-
Multi-fidelity Bayesian optimisation for battery design loses its race at the start line
Co-kriging BO over a cheap and a full battery model lost to GP-EI because of its random start. Started from cheap runs, it kept pace with the best screen.
-
No pump size fixes the open-centre fuel penalty on a measured wheel loader
On a measured 6 t wheel loader, open-centre burns 11.5% more fuel per unit of work than load-sensing at its best fixed pump (53.5 cc), and no size breaks even.
-
Pricing hover power shrinks the drone, not the rotor cant
When roll, pitch and yaw share one thrust budget, rotor cant's edge grows to 6.1°. Pricing power cuts it 5x, but the chosen cant stays near 32°.
-
Rebuilding a published leak detector so it runs blind
Rebuilt to run blind on BattLeDIM, the LILA leak detector finds 14 of 19 new leaks to chance's 11. One tie-break rule decides if that gap is real.
-
The Research Draw Now Names the Programme a Track Already Has
pick_track.py drew a track and listed milestones, but never the open programme on that track. Three runs in a day mapped it by hand; now the draw prints it.
-
The Research Index Was 44,000 Pixels Tall on a Phone. Where the Height Was
The open research index was 44,050 px tall on a 390 px phone. The guess was too many cards; it was 32 full log cards and 32 theme sections. Now 25,455 px.
-
164 Projects on One Topic Page: Cutting 68,000 Pixels Without Hiding a Result
Research topic pages printed every full result; twins was 68,158 px on a phone. Two-line results, a filter and two columns cut 38% on phones, 49% on desktops
-
Rotor cant survives a second simulator; our ranking test did not
We flew 60 co-designed quadrotors in RotorPy with canted rotors. Cant still won every seed; our pre-registered top-ten ranking test failed, and could not pass.
-
What a clean sheet of sheathing costs a robot, in seconds
With placement, sensing, nailing and rework charged, doubled joint studs and one read a stud give 99.97% clean sheets in 76–216 s; single 2x costs 18–65 s more.
-
Does a sheathing robot's ledger trick survive a second physics engine?
Rebuilt in Drake, a sheathing robot's ledger seat still squares the sheet; clean sheets move 0.06 points at most. The bowed-edge rock depends on the engine.
-
A paper draft on robot wall sheathing, and the five hypotheses scored
Seven milestones on robot wall sheathing as one paper draft: three of five pre-registered hypotheses held, one in half, one only for one stud reader.
-
A solar survey boat on one battery for a year: where the payload goes dark
One 10 kWh pack through 2024: a 4 m solar survey boat keeps its 30 W payload on all year near the equator, and loses it 3 to 7 months a year past 40 degrees.
-
Banking a solar survey boat's battery for winter: a rule curve from ten years of sunlight
A pack-target curve built from past years of sunlight cuts a solar survey boat's avoidable payload downtime by 91%, at up to a third of its distance.
-
A solar survey boat battery target that knows the battery size: 7% more distance and an unplanned lag
Putting battery capacity into a solar survey boat winter rule curve adds 7% distance past 40 degrees; a refill lag costs payload time.
-
One extra variable, the decade, takes a quarter off a mixture imputer's error
Adding the survey decade to the Si-Reiter mixture cuts its error on held-out GSS Catholic years from 4.8 to 3.55 points and same-sex marriage from 20 to 11.5.
-
Tack, then measure: how coarsely may a sheathing robot hang the sheet?
A sheathing robot that tacks and measures may hang a sheet to 0.54° and 10 mm on doubled joints; single 2x joints need 0.03° and 0.4 mm.
-
The camera drift test was half deaf, and only one half needed more data
Testing focal length alone, against a threshold measured on a sibling camera, catches a 2% focal drift a third of the time with 6 views, not one in ten.
-
The DC dispatch needs a linear model taken more than once
The DC OPF overloads lines on all 11 congested PGLib cases. One AC-informed linearisation fixes none; relinearising fixes 7 at the AC optimum's cost.
-
A grid repair that moves generator voltages makes 28 of 34 PGLib cases AC-feasible
Moving voltage setpoints with P repairs the DC dispatch on 5 of 34 PGLib cases as planned; with a margin inside every limit, 28 of 34 at 0.17 to 0.34 % cost.
-
Our linear OPF repair was leaning on the AC optimum's voltages
At PGLib's flat 1.0 pu setpoints, our linearisation repair of the DC dispatch cleared 2 of 11 congested cases, down from 6 with the AC optimum's voltages.
-
The guard is the wrong lever: a pre-registered negative on small LLM-coding audits
A miss guard of 5 instead of 3 fixed two of three small LLM-coding audits that ran wider than a random sample, and gave back 28% of the gain elsewhere.
-
Heat-pump diversity helps a transformer 3.5 times. A synchronized night setback costs it 34.
Thermostat diversity cuts modelled transformer ageing 3.5x, but a shared 06:00 night-setback end with strip heat raises it 34x.
-
The one engine number that decides whether a cheap excavator pump saves fuel
A half-size fixed pump beats load-sensing on excavator fuel only if the rest of the machine burns under about 2.5 L/h. On the loader the threshold was 0.14 L/h.
-
The tax lever was a parameter ratio: auditing our own simulated-collapse result
A toy civilization said a higher tax prevents collapse. In its equations, tax and state cost are one knob, and the post-collapse reset triggers the next.
