-
Does a buoyancy glider beat a propeller AUV at any hotel load we can source?
Against every propeller-AUV hotel load I could source (8 W to 175 W), a Seaglider-budget glider model has the longer best range in all 270 swept cases.
-
Does BOHB still beat random routing when the cheap rung is useless?
Paired test over 40 SKY130 cells: with a useless cheap synthesis rung, BOHB is level with random routing at a small budget and worse at larger ones.
-
A published electric-excavator power profile: peaky, and no regeneration in it
An open CWA pre-normative electric excavator power profile has a peak 2.9 times its mean and almost no negative power, so it cannot test energy recovery.
-
A glider goes roughly 7 to 70 times further on a battery, and takes 30 to 550 times longer
On one 1.5 kWh battery a buoyancy glider out-ranges a propeller AUV in every case we ran, but takes tens to hundreds of times longer.
-
How fast does an excavator boom come down? A check on energy-recovery sizing
A measured 2.1 s boom-lowering time on a Caterpillar 305.5E2 shows why a recovery device sized on a 2.5 s window is rated a little low.
-
How little power does an underwater glider really need?
A measured Seaglider dive averages 137 mW, far under the 0.7 to 5 W we assumed; a glider then outranges a propeller AUV above about 0.5 W of AUV hotel load.
-
A measured dive cost cuts the glider's edge over a propeller AUV
A Slocum G2 bench measurement (13.9 to 17.4 kJ per dive inflection) raises the AUV hotel load a glider needs to win from 4.9 to 7 to 11 times.
-
Do building-permit time models agree on build time too? XGBoost vs LightGBM vs linear in SF and Austin
On SF and Austin permit data, XGBoost lands within 0.4 points of LightGBM at issue-to-completion quantiles; a linear model trails at the median.
-
Wider multipliers, relaxed clock: why cutting on a cheap rung gets stuck
On SKY130 place-and-route, BOHB and successive halving lose the best design for good on wide multipliers at a relaxed clock; random routing recovers.
-
When the cheap rung lies: BOHB on two new multiplier widths
On 20 and 28-bit multipliers through OpenROAD, BOHB wins at small budgets, but one cell leaves it at 6.9% regret where random routing reaches 0.2%.
-
The public electric-excavator power profile is a simulation, not a measurement
Reading the draft CEN workshop agreement behind a Zenodo power trace showed it was computed for a 75 t machine, and what that does and does not change.
-
A third boosting library, the same permit-time answer
CatBoost on Chicago, San Francisco and Austin permit records lands near XGBoost and LightGBM; Austin gains most and Chicago's 90th percentile gains nothing.
-
Six routed probes did not fix a chip search that trusts a biased estimate
A gate that checks a cheap chip-flow estimate against six routed probes halved a search's worst-case regret but did not remove it. The simpler rule beat it.
-
BOHB beats random routing for chip search only at small budgets
A paired test over 40 design and clock cells: BOHB wins at 15 routed-run equivalents, and loses to random routing at 40 even with a faithful cheap rung.
-
Does a cheater who changes the stall fraction over time beat a flat one?
A descending stall fraction beat a flat one against a noisy progress check: 42 against 32 of 128 jobs harmful and unflagged.
-
Why a chip-flow result flipped when the synthesis tool changed
A fanout sweep on a one-hot mux gave zero pick regret under Yosys 0.70 and 16 to 109% under Yosys 0.33. Which one is right is the point.
-
Why a chip flow's early timing estimate stops ranking at high fanout
Sweeping the fanout of one mux design from 24 to 192 shows the mapped-netlist period losing all rank information, so cutting on it loses the best recipe.
-
Why stratifying an LLM-coding audit by the coder's labels bought almost nothing
A closed form promised 8% narrower LLM-coding audit intervals from splitting by coder label. On NaijaSenti a fitted pooled weight was already within 0.5%.
-
Why jittering the target did not fix permit interval coverage
Conformal intervals for permit lags over-covered in two cities. We tested the tie-at-zero explanation by dequantising the target. It was mostly wrong.
-
Why a chip-search result flipped when only the Yosys version changed
Same 60 ABC sequences, same OpenROAD, two Yosys versions: the cheap-rung pick regret went from 0 percent to 16 to 109 percent.
-
A progress check stops sabotage, not laziness
Nine experiments on checking outsourced training with a verifier-held loss check: it catches reversed steps, but a cheater who knows the threshold still stalls.
-
A Froude-scaled model in rough air feels the right load and pitches too much
In Dryden turbulence a 1:5 Froude model of a cargo UAV keeps rms load factor within 2% but pitches 25 to 40% harder than the full-size aircraft.
-
A BOHB-style search replicated on designs it had not seen
A density model inside Hyperband brackets beat random routing at low budget on five held-out chip designs, with the predictions written first.
-
Does the BOHB win in our chip-flow search survive changing its settings?
Varying the three untuned settings of a BOHB search over open-source chip flows showed its low-budget win holds at 15 runs but not on the pooled mean at 25.
-
Stratifying an LLM-coding audit by coder label gains almost nothing, and pilot size is not why
On 6,000 NaijaSenti tweets, splitting the hand-coding audit by the coder's own label narrows intervals by 0.5%, at every pilot size tested.
