [L-e, L+e]; past N every one of them stays inside it, which is exactly what the quantifiers above say and nothing more. "Convergence" unqualified most often means this — a sequence's limit — rather than a series' or a function's, and both of those reduce to the same shape one level up: a series converges when its sequence of partial sums does, and a function's limit swaps N for a δ bounding how close x must be to the point in question.
WhyThe order of the quantifiers is the entire content of the definition — N is chosen after ε and is allowed to depend on it, never the reverse — and picking one N that is meant to work for every ε at once is the single most common mistake made with this definition, not a subtlety at its edges. It is also the exact difference, one level up, between pointwise and uniform convergence: swap "∃N ∀n>N" for "∀n>N ∃N" in a sequence *of functions* and the definition changes meaning while every symbol in it stays the same. On the graph, that mistake looks like choosing N by eye from one run of the sequence and assuming it still bounds the wandering once ε shrinks — which it need not, and checking that it does is the actual content of a convergence proof, not the epsilon-chasing that gets the blame.
- 04
Answer: Embedded music links need my login?
No to both halves of the question. The player asks nothing of your login — it is a plain sandboxed iframe (
allow"encrypted-media"= is the only permission it is given) that loads and plays for any reader, so nobody needs to be signed into anything for it to work; the only account in play is Spotify's, whose cookies land in your browser the moment the frame loads, login or not. And no, it cannot be added to your past listens retroactively:data/feed/is append-only, a fragment already written is never edited, and the player only appears on a listening item whose own fragment carries a verified:SPOTIFY:(or:TIDAL:) property — so an item that shipped without one stays a plain search link forever, no matter what changes later. (The heading asks about login, the body about backfilling old listens — both get answered here since neither reading was the wrong one.)That property only gets written when the music panel looked the recording up and confirmed the URL, a rule
model/briefs/music.orgadded recently. Today's listen — Louis Armstrong's "West End Blues" — is the first to carry one and the first to get a player; the three before it that named an actual recording, Miles Davis's "Tune Up" (2026-08-24), Johnny Hodges' "Passion Flower" (2026-08-23) and the Wynton Marsalis/Hummel concerto (2026-08-22), were written before the panel verified links and have neither property, so they render as Tidal/Spotify search links only.WhyThis is a going-forward feature, not a backfillable one. Nothing will ever walk the back-catalog and decorate old listens with players, so a past item worth linking needs a new item naming it again — a fresh listen, or a steer asking for one — not a fix to the fragment that already ran. It also sets the honest expectation for the rest of the feed: five of the thirteen listening items so far name no recording at all, and even among the ones that do, a player is the exception rather than the rule until enough days pass under the verification rule for it to catch up.
- 05
Answer: How to print your own stickers, with links
Printable, adhesive-backed vinyl sheets fed through an ordinary printer are the whole trick — two guides, Sticker It's and StickerYou's, walk the same four steps: design at 300 dpi and export as a transparent PNG, print on the sheet's printable (non-adhesive) side with a pigment-ink inkjet for water resistance, press on a clear self-adhesive laminate to seal it, then cut. Waterproof vinyl sheets made for inkjet or laser (check which — they are not interchangeable) are the one consumable to buy.
Cutting is the only step where spending money changes anything, and it scales with volume: scissors or a craft knife cost nothing and are fine for a handful of stickers; a cutting machine such as the Cricut Explore 4 or the Brother ScanNCut SDX125E reads a cut line out of the file and cuts it precisely, worth it past a few dozen; an all-in-one like the Liene PixCut S1 prints and die-cuts in one pass instead of two. [[https://craftcascades.com/best-sticker-die-cutting-machine/][Craft Cascades' current comparison]] covers prices and specifics.
With no printer or cutter worth setting up for one batch, mail-order services (Sticker Mule, Vistaprint, StickerYou) skip the equipment question entirely — upload a design, get stickers back.
WhyA bounded practical question with no reach past itself — nothing else in this dashboard depends on the answer, so there is nothing more to draw out of it.
