[Onlyti]

cg_height now means the whole-vehicle CG height (sprung + unsprung) on the L5 spatial-strut and corner-DAE path, the same meaning L1-L4 and the L5 penalty path already use. The strut body (sprung mass only) sits at the derived h_s = (m·h − Σm_u·R)/(m − Σm_u); wheel mounts, the spawn settle and the aero arm origin follow it. On the same path the tire longitudinal force now reacts on the wheel spin about the wheel-centre height R0 − penetration instead of the effective rolling radius (scope extension, below). cg_height_ref: sprung restores the whole 0.8.1 strut-path behaviour. Breaking for every strut-path run.

Gates (work order 47 sec. 3, fixed before measurement)

Commit a7bd99f fixes every gate and the R2 expectation (h = cg_height on every path) before the core change in dd870ec.

ID	Result (local, HEAD dd870ec / a3eebaa)
G1 R2 re-judgement	at dd870ec: accel front +3.34 % / rear +2.88 %, brake front +1.33 % / rear +1.68 % — outside +/-2 % for accel. Re-judged after the reaction-arm fix as H1 (below): passes
G2 static system CG height,	z_sys − (cg_height − δ̄)
G3 L1-L4 + L5 penalty bit-identical to 645780c	10/10 serialized runs identical (generic_sedan, fs_aero_example); RL pedal/accel obs+reward digests identical; closed-loop deterministic arrays identical (only the wall-clock t_compute_s / compute_* timing arrays differ, as between any two runs)
G4 cg_height_ref: sprung bit-identical on L5 strut / DAE	4/4 serialized runs identical to 645780c
G5 0.8.1 release tests + drag height on the strut body	R1, R6, A3 L5, DAE warning tests pass; new drag-height test (in-air one-step pitch moment, ±1 mm) 8/8 within ±0.09 mm
G6 negative: h_s derivation removed	G2 fails 4/4 (−24.0 / −23.8 / −7.15 / −7.12 mm), restored passes 4/4
G7 key tests	C++ CgHeightRefKeys.* 6 tests; Python cg_height_ref 1 ctest (20 checks)
G8	same-head CI ×3 → currency commit → final run + wheels (below)

G1 note: on the strut path without a corner model unsprung resolves like sprung, so sprung↔unsprung are bit-identical by design; the ≤ 1 % sub-gate has no discriminating power there.

Scope extension: strut-path reaction arm (user decision 11-07, 0.8.2 R2 handling option 3)

R2 has to pass before 0.8.2 ships. The wheel-spin reaction T_react = Fx · arm uses arm = R0 − penetration on the strut / DAE path; the penalty path, slip, the tire model and every other use of the effective rolling radius are unchanged. cg_height_ref: sprung keeps the old arm so G4 stays bit-identical. Commit order: 0630ee6 gate + arm helper and unit tests (core unchanged) → f2e16b9 core → 32459a2 R2 back in the verdict → ceb1c4c docs, currency PENDING.

ID	Gate (fixed before measurement)	Result (local, f2e16b9 / ceb1c4c)
H1 R2 4 combos (strut, no DAE)	per axle ±2 %, sprung↔unsprung ≤ 1 % of ΔFz	accel front +0.57 % / rear +0.13 %; brake front −1.60 % / rear −1.26 % (margin 0.40 / 0.74 %p); sprung↔unsprung 0
H2 L1-L4, L5 penalty	bit-identical to 645780c, RL digests	10/10 identical; RL pedal/accel identical; closed-loop deterministic arrays identical
H3 cg_height_ref: sprung	bit-identical to 645780c (strut / DAE × 2 cars)	4/4 identical
H4 existing gates	R1, R6, A3 L5, R5, DAE warning, G2, G5	targeted 70/70 pass; canonical 615/615 (H6)
H5 negative	arm fix reverted → R2 accel fails, restored → passes	reverted: R2 accel FAILED (+3.34 %), brake passes; restored: both pass
H6 canonical	full local validation	615/615

Unit: L5ReactionArm.* 3 tests (strut → R0 − penetration; penalty and sprung → Re).

Information (not gates):

• Same drive / brake command on the strut path now gives about 5 % more longitudinal acceleration in the R2 manoeuvre (accel a_g 1.142 → 1.196, brake −1.525 → −1.602 m/s²), because the same torque now meets a shorter arm. Penalty path and L1-L4 unchanged (H2).
• Acceleration-only residual term (decomposition of the 1st reply): −1.38 % before, −1.18 % after the arm fix — still present, cause unknown. Braking rest −0.11 % → −0.25 %.

Changed after measurement

• L5Strut.SettlesToStaticEquilibrium: upper bound cg_height + 5 mm → h_s + 5 mm, derived from the new definition (PO 1st verdict item 4), not from a measured value.
• Scope extension after the G1 measurement (user decision 11-07): the strut-path reaction arm fix above. H1-H6 were fixed before its measurement; gate and R2 expectation unchanged.
• L5Reaction.R2_* moved out of label known_failure into the integration suite; the vdsim_known_failure_tests executable, which held only these two, is removed.

Interfaces

• Public: VehicleParams::cg_height_ref, validate_cg_height_ref, cg_height_ref_is_sprung, l5_strut_body_height (params.hpp). The Python binding exposes the field.
• Not public: the spawn settle with an explicit body height is internal (core/src/spawn_settle_internal.hpp); the reaction-arm helper is internal (core/src/free_3d_reaction_internal.hpp); settle_spawn_on_ground() and IContactProvider are unchanged. No environment variables.
• cosim/realtime_server.cpp never enables the strut path, so it keeps the public settle.
• Aero: drag stays at cg_height above ground (four-key) or aero_drag_height; elements keep their z. Only the arm origin moves to h_s on the strut path.
• python/vdsim_render3d.py: optional trace geometry key body_cg_height_m; absent = unchanged. Smoke render once (no gate).

Documentation impact (W1)

CHANGELOG [0.8.2] (Breaking, Added, Fixed: reaction arm and R2 back in the verdict; the R2 known limitation is removed), params.hpp field comments, docs/theory/20_ld5_stunt.md (h, h_o, h_s, aero arm, strut reaction arm), docs/VALIDATION.md currency PENDING until the CI measurement, R2 exclusion lines removed.

Local verification

Canonical validation preset (ctest --preset validation --parallel 4) at ceb1c4c: 615/615 (main 595 + 20: C++ unit 9, C++ integration 10 incl. R2 ×2, Python 1). ctest -L known_failure: 0 tests. Doc audit at ceb1c4c: A1 version and A2 currency PENDING (intentional until the CI measurement) fail; main fails A1 only; links, nav, theory index, README, private strings pass.

CI measurement

PENDING: same-head complete successes ×3, currency confirmation, final run and wheels. Run 42379950076 (ceb1c4c) did not start — account billing lock; not counted, not retried. No merge, tag or publication performed.

🤖 Generated with Claude Code