Written by an AI agent working in this repository.
The heavy-hydraulics programme asks when a small excavator should carry a cheap fixed gear pump instead of a load-sensing piston pump. The answer is a threshold efficiency, E50, that a selection rule predicts from data-sheet numbers. A one-number idle rule beats today's two-number rule (0.0047 against 0.0056 left-out RMS) when each pump's idle drag is averaged over the pump sizes the fuel optimiser picks for that pump. A builder cannot run that optimiser, so the programme has been looking for something a publisher could print once.
Where the last step left it
Milestone 7q tried a pick table: a publisher runs the model on the pumps it has and prints their picks, and a builder reads the row whose idle drag curve bends most like theirs (drag at 79 L/min over drag at 64 L/min). It failed, 0.0075. One pump, the Danfoss family, bends far more than any other (1.78 against 1.28 to 1.50). The optimiser sizes it down to 58.5 L/min, while every row in the table sat near 72 L/min.
What 7r tried
If no measured pump bends like Danfoss, the publisher can make rows for bends nobody measured. For each donor pump, multiply its own idle curve by a power of flow so that its bend is exactly 1.2, 1.3, ... 1.9, run the motion model on the bent curve and tabulate the picks. Danfoss's own curve enters nothing except as its data-sheet bend when it is the pump being predicted. Eight predictions were written in the log before any number was computed, and 7q's results reproduced byte for byte first.
What came out
| Rule | (A) 308 diagonals | half idle speed held out | idle shares held out |
|---|---|---|---|
| two-price rule (owh) | 0.0056 | 0.0059 | 0.0057 |
| 7q pooled table | 0.0061 | 0.0064 | 0.0067 |
| 7r bend table | 0.0052 | 0.0056 | 0.0055 |
| leak-free own picks (needs the model) | 0.0047 | 0.0049 | 0.0047 |
The bend table beats the two-price rule in all three scores. Almost all of its gain over the pooled table is the Danfoss family, whose error falls from 0.0122 to 0.0063. Five of eight predictions held.
It is not the whole fix. Bending the donors' curves does pull their picks down, but slowly: the row's median falls from 74.5 L/min at bend 1.2 to 68.0 at 1.8, about a third of the way to Danfoss's own 58.5. So the row still charges Danfoss about 1.19 times the drag it really pays, and Danfoss's error stays just above the two-price rule's (0.0063 against 0.0061). Whatever else makes the optimiser shrink that pump (its drag level, how its efficiency falls at small sizes) a power-law bend does not carry.
What it means
This is a sensitivity, not a result: the bent curves are synthetic, and a different synthetic shape could move a 0.0004 margin either way. It does say something a builder can use. A table can carry part of the own-pick gain once it has a row for the pump's bend class, and the thing that would turn the sensitivity into a result is a measured second gear-pump family whose idle drag bends like Danfoss's. Until then the two-price rule stays the recommendation.
Code: gear_pump_bendtable.py; results: gear_pump_bendtable_results.txt.