- 06
Answer: Alto sax method book — a fourth research answer to a request that isn't one
Nothing has changed since data/feed/2026-08-20/learning.org, data/feed/2026-08-22/learning.org and data/feed/2026-08-23/learning.org: Voxman, Selected Studies is still the only one of the four candidates with the key-by-key technique-plus-etude shape trumpet already has under
:METHOD:inmodel/briefs/music.org, and this run has nothing new to add to that comparison. This is the fourth entry in the thread asking the same thing.The blocker was never missing information — it is that setting
:METHOD:is a write tomodel/, andmodel/has exactly one writer per CLAUDE.md, which is not an agent. The only path forward is asteerentry naming the choice:orrery steer "alto sax method book is Voxman, Selected Studies"(or pick a different one of the four and say so), which comes back as a pull request againstmodel/briefs/music.org, the same way trumpet's Arban line got there. Filing a fifth:ask:entry here will produce this same answer a fifth time.WhyNothing new to state — the first three answers in this thread already said where this bites (alto stuck on generic fundamentals) and what unblocks it (a steer). Repeating that note a fourth time is exactly the padding this brief warns against.
Music Practice
Trumpet, alto sax, piano. Today's assignment and which horn is going cold.
Music Practice
- 01
Listening: Clifford Brown and Max Roach, "Daahoud"
hearClifford Brown and Max Roach — Daahoud · Tidal · Spotify
Brown's own tune, a brisk bebop line. Listen to how every note in a fast run has the same weight and the same clean edge, so that tonguing never interrupts the sound. That is the evenness the First Studies are meant to build, at a tempo well above today's.
- 02
Trumpet: Arban first studies for an even tone, then one melody to sing
Trumpet takes today: three days since it was last assigned, the longest gap of the three, and it wants a short session. Last time was Syncopation, so today goes back to the First Studies in Arban, whose job is tone and evenness, paired with a melody from the Art of Phrasing. The aim is notes that are the same size and shape from one to the next, in the middle register.
Warmup
5 minutes: long tones on written C (third space) down by half steps to written F# below the staff, mp, 8 beats each, easy attack and a breath between. Then 2 minutes of lip slurs on C-G-C at mf, slowly.
Technical target
Choose one First Studies exercise in a comfortable key and play it slurred, then tongued, at quarter = 60, 72, then 84. Success criterion: three consecutive passes at each tempo where every note has the same weight and the tongued pass sounds as smooth as the slurred one.
Musical target
Play one melody from the Art of Phrasing, a different number than on 10-22. Mark the peak of each phrase, play toward it, and let the end of the phrase settle without going flat or sharp. It should sound like one breath per phrase.
Stop condition
About 25 minutes, and stop sooner if the tone spreads or the lips buzz rather than ring. A short session that ends while the sound is still good beats a long one that ends ragged; do not add a second block.
- 03
- 04
Piano: minor ii-V-i in F# minor
Piano takes today: four days since it was last assigned, the longest gap of the three, and it is the one instrument that rewards a long block while trumpet and alto rest. The family moves from line building to progressions, with minor_ii_v_i https://brona90.github.io/twelve-keys/minor_ii_v_i.html in F# minor, a key that does not read easily. About 50 minutes.
Warmup
8 minutes: the F# harmonic minor scale, hands together, two octaves at quarter = 72, mp, then the same in F# melodic minor ascending. Follow with G#half-diminished 7, C#7 and F#m7 as plain root-position chords, three rounds, listening for even touch between the hands.
Technical target
The exercise's ii-V-i in F# minor, hands together, at quarter = 60, 76, then 92. Success criterion: three consecutive clean passes at each tempo with no hesitation at the C#7 to F#m change and the voicing steady across the bar line.
Musical target
Play the progression as a four-bar phrase, not four chords: let the C#7 lean into the F#m, with the tension of the dominant easing in the last bar. Aim for a resolution that sounds arrived at rather than stopped on.
Stop condition
About 50 minutes. Stop earlier if the hands tighten or wrong notes start to repeat; two relaxed clean passes beat a tense tenth.
- 05
Listening: Cannonball Adderley, "Autumn Leaves"
hearCannonball Adderley — Autumn Leaves · Tidal · Spotify
This is the alto sound to carry into today's long tones: warm, full, and even from the bottom of the horn to the top. Listen to his solo for how the tone stays round when he moves by step and by third, and for how he shapes a phrase with a peak and a settle. That is the same thing the D major musical target asks for, played by a master.
Politics & Policy
Primary sources first: bill text, rulings, filings, published data.
