[Onlyti]

Integration of the three independent feature lines onto main (PO Q13, briefing §14).
main and the feature lines had diverged in both directions since 34db50c, so the
resolution happens here rather than on main.

What is merged

#	line	commits	brings
A	feat/scene-sensors-parser	68	trace container, headless BEV renderer, comms/scene parsers, AC importer M1, D0–D2 contact work
B	feat/render-preset-overview	47	render presets, panel selection, layout violation check
C	docs/validation-canonical-config	50	validation preset, currency gate, pinned CMake, GUI v3 test switch

chore/repo-hygiene is already contained in line A and needs no separate merge.

Conflict hunk classification

class	hunks	how it was resolved
physical baseline (ISO target, benchmark reference, analytical comparison)	0	nothing to resolve — no such hunk conflicted
measurement metadata (test counts, currency block, evidence logs, dates)	20	no number taken from either side; the currency block is PENDING
code, same feature under two names	7	main's generalisation kept (SessionKind over direct_control_path, skip_kinematic_blend() over vp_.plant_path)
code, genuinely different features on the same lines	8	composed, not chosen (contact-frame work + T6 instruments + preset BEV toggles)
build/registration surface	6	union of both test and target lists
historical dated evidence (docs/evidence/review/2026-06-21/**)	15	main-lineage record kept; the thesis-branch copy of the same filename documents a tree not in this history
additive documentation rows	11	union

Total 67 hunks over 21 files across the three merges.

Test counts

docs/VALIDATION.md carries no number. The last canonical figure was measured on a tree
that no longer exists, so it is void rather than stale, and the currency block sits in an
explicit PENDING state with the reason recorded. The gate learned PENDING as a state
(it demands a pending: line, and the duplicate-count scan over docs/ stays active) so
that the block did not have to be deleted or bypassed to get through the merge.

Post-merge fixes in this PR

1. The build directory name was hardcoded as build in the ctest PYTHONPATH and in
   tools/module_workshop.py, so the canonical preset — which builds into
   build-validation — only worked next to a stale build/. Both now follow
   ${CMAKE_BINARY_DIR}.
2. D1ContactFrame.InvalidContactResetsTireTransientBeforeRecontact rebuilt a level from
   a State snapshot and required an exact match. That stopped being exact at L4 when
   main added KC bushing compliance, whose history lives in the axle DAE rather than in
   State. The check now measures that reconstruction residual on a control run that never
   loses contact and requires the airborne run to be no worse — no tolerance was fitted to
   the observed failure. L3 remains exact.
3. tests/fixtures/d0_flat_contact_l4_baseline.csv re-captured, with the cause measured
   rather than assumed: suppressing compliance_targets_rad() makes this tree reproduce
   the pre-merge L4 fixture bit-for-bit, so main 9a40c1b moved it and the contact-frame
   work did not. The L3 fixture is untouched and still bit-identical.
4. The validation CI job built the Python bindings with no pybind11 and ran the
   catalog/plant/trace tests with no numpy, pyyaml or matplotlib. Both jobs now install the
   same pinned set.

Local result

Clean configure + build + ctest --preset validation on ailab-12: 485 registered,
485 passed, 0 failed. That number is deliberately not written into the document — the
canonical measurement is the CI validation job on this PR.

🤖 Generated with Claude Code