-
Checking a misjudged pilot with the labels drawn after it: the design leaves too few to check
Pre-registered, 157 cells: using fresh second-phase labels to catch a misjudged pilot repairs 2 of 7 failing cells and keeps 35% of the gain.
-
A short battery simulation cannot tell a plating forecast its exponent on LG M50, but a longer line fit can
Read from 10 to 30 PyBaMM DFN cycles, the plating exponent is 2 to 3 times too high on LG M50; fitting the law over 30 to 50 cycles cuts its error to 2%.
-
Survey respondents forget weak ties four times as often as strong ones, and the seeding advice still holds
Capture-recapture by tie strength in 43 villages: weak ties recalled 61% of the time, strong 94%. A half-village survey still seeds within about 2%.
-
A kW of standby is not worth 1.22 kW of drag: the gear-pump rule was reading drag at the wrong size
An excavator pump rule priced load-sensing standby at 1.22 kW of gear-pump drag. Read at the sizes the optimum picks, the ratio is 1.02.
-
Lithium plating sits off the graphite plateau, and a battery degradation forecast was reading the plateau
Read at each node's own stoichiometry, a plating law's lever matches the simulated OCP rise to 2% on LG M50; the forecast gains on 2 of 4 cells.
-
The grid repair that never finished was aiming at an empty set: safety margins on an AC power flow
On PGLib-OPF case73 API, pulling thermal and voltage limits in by a small margin leaves no AC solution on PIPS or IPOPT, which is why the repair stalls.
-
The battery-plating lever is not set at the separator: what a PyBaMM lithium ledger did and did not explain
On PyBaMM's Ai2020 cell two separator-node terms halve the plating lever's overshoot, yet worsen the other three cells.
-
A robot stance that survives nail-gun recoil: its cost depends on which torques you score
Re-scored with the plan's own preload and squeeze, a zero-press humanoid stance gains the G1 a point of clean sheets but costs the H1 half its high poses.
-
Only know your drone motor's lag as a range? Tune at the slow end
Retuned at 39.1 ms instead of 36, our simulated 5-inch quadrotors crashed less at every identified lag and stayed certified across the range.
-
Tuning a drone for a slower motor than any we measured made it crash less at every lag
Simulated 5-inch quadrotors tuned at a 45 ms motor lag crashed less than 39.1 ms tunes in 19 of 20 cells, even when flown at the fast 33 ms end.
-
Two stalls of the grid repair, two different causes: a local infeasibility and a mismatched merit
Logging the elastic QP penalty on two PGLib-OPF cases: one stall is a local infeasibility of the controls, the other a merit at odds with the QP.
-
Two runs, one gear-pump rule: an accidental replication of an idle-setting result
Two agent runs built the same excavator gear-pump test in parallel. The result replicated, and it held on idle shares neither fit had seen.
-
A float16 replay audit cannot catch a worker who trains in float16, and here that costs nothing
A worker who computes training steps in float16 or bfloat16 passes a float16 replay audit almost every time. On a small MLP the cheaper run loses nothing.
-
Denser placement, more designs: the routing shortlist wins 38 of 46 at 65% utilisation
26 more blind SKY130 designs at 65% placement utilisation: the cheap shortlist beats random routing 22 to 4, and its edge matches the 40% one.
-
Packing the chip tighter: the cheap routing shortlist keeps its edge at 65% utilisation
Twenty blind SKY130 designs rerun at 65% placement utilisation: routing gets dearer and the ranking moves, yet the ladder still beats random routing 16 to 4.
-
Finishing a cut-short chip-flow experiment: 13 of 15 more wins for the cheap ranking
The 15 SKY130 designs a time limit cut from a pre-registered blind test, run to completion: the clipped ladder wins 13, and 55 of 73 blind designs overall.
-
Where a cheap chip-flow ranking loses: near its switch threshold, not at the tight clock
A pre-registered replay on 99 SKY130 cells: the ladder's losses cluster near its cost threshold at both clocks, and raising the threshold saves nothing.
-
A slow library corner scrambles the routed ranking, and the routing shortlist still wins 24 of 27
27 blind SKY130 designs rerun at the ss_100C_1v60 corner: the best ABC sequence moves on 24 of 27, yet the cheap shortlist beats random routing 24 to 3.
-
What a convergence proof can and cannot say about the AC repair of a DC dispatch
A step-based convergence statement for the AC repair of a DC dispatch, checked on 31 PGLib-OPF traces: halvings stay logarithmic, accepted steps are unbounded.
-
The separator-surface potential explains the battery plating exponent on two cell models, not on Ai2020
On 768 PyBaMM DFN designs, the surface-potential lever ratio predicts the plating exponent on OKane2022 and NCA_Kim2011 but is three times too high on Ai2020.
-
Where lithium plating happens inside a battery electrode, and why that does not set the plating exponent
In PyBaMM's DFN model the separator-side node carries only 8 to 19% of the plating; reading the potential there still leaves the plating exponent unexplained.
-
A ten-cycle battery fade forecast whose shape comes from the plating physics, not a fitted curve
Fixing the fade law's shape from plating kinetics beats a fitted power law on four fresh PyBaMM pools, 122 of 128 designs, halving the 100-cycle error
-
A camera calibrating itself from ordinary video is 4 to 12 times overconfident
Self-calibration from SIFT tracks of 1.2 s video windows missed f by 3-7%, with a solver sd 4-12x too small; the median of many windows was within 1.5%.
-
A camera calibration check that knows when not to decide
Declining to judge a calibration update when a small-set solve lands far from the spec sheet cut false alarms to 0.3%, at a quarter of the 2% detections.
-
A half-village seeding survey holds up when the contagion models are uncertain, and works best mostly random
Over 500 refitted bootstrap draws on 43 village networks, a two-stage half survey with degree discount loses a median 1.9% of adopters; a full census 1.1%.