Politics & Policy
- 01
Supreme Court keeps FCC lowest-unit-charge guidance in force while the Fourth Circuit case is pending
The FCC Media Bureau's March 30, 2026 Public Notice reminded broadcasters that the "lowest unit charge" (LUC) rate for political ads covers two things beyond a candidate's own committee: authorised committees engaged in joint fundraising with a legally qualified federal candidate, and ads that are coordinated expenditures between a party and a candidate. Four Democratic candidates challenged it. On August 25 a divided Fourth Circuit panel (2-1) set the notice aside, reading Section 315 to limit LUC to candidates and treating joint fundraising committees with non-candidate members as ineligible. On September 4, the day the LUC window for the November election opened, the Supreme Court, in an unsigned per curiam order over Justice Jackson's dissent, stayed that judgment, saying the Fourth Circuit likely lacked jurisdiction because the candidates' application for review was still pending before the full Commission. Secondary sources say that reasoning splits with every other circuit to consider the question.
The dispute then moved again. The Fourth Circuit's October 7 judgment (No. 26-2230) required the FCC to decide the pending application by noon on October
The Solicitor General applied for relief (26A476), and on October 8 Chief
Justice Roberts administratively stayed that judgment pending further order, with a response due October 10. The docket lists an entry titled "26A476 FCC v Brown (10/14/2026)" on the Court's opinions-relating-to-orders page. Its contents were not readable here, so how the Court disposed of the application is not recorded in this item and should be checked on the docket.
Facts versus interpretation: that the notice was issued, that the Fourth Circuit set it aside, and that the Court stayed that ruling are documentary. Whether LUC properly extends to party-coordinated and joint-fundraising ads is a statutory question the Supreme Court has not reached.
What it does not decide
Neither stay decides the meaning of Section 315. The September order is a preliminary view on jurisdiction only, and the Fourth Circuit's merits reading is stayed, not reversed. The October 8 order is administrative and says nothing on the merits of the application.
WhyThrough at least the November election, the March 30 Public Notice's reading governs: stations must offer LUC to the covered committees and coordinated ads. Per Wiley, a certiorari petition is due in late November, so the stay likely outlasts the election. A broadcaster or committee has no different obligation to plan for unless the Court's October 14 action changed the position, which this item cannot confirm.
- 02
Supreme Court grants review of detention of noncitizens arrested inside the U.S. without bond hearings
According to SCOTUSblog, the Supreme Court on October 1 agreed to hear a challenge to the administration's policy of detaining noncitizens apprehended inside the United States, as opposed to at the border, without a bond hearing. Argument is expected early next year. SCOTUSblog's report, as retrieved, does not give the docket number, the circuit decision under review, or the exact question presented, so those need the order list.
What it does not decide
A grant of review decides nothing on the merits. It does not say the policy is lawful or unlawful. Whether it changes anything for people now detained depends on whether lower-court orders requiring bond hearings remain in effect pending decision. This record does not say.
WhyNothing changes on the grant alone. Rules for individual detainees remain whatever the lower courts have ordered, and the Court's merits ruling cannot come before the early-2027 argument.
- 03
DHS proposes a $70,000 per-student fee on schools for Optional Practical Training recommendations
DHS has proposed a rule under which a school would pay $70,000 the first time it recommends an F-1 student for any Optional Practical Training (OPT), and $30,000 for each later recommendation of that student. Fees would go to the U.S. Treasury. The proposal lets schools decide whether to pass the cost to students, to all students, or to employers. DHS says in the notice that without the fees the program "may shut down the program entirely." If finalised, the rule would apply prospectively, starting 60 days after the final rule is published. Comments are due November 9, 2026.
Date caveat: the secondary source says DHS issued the notice on October 10, but the Federal Register document it cites is dated October 8. Verify against the notice. Every figure above comes from that one secondary summary.
What it does not decide
This is a proposal. Nothing is owed by anyone, and the text can change after comments. It is not known from this source what statutory authority DHS relies on to charge a fee of this size, or what DHS's cost basis is for the amounts; those are in the notice and should be read there.
WhyNobody is bound yet. If finalised as proposed, a school recommending an F-1 student for OPT would owe the fee from 60 days after publication, and would have to choose who bears it. The practical step available now is a comment by November 9. A final rule would also be open to challenge on its authority, but no such suit exists on this record.
- 04
September jobs report: payrolls +29,000, unemployment 4.2%, with July and August revised down
BLS reports that nonfarm payroll employment rose by 29,000 in September and the unemployment rate was 4.2 percent, with 7.1 million unemployed. BLS describes both as little changed. Consensus forecasts cited by CNBC and others were about 90,000, but a forecast is not BLS data. The release revised July down by 31,000, from +21,000 to -10,000, and August down by 29,000, from +162,000 to +133,000. Average hourly earnings rose 0.1 percent on the month to $37.81 and 3.0 percent over twelve months. The unemployment rate has stayed between 4.1 and 4.3 percent since March. By industry, health care added 17,000 jobs, construction 11,000 and manufacturing 9,000, while financial activities lost 7,000. Household employment rose 406,000 and the participation rate rose 0.2 point to 61.8 percent.
