The heavy-hydraulics programme asks when a small excavator should carry a cheap fixed gear pump on an open-centre valve instead of a load-sensing pump. Its selection rule gives the gear-pump efficiency at which the two tie on fuel. Two earlier steps today tried to shrink that rule to one number per idle hour, the gear pump's idle drag minus the load-sensing pump's standby power, using drag read off a data sheet at one flow and then at two. Both fell short of the two-price rule, which prices drag and standby separately. This step asked why: is the one-number form wrong, or only the way the drag was read?
The test
The earlier reads all sampled each pump's drag curve at fixed flows: the median and quartiles of the pump sizes the fuel optimum picks, pooled over six pump families. This step instead read each of the 14 pump variants at the sizes the optimum picks for that variant, on its own break-even cell, over the same 384 excavator geometries, 22 idle conditions and 4312 cells (built from a measured excavator's motion, fuxi-robot/excavator-motion on Hugging Face). Seven predictions were written down first (LOG.md, milestone 7n), and the previous step's results file reproduced byte for byte before any new number was computed.
What came back
All seven held.
| Rule | Left-out RMS, 308 diagonals | Unseen half idle speed | Unseen idle shares |
|---|---|---|---|
| Two-price (drag at 85 L/min, standby) | 0.0056 | 0.0059 | 0.0057 |
| One number, pooled-pick drag | 0.0056 | 0.0059 | 0.0059 |
| One number, two data-sheet reads | 0.0066 | 0.0074 | 0.0070 |
| One number, own-pick drag | 0.0042 | 0.0047 | 0.0043 |
Standby and drag trade 0.98 for one in the fitted rule, as the break-even model says they should.
The reason the fixed reads failed is visible in the picks. For 13 of the 14 variants the optimum's median pick sits between 68 and 75.5 L/min, close to the pooled 72. Danfoss, whose idle drag curve bends, is the exception: its picks split, with a median of 63.5 L/min and a lower quartile of 57. The optimum sizes that pump down, so at its own picks it pays 0.782 of the drag the pooled picks charge it. The spun and geared Rexroth AZPN variants do the same thing for a different reason: their lower quartile drops to about 55 L/min to stay below the size where the drive's extra drag starts. Across pumps, the own-pick drag carries only 0.743 kW per kW of the pooled-pick read. No fixed read can match that, because the shift depends on how high or steep each pump's own drag is.
What it does not show
This read is an oracle. The own picks come from the same model and motion record that produce the crossings being predicted, including for the pump family held out of each fit, so some of the gain may be that leak. No builder has those picks. The result says the one-number form is sound and locates the problem in the read; it does not give a builder a new rule. The practical advice stays the same: price idle with two numbers, the drag at 85 L/min and the load-sensing standby, each per idle hour.
The next step is a builder-side stand-in for the own picks, for example shifting the pooled pick weight down by the slope of the pump's drag curve, scored against this oracle and against the two-price rule.
Code: experiments/gear_pump_ownpick.py; results: gear_pump_ownpick_results.txt.