[JamesL787]

Depends on #24: stacked on it (its commit is the first one here) and should merge after it; only the last two commits are new.

What it does

• modeld (both runtimes) records where the steering came from: modelV2.action.lateralSource is plan, actionHead or curvatureHead. delayInput records how the told delay reached the model.
• On a plan model (Pop v2, TCP v3) the Firmware Controller's command delay becomes max(0.3 - lateralActionTime, 0). That undoes the MIN_STABLE_DELAY floor and nothing else. Action heads keep the calibration's delay. The delay slews at 0.3 s/s when the model changes.
• The TEG-A010 and the C020 get their own prediction schedules:
  • TEG: 0.20 s flat.
  • C020: 0.28 s in town, 0.26 s on the highway (provisional).
• The cars PR Firmware Controller: corrected-PTM calibrations, R6 from Table A, each EPS's own E4 #24 adds take their measured sibling's schedule: the A030 the TEG's, the C120 and TGG the C020's. The CR-Vs keep live/manual delay; no CR-V timing has been measured.
• With a schedule set, lagd is locked out for those cars, as it already is on the Clarity and the Insight.
• The Clarity's command delay defaults to 0.125 s above 15 m/s (was 0.025): on an action-head model it ran 0.10–0.14 s early there without it (routes 3f1/3f5). A saved value is kept.
• The Civic command delay (0.175 s) now holds above 15 m/s. The Insight's and the CR-Vs' still fade out there.

Why (measured by speed band on every local route)

• TEG (its owner's routes 12e–13d, Pop v2):
  • The plan is read at lagd + 0.075 ≈ 0.44 s, but the car reaches a command 0.26–0.31 s after the frame.
  • So it turned in early, and above 15 m/s, where the delay faded out, it ran 0.13–0.17 s early.
• C020 (route 380): 0.16 s early at 20–40 m/s.

Not yet confirmed

• The C020 schedule comes from a drive on an older controller, corrected for this one's output filter. It should be re-measured on a drive with this controller.
• No drive has been made on this change yet.

Tests: new test_model_source.py, plus updates to the Civic profile, starting-calibration, runtime and timing-metadata tests. No new failures. Three tests read the opendbc submodule's files from disk and need #24's opendbc pin checked out. settings_ui.json was recompiled for the delay text. The NrdrYawCommandDelayHigh default changed, so the params registry is regenerated (a native rebuild picks it up).

🤖 Generated with Claude Code

[nrdr]

Already picked up in nrdr-development: dd3daff and 92fab80 match the two changes here. The later model-source fix is also in db87bbb. Closing this one so it doesn't look like we're still missing it. Thanks!