What it does not decide
One month of payroll data from the establishment survey carries sampling error, and later revisions have moved recent months materially. Both of the prior two months were revised down in this release.
WhyThe effect is on inputs, not on anyone's legal obligations. Anything indexed to or triggered by the unemployment rate sees a 4.2 percent reading for September, and the July figure is now negative. The next release is for October, due in early November.
- 05
Supreme Court stays the injunction on third-country removals and sets the merits for December 9
In an unsigned order on September 29, the Supreme Court granted the government's application for a stay in DHS v. D.V.D. and set the case for oral argument on December 9. The stay suspends a district-court ruling by Judge Brian Murphy that required DHS to give a noncitizen written notice of a third country, meaning one not named in the removal order, and a meaningful opportunity to contest removal there. Murphy's ruling also required DHS to assess torture claims against a "reasonable" fear standard rather than "more likely than not." Justices Sotomayor, Kagan and Jackson would have denied the stay. SCOTUSblog reports they gave no written explanation; ABC News quotes a line from Justice Sotomayor, so check the order text on whether a written dissent exists.
Outlets agree on the facts of the order. The disagreement is over interpretation, and mostly in headline framing: "major victory" for the administration (IBTimes) against "for now" (UPI). The order says nothing that supports either reading beyond the interim relief itself.
What it does not decide
The Court has not ruled on whether third-country removals without these procedures are lawful or constitutional. That is the question for December
The stay is interim relief, not a holding on the merits.
WhyFrom September 29, DHS may again carry out third-country removals without the written-notice and challenge procedures Murphy's ruling required, and the lower standard for torture claims no longer applies. This lasts until the Court decides the case, which cannot happen before the December 9 argument. A noncitizen facing removal to a country not named in their order has, in the meantime, no court-ordered notice or hearing on that destination.
- 06
Supreme Court stays a district court order blocking the modified SAVE citizenship database, but limits its use to individual checks through Election Day
SAVE (Systematic Alien Verification for Entitlements) has run since 1986 to verify immigration and citizenship status for benefits eligibility, and 27 states have also used it since 2005 to check voter rolls. Following a March 2025 executive order, DHS and the Social Security Administration modified SAVE to check against SSA's NUMIDENT records, including Social Security numbers, and to allow bulk rather than only individual lookups. The League of Women Voters and the Electronic Privacy Information Center sued, arguing the changes violated the Social Security Act's confidentiality provision (42 U.S.C. § 405(c)(2)(C)(viii)(I)), the Privacy Act, and the APA. On June 22, the U.S. District Court for D.C. (Judge Sparkle Sooknanan) granted summary judgment to plaintiffs on all three claims and vacated the modified program; a divided D.C. Circuit panel denied a stay on September 4. In a 6-3 decision, the Supreme Court granted the government's stay application, holding it is likely to succeed on the merits because 8 U.S.C. § 1373's "notwithstanding any other provision of law" clauses require DHS to share and receive citizenship and immigration-status information in a way the Court held supersedes the Social Security Act and Privacy Act restrictions the district court relied on. The Court found standing for at least one plaintiff — a naturalized citizen registered in Louisiana whose SSA record had not updated — but held that risk did not outweigh the government's showing of harm. The order preserves the National Voter Registration Act's 90-day pre-election moratorium on "systematic" list-purging, but allows "individualized" citizenship-verification inquiries to proceed during that window. Justice Jackson's dissent, joined by Sotomayor and Kagan, argued the majority granted emergency relief without full briefing or argument on two never-before-interpreted statutory provisions.
What it does not decide
This is a stay of the district court's injunction pending appeal to the D.C. Circuit, not a final ruling on the Social Security Act, Privacy Act or APA claims, which remain to be litigated there. The Court did not decide whether the NVRA's 90-day moratorium applies to programs that specifically target noncitizen registrations rather than "systematic" removals generally — that question is pending separately in Republican National Committee v. Mi Familia Vota, No. 25-1017, not yet argued. No state is compelled to use SAVE; its use for voter-roll checks remains voluntary.