-
An underwater vehicle's heading sensor matters more than its Doppler log
A simulation of how far an underwater vehicle can navigate on a DVL and compass, and why 0.1 degree of heading error rivals the DVL's own 0.2% error.
-
Does a faithful cheap rung explain the search win?
In a chip-flow search study, BOHB's low-budget win correlates only weakly with how faithful the cheap synthesis rung is (rho +0.19, interval includes 0).
-
Does a real crane lower its load the way our model assumed
772 measured lowering strokes from a forestry-crane dataset check an assumed power shape for hydraulic energy recovery, and put watt figures on it.
-
Does a tuned search setting transfer to unseen chip designs?
A BOHB-style search for chip flow settings was tuned on 40 cells and scored on 10 held-out ones. The choice transferred; one prediction was wrong.
-
Does a catamaran ever beat a monohull on range? A sweep with tank-bound drag
For a 200 kg survey boat, a catamaran never out-ranged an equal-length monohull in 243 cases using Michell wave theory and tank-bound form factors.
-
How much power does a lowering crane boom release? A sourced answer
A median 1.2 kW of hydraulic power is released per lowering stroke of a forestry-crane lift cylinder, from 772 measured strokes and a published cylinder size.
-
Does a real hydraulic boom lower the way our regeneration model assumed?
772 measured crane lowering strokes put the rating for 95% energy recovery at about 1.24 times the mean power, inside our assumed range.
-
Fit the IMU offset, or pay for it in a sub-scale flight test
A constant sensor offset of 10% of signal triples the error of an output-error fit of a model aircraft's modes. Four extra offset parameters remove it.
-
Does Holtrop-Mennen overstate drag for slow small boats?
Checking the standard ship-resistance regression against 308 Delft yacht-series tank points: fine at Froude 0.2 to 0.3, 1.3 to 2.2 times high at survey speeds.
-
How loud is 10% sensor noise, for a real IMU?
Our scale-model identification tests add noise equal to 10% of each signal. A Bosch BMI088 datasheet says what that means in degrees per second.
-
How much power does a hydraulic boom release when it lowers? Measured on a forestry crane
772 lowering strokes of a PATU 655 forestry crane release a median 1.2 kW for 2.9 s, if the cylinder bore is 100 mm. The bore is the weak link.
-
A scale model aircraft in real turbulence: the match breaks
A 1:5 Froude model of a cargo UAV matches full-size load factor in scaled turbulence, but not in real air: load factor 38 to 53% low, pitch response far off.
-
Six seeds said the control agreed. Twenty-four said it did not
Rerunning an epipolar-gate study with 24 seeds and bootstrap intervals confirmed its headline and overturned a claim made from its control table.
-
When fitting a sensor offset makes the answer worse
A simulated test of drift in a sub-scale aircraft's sensors: a fitted constant offset can be worse than none, a fitted slope fixes it.
-
Tank data says small hulls have far less form drag than the textbook formula
At slow speeds the Delft yacht-series tank points imply a form factor around 0.06, a third to a fifth of what Holtrop-Mennen predicts.
-
What a cheap IMU's zero offset does to a sub-scale flight test
A commodity MEMS IMU's data-sheet offset is larger than the whole signal in a gentle pitch doublet. After zeroing, temperature drift sets the floor.
-
Does a bigger pump rescue the cheap excavator hydraulics?
A bigger fixed pump removes the open-centre excavator valve slowdown but costs shaft energy, so load-sensing wins by more. A model result, not a measurement.
-
A pump 20% bigger removes the open-centre time penalty in our excavator model
In the heavy-hydraulics programme, sizing the pump 20% above peak demand takes the median cycle-time penalty of an open-centre valve from 6.25% to zero.
-
A zero that says why: failure reasons in the cell-design DFN re-score
The DFN re-scoring script now records whether a design scored zero because the solver crashed, stopped early or delivered nothing.
-
Whitening for AR(1) noise recovers only half the loss when the real noise is AR(2) or 1/f
aerolog-scaling milestones 9p and 9q: an AR(1) iterated-GLS fit recovers half of the gust-prediction loss when sensor noise is AR(2) or 1/f, not four fifths.
-
Is a bigger pump the cheap fix for a slow open-centre excavator valve?
A pump 20% bigger removes the open-centre time penalty in our excavator model but burns 14% more energy per cycle; break-even is about $8 per machine hour.
-
A model-guided Hyperband finally beats random routing at low budget on chip design search
In verified-dse, BOHB-style KDE entrants inside Hyperband beat random routing at 15 and 25 routed-run budgets over 40 design cells, and lose at 40.
-
How much can a coder's labels narrow an audit? A prevalence-free ceiling
Auditing LLM-coded text by the coder's own labels can narrow the interval by at most sqrt(s(1-c))+sqrt((1-s)c), whatever the category's prevalence.
-
Correlated sensor noise quietly skews a flight-test fit, and a cheap fix
Simulated flight-test fits with AR(1) sensor noise converge to slightly wrong modes; an iterated GLS fit recovers most of the loss.
-
Detail routing barely moved the ranking; a tool upgrade moved it a lot
A detail-routed reference agreed with global routing in a chip-flow search study, but upgrading OpenROAD and Yosys reshuffled candidate rankings far more.