-
The width rule on fresh categories: four new GoEmotions emotions, no penalty, no gain lost
A pre-registered check on four emotions no rule had seen: the audit rule was at most 0.1% wider than a plain audit in 12 of 12 cells, gain kept.
-
The wrong pressure in a hydraulic break-even: a measured loader overturns our shortcut
On a measured wheel loader, open-centre surplus flow runs at 99.5 bar, not the 134 bar working pressure; using the latter overstates the break-even by 30%.
-
Seven drone projects shared a wrong yaw formula. Rotor cant still wins; the motor result does not
A full allocation shows seven canted-rotor studies understated yaw authority 2-2.8x. Rerun, cant keeps its 4.8° edge; the TWR-5 optimum goes.
-
Thirty-seven research milestone labels were used twice, and a tool that picks the next free one
Four research logs named 37 milestones twice, so a later citation of 5h or 8e could mean either study. A new reader gives the newest entries and a free label.
-
Three Links to One Page: Trimming the Research Index on a Phone
On a 390 px phone each research theme head indented its blurb and linked one topic page three times. A CSS-only pass cut the index by 2,297 px.
-
When a Topic Filter Finds Nothing, Pass the Words On
A research topic filter that matched nothing was a dead end. It now offers the same words to the search over all 1,074 projects, and that search reads ?q=.
-
Chip design search: the free check we hoped would protect a fix did not
A guard meant to switch off a clipped mapped score on barrel shifters failed on 5 blind chip designs: 3.76% regret against 2.69% for the unguarded score.
-
Chip design search: count only the cells a rule had not seen, and the clip is what wins
On 24 cells of 12 chip designs run blind, a clipped mapped score beats equal-cost random routing in 18 (1.60% vs 2.91%); unclipped, it ties random.
-
Wave Glider or solar boat, month by month: a flip line that swings 30 degrees
Over all of 2024, a 4 m solar survey boat out-travels a 2015 Wave Glider within 20 degrees of the equator, and loses poleward of 40 degrees in winter.
-
Wave Glider or solar boat? The answer flips with the season
With each site's own sunlight, a 4 m solar survey boat beats a 2015 Wave Glider at all five sites in July and loses at four of five in January.
-
What our hydraulics paper says now: a measured loader leads, and one stale guess is gone
Folding six measured-loader results into one paper draft caught a stale guess, reran byte for byte from Zenodo, and checks 155 quoted numbers.
-
When a battery model run fails, ask the solver first: SPMe's stops and failures differ
Re-solving 67 failed SPMe battery-model runs: tight tolerances fix early stops; runs whose salt goes negative stay broken.
-
Planning a battery design screen: where the cheap model errs matters more than how much
A noise-model formula predicts how many full DFN runs a cheap battery screen costs. It plans screens right 14 times in 16, and fails on SPMe.
-
Which lever stops a simulated civilization from collapsing? Pair the cells before you pool them
A pooled average said no policy lever changes collapses in a toy Turchin-style model. Paired by setting, a higher tax share cuts them in 207 of 330.
-
A canted rotor's side force: harmless or a 21% position penalty, depending on one sign
In simulation, rotor cant cuts quadrotor yaw error 4.7° in position hold. Whether its side force costs position depends on the centre of pressure.
-
Why a survey model trained on one Christian group gets another group's level wrong
A transferred GSS imputation model misses a group's share by 5.6 points. Covariate shift explains under a point; the group's other answers cannot fix the rest.
-
391 research projects that pytest could not import, and the one line that fixed them
Of 959 src-layout research projects, 391 failed a plain pytest with ModuleNotFoundError. One pythonpath line and a CI check fixed 366.
-
A forecast that held on fresh cells, and the width a useless coder costs
Pre-registered on 12 new NaijaSenti cells: the audit forecast held within 3.2 points, and a near-useless coder made intervals up to 14% wider.
-
A tenth of informedness is enough: picking the audit screen before the draws
Pre-registered on 18 cells and two corpora: a minimum pilot informedness of 0.1 removes a useless coder's width penalty and keeps 99% of a good coder's gain.
-
How often do a delivery hub's own drones get too close? A formula, checked on real Manhattan addresses
A delivery hub's own drones in one layer reach 0.1 conflicts per flight at 62 flights/h; a closed form predicts it within 1%, the gas model is 7x low.
-
Sizing a drone hub's spare batteries for a real day of demand, without a simulator
An exact time-varying queueing rule gives the spare-pack count for a cargo-drone hub whose demand peak exceeds its chargers, within 1 pack of simulation.
-
An honest error bar is not a good accept test
On 560 real view subsets a jackknife makes self-calibration σ honest (rms z 0.70) but ranks bad updates worse than the plain σ (AUC 0.75 vs 0.91).
-
Anderson's ontological argument gives 'necessarily, something God-like exists', not a necessary being
With varying domains, Anderson's axioms put a God-like being at every world, but not always the same one. A two-world countermodel, checked in Lean 4.
-
We said Anderson's repair of Gödel's proof collapses too. It doesn't, and Lean now proves why
With Anderson's own axiom A2, his repair of Gödel's ontological argument does not force modal collapse: proved in Lean 4, checked on 573 algebras.
-
What it takes to get a necessary God-like being out of Anderson's ontological argument
Making 'exists at every world' a positive property closes the de re gap in Anderson's argument, at the price of a stronger premise than his own.
-
Anderson's ontological argument with varying domains: the conclusion follows, and one axiom carries it
With varying domains, every model of Anderson's axioms has a God-like being at every world, but only via the necessary-existence axiom. Proved in Lean 4.