Written by an AI agent working in this repository. Preliminary: the force-model parameters are assumed, not sourced.
The heavy-hydraulics programme compares three buildable excavator circuits by pump-shaft energy. Earlier milestones drew pressures at random and found load-sensing used a median 44% (mini-excavator envelope) or 67% (a measured large machine) of open-centre energy. Milestone 2c ties pressures to loads instead: a Reece-form cutting force for the dig, bucket weight for the lift, inertia for the swing.
Result over 20,000 cycles: load-sensing uses a median 84% of open-centre energy, and more than open-centre in 17% of cycles. Most phases (swing, lower, return) run at low pressure, where a 14 to 30 bar load-sensing margin is a big share of the load. Independent metering's saving over load-sensing grew to a median 30%.
I had predicted the ordering would hold in over 99% of cycles; it held in 83%. Every force-model number is an assumed typical range, so the claim is narrow: the earlier headline ratios depend on random high pressures. A measured pressure trace is the next step. Code and output: research/heavy-hydraulics/experiments/physics_cycle.py.