WhyStates that use SAVE for voter-roll checks may resume using the modified, Social-Security-linked version for the rest of the 2026 cycle, limited to one-by-one checks rather than bulk searches through Election Day. A voter flagged by the tool — including a naturalized citizen whose SSA record has not caught up, the harm the named plaintiff alleged — would now have to affirmatively produce proof of citizenship to stay registered, starting immediately, while the underlying legality of the program is still being litigated in the D.C. Circuit.
Typesetting
TeX, Typst, type design. One specimen and one small challenge per week.
Typesetting
- 01
Assignment: a titlepage to numbers, in LaTeX and in Typst, with the calligraphic capitals checked
Set a titlepage for a short set of notes: title, author, date, one-line abstract containing
\mathcal{L}(Typst:$cal(L)$). Do it once with thetitlepageenvironment and once with Typst'spageandalign, in under an hour, from a document you already compile.Success criterion: the title block's optical centre sits at 38 to 42 percent of the page height measured from the top; exactly three sizes are used (title, author and date, abstract), each at least 1.25 times the next smaller; the abstract measure is 45 to 60 characters; and the calligraphic L in the abstract is the same glyph in both outputs, checked at 400 percent zoom. Write down the measured numbers for both.
WhyPlacing a block by numbers instead of by eye is the technique that transfers: the next preprint or lecture handout gets the same ratios for free. The calligraphic check is the second half: it is the cheapest test of whether the two toolchains agree on a glyph you actually write every week.
- 02
Typst 0.15.1 ships New Computer Modern 8.1.1, fixing a regular-weight math font that drew old calligraphic letterforms
Typst 0.15.1 (2026-07-17) updates the bundled New Computer Modern fonts to 8.1.1. The changelog says this fixes the regular weight of the math font, which still used old calligraphic letterforms that were meant to live in stylistic set 6. The same patch fixes two math regressions: alignment points inside
lr(matched delimiters) and vertically misalignedopelements. I have not compared specimens myself; the changelog gives the cause, not a picture.WhyIf you set a Lagrangian, a Hamiltonian or a Hilbert space with
cal(L),cal(H), the glyph you got from 0.15.0 may not be the one Computer Modern users expect, and a document compiled before the patch can change shape when recompiled after it. Pin the compiler version for anything you will reissue, and look at one page of calligraphic capitals before and after. The alignment regression matters more for derivations: an&inside a delimited expression that silently stopped aligning is the kind of bug that survives to the PDF. - 03
Assignment: diagnose one badly set page by measure, leading and hierarchy, then fix it to numbers
Take any page of your own notes compiled with default settings, or a
\documentclass{article}page with\usepackage[margin0.75in]{geometry}= at 11pt on letter paper. Measure three things on the printed or zoomed page: characters per line on a full line (count one, then trust it), the baseline-to-baseline distance divided by the type size, and how many visually distinct levels the headings, body and captions have.Success criterion, all numbers: the measure sits between 60 and 75 characters; the leading is 1.4 × the type size (15.4pt on 11pt type); and no more than three levels are distinguishable, each by one change only (size, weight or space, not all three). Fix the page by changing the margins or column width first, then
\linespreadorset par(leading: ...), and re-count. Write down the before and after numbers; if the measure was over 90, say which of the two fixes (narrower text block, larger type) you chose and why.WhyThe diagnosis transfers: counting characters per line and dividing leading by size takes a minute and works on every document you will ever be handed, including the next paper draft. Wide measure with tight leading is the most common reason a page of mathematics feels dense, which is the failure to look for first in your own notes.
- 04
Typst 0.15 allows several bibliographies in one document — the per-chapter reference list that LaTeX needs =refsection= for https://github.com/typst/typst/releases/tag/v0.15.0
Typst 0.15.0 (2026-06-15) lets a single document carry more than one bibliography, so a thesis or a collected set of lecture notes can end each chapter with its own reference list instead of one list at the back. The same release adds a
withinselector, adividerelement and an experimental bundle export target that writes several output files from one source. The release notes describe the feature; check the 0.15 bibliography reference for the exact syntax before relying on a form remembered from here.WhyThis removes one of the few structural reasons a long document went to LaTeX: per-chapter bibliographies there mean
biblatexwithrefsectionor themultibibpackage, and a second pass of the bibliography tool. The open question is what happens to a citation that appears in two chapters, which the release notes do not answer; test with one shared reference before converting anything. For a single paper with one reference list this changes nothing.