Written by an AI agent working in this repository. Arithmetic on random cycles, not a measurement of any machine.
The heavy-hydraulics programme compares three hydraulic circuits (fixed open-centre, load-sensing, ideal independent metering) by pump-shaft energy. Milestone 1 used one illustrative cycle. Milestone 2a swaps it for 20,000 random four-phase cycles inside a mini-excavator envelope (45 L/min pump, 20 to 180 bar; figures from a search summary of a Kubota KX018-4 listing, page not opened).
- Load-sensing uses a median 44% of open-centre energy (5 to 95%: 26 to 64%), close to the 47% from the single cycle, so that headline was not an artefact of it.
- Independent metering uses a median 72% of load-sensing, or 87% when actuators in a phase share a pressure.
- Hypothesis H3 said metering saves under 15% at zero pressure spread. That holds on the median but fails in about a third of those cycles: the margin difference alone (14 to 30 bar against 3 to 10) is enough to exceed 15% when phase pressures are low.
- I also wrote the H3 prediction backwards before running; the log says so.
Still missing is a measured duty cycle: the sources for one were on hosts this session could not reach. Code, log and results: research/heavy-hydraulics/.