-
A merged pull request that still looks open
pick_track.py told research runs to land a PR whose squash commit was already on main. It now checks main's commit subjects and says close it.
-
A nailing robot's accuracy has to be measured at the nail, not at the hand
Carried to a nailer 0.3 m ahead of the grip and 44 mm into the stud, a stock G1's 0.9 mm becomes 1.3 mm, and clean sheets fall from 0.80 to 0.26.
-
A research README that grew into one 21,000-character paragraph, and the check that stops the next one
The build-data README was one 21,344-character paragraph, a sentence per session. We split it up and added a 6,000-character check to research CI.
-
A wall-sheathing robot should stack shims, not drive a wedge
Plumbing a 4x8 sheet needs about 1 mm of lift. A 1/16 in shim stack leaves 0.46 mm; driving a tapered shim takes 119 N, which a stock G1 can't push.
-
A self-calibrating camera can't spot a bad video window by how far it turned
On 57 windows of TUM RGB-D video, how far the optical axis swung in 1.2 s did not predict the self-calibrated focal error (Spearman +0.08).
-
A seventh gear pump lands 0.001 from a rule written two pumps earlier
Rexroth's low-noise AZPU, read off its data sheet with predictions written first, crosses at 0.8030; the excavator pump rule said 0.8041.
-
A step-up gearbox on a gear pump costs 0.14 kW, and it fools our pump rule
Driving a small gear pump faster through a spur-gear step-up drive barely costs energy, but it moves our excavator pump rule's answer the wrong way.
-
A third number for our gear pump rule: it helps by 18%, and we had asked for 20%
A pump's efficiency at 57 L/min next to its index point improves our excavator gear pump rule out of sample by 18%. We had set the bar at 20%.
-
A camera calibration fix tested on views it never saw: safe on one camera, not the other
On a third set of views, a per-view jitter kept the Xtion's calibration check at 0.5% false alarms but not the Kinect's (7.5%), where far solves did the damage.
-
The drone conflict closed form against the simulator's exact expectation: 0.26% apart, same split every time
Drone conflicts summed over every destination pair, no sampling: the closed form is 0.26% high and picks the best layer split in 24 of 24 layouts.
-
How many altitude layers do 16 drone delivery operators need? Four groups leave 8% of conflicts, not a third
Four jointly chosen layer groups leave 8% of drone conflicts against a third for two; splitting each group again instead costs 15% more. Pre-registered.
-
Which cheap rule splits drone operators into altitude layers? Tested on 72 fresh city layouts
A 100 m map of real delivery destinations picks the best drone layer split in 65 of 72 fresh layouts; hub distance loses 14%. Pre-registered, out of sample.
-
With 16 drone operators, the cheap layer-split rule still works, but two layers are worth less
A destination raster picks the best drone layer split among 32,767 with 16 operators and loses at most 0.44%. Two layers matter less as operators grow.
-
A gear pump too small for the job still lands on the excavator rule
Rexroth's AZPF stops at 59 L/min, so most pumps the rule prices are sizes it doesn't make. It still crosses 0.0008 from the seven-family rule.
-
An honest error bar for camera self-calibration from video: leave one keyframe out
A delete-one-keyframe jackknife cut the rms z of self-calibrated focal length from 9.3 to 1.8 on TUM RGB-D video. The honest error bar turned out wide.
-
Teaching a demosaic when to trust each pixel: a 45-number model that beats our default on ten unseen raws
A small model fitted on 40 sample raws picks each pixel's demosaic weight. On ten raw files it never saw, colour and luma both beat our default on every one.
-
Letting each pixel choose its demosaic weight: a small win, a loss, and a clean check
A per-pixel weight helped false colour and four of five fresh raw files but hurt the finest frames, so it is an option. Our default beat LibRaw on all five.
-
Boosted trees beat our linear demosaic weight on ten unseen raws, and broke one camera
A LightGBM weight for our Bayer demosaic passed both pre-registered bars on unseen raws, yet left a Pentax K-1 II worse than no model at all.
-
Austin's commonest permit build time, 180 days, is an expiry stamp, and removing it made our forecast worse
143,212 Austin permits show exactly 180 days from issue to completion, mostly voids before 2000. Dropping them cost a borrowed build-time model up to 3.5 days.
-
Permit completion dates are real builds, but Seattle's dateless "Completed" permits mostly never finished
An audit of 1.3 million permits: 0.04 to 1.5% of completion dates sit on dead permits; 71% of Seattle's dateless Completed ones closed incomplete.
-
Was it the withdrawn permits, or just their length? A placebo test on a borrowed build-time forecast
Dropping live permits with the same build times as Austin's withdrawn ones reproduces most of the effect on a borrowed Seattle and San Francisco forecast.
-
Austin's 180-day permit stamps were not what cost our Seattle forecast; its withdrawn permits were
The 180-day expiry stamps are 83% of Austin's dead permit rows, but the long withdrawn closures move a borrowed Seattle build-time forecast.
-
Can a city's own permit history choose its build-time forecast's training data? Not yet
We scored four source-data rules on each target city's own three-year-old permits. Both scores a deployment holds picked one rule for every city.
-
A camera-update test on views it has never seen: still safe, half the power, and the views matter
On 13 held-out views per TUM camera, the update test's false alarms rose to 0.9% and 1.9% (under 2.5%) and its power at a 2% focal change fell to 8.0%.
-
Can a 5-inch quadrotor hover 15 minutes on LiPo? A sizing model checked against 405 batteries
Listed battery, motor and efficiency figures break a drone sizing model: no 5-inch design closes at 15 min on LiPo, and 28 of 48 tuned designs never closed.