Ground Study + Rate of Closure Suites: regional execution, camera tracking, job workspaces & static web distribution (consolidates 32 closed PRs, supersedes #4409/#4410)

[dieterolson]

Supersedes #4409, #4410

One carrier for both suites. #4409 and #4410 consolidated the same already-closed
#4332-#4402 range: they shared ~360 commits, diverged by only 49, and neither was an
ancestor of the other, so each was re-running the full CI matrix over largely identical
content. Both were red. This branch is consolidated/rate-of-closure-batch with
consolidated/ground-study-batch merged --no-ff on top, with every blocking defect fixed.

quality-gate failed on both PRs at Reject Raw Merge Conflict Markers, despite head
commits on each branch claiming the markers were resolved. That step sits near the top of the
job, so it short-circuited everything after it — which is why fixing it exposed two further
blocking failures in the same content that nobody had ever seen.

1. Raw conflict markers — 5 files

File	Markers	What the conflict actually was
SPEC.md	73 open / 104 close — unbalanced	Identity version rows + dated §12 Change Log entries
AGENT_HANDOFF.md	8 / 8	Prepended handoff entries
docs/development/RATE_OF_CLOSURE_CAMPAIGN_HANDOFF.md	6 / 6	Prepended handoff entries
src/rate_of_closure/AGENT_HANDOFF.md	6 / 6	Prepended handoff entries
.secrets.baseline	15 / 15	detect-secrets entries (live merge conflict)

No source file carried a marker.

Every conflict had one cause: ~25 stacked feature branches each prepended a dated entry
just below a section heading and bumped the same version rows, so every merge collided at
exactly that spot. Because each side is an independent entry rather than a competing edit, the
correct resolution is the union of all sides, with exact-duplicate blocks collapsed.

SPEC.md needed a tolerant parser: prior "resolutions" had deleted markers arbitrarily,
leaving 14 orphaned ======= and 31 orphaned >>>>>>> lines with no matching <<<<<<<.
It was verified line-for-line afterwards that no original content was dropped from any of
the four documents. The single-valued Identity rows collapse from 33 / 33 / 6 duplicates to
one each, at the highest observed version (1.14.94) plus a bump to 1.14.95.

2. .secrets.baseline had Windows path keys

The baseline inherited from #4410 was regenerated on Windows and committed with backslash
path keys (30 of 33), replacing the POSIX keys on main and #4409. No Linux CI scan can match
a docs\SECRETS_MANAGEMENT.md key, so the detect-secrets job would have reported every real
audited entry as a newly added secret. Regenerated with CI's exact invocation and POSIX keys;
the result is a superset of both main's and #4409's entries and is idempotent under rescan.

3. ruff-format version drift — 100 files

The branches were formatted with a ruff other than the pinned ruff==0.14.10, which reformatted
100 changed Python files into output the pinned formatter rejects — including files with no
connection to this work (launch.py, scripts/bump_vendor_pin.py, src/shared/**). This was
not pre-existing debt: main's versions of all 91 of those files that exist there pass
ruff format --check under 0.14.10 cleanly. Re-running the pinned formatter restores 85 of
those 91 to byte-identical-to-main content; the other 6 keep genuine changes.

4. A rewrite that contradicted its own suppression

tests/unit/sidekick/agent/test_action_audit.py had an automated UP017 fix rewrite
tzinfo=timezone.utc to tzinfo=UTC and import UTC from datetime, while leaving in place
the very suppression forbidding it: # noqa: UP017 - Python 3.10 CI lacks datetime.UTC. The
tests matrix still includes Python 3.10, where datetime.UTC does not exist. Reverted. The
rest of the branch was swept: every other from datetime import UTC sits inside a guarded
sys.version_info compatibility shim.

5. Accidental tree entries, absent from main

• .codex-worktrees/* — six mode-160000 gitlinks pointing at stale local Codex worktrees
  from June (PRs Fix Swingset legend plot clearance #3602-[codex] Seed transfer entropy permutation tests #3784, long since merged). There is no .gitmodules, so they are
  undeclared, and the commits they reference exist only in one local clone and are on no
  remote branch — a fresh git clone would get six gitlinks to objects the server does not
  have. This is the "unregistered historical gitlink" that the campaign handoff records as
  having broken actions/checkout's cleanup and forced a credential-free manual fetch
  workaround.
• batch_plan.txt, non_draft_prs.json, pr_worker_final_results.json,
  specified_summary.txt — raw gh API dumps and PR-batching notes written into the
  repository root by the prior consolidation agent. CRLF-encoded, and they trip
  git diff --check.

6. 138 mypy errors in 42 files, previously masked

With the markers fixed, quality-gate reached its Type Check step for the first time and
reported 138 errors across 42 files — all 42 of them new files added by this work, none
pre-existing on main. None came from the merge: exactly 5 files needed hand resolution
(§1 above) and not one was Python, so no call site was resolved differently between the two
branches.

• arg-type (84) — two dominant patterns. Indexing a 2-D array (theta[i, 1], y[0])
  yields a scalar at runtime but numpy's stubs type it as ndarray, while the callee genuinely
  needs a scalar because it calls math.cos; those sites now hoist float(...) locals. The rest
  are dict literals inferred as dict[str, object] and then splatted with **kwargs, which
  defeats keyword type-checking entirely — fixed with TypedDict annotations on the literals
  rather than by loosening any callee signature.
• no-any-return (26) — bound to an annotated local and returned, rather than cast.
• redundant-cast (18) — dead cast(...) wrappers removed where narrowing already applied,
  each replaced with an annotated local so the fix also holds where the inner expression
  degrades to Any, with unused typing.cast imports pruned so F401 stays clean.
• var-annotated (8), has-type (1), return-value (1) — explicit annotations, one
  declaration-only class annotation, and one documented cast where the parameter must stay
  object to remain contravariant with the CompleteBatchExecution.validator signature. A cast
  is a runtime no-op, so the golden-fixture-pinned output bytes are untouched.

7. Two more masked steps

• Six empty tests/__init__.py package markers under src/shared/python/swing_sim/** failed
  Changed Test Assertion Check; allowlisted alongside the equivalent ground/tests marker
  already listed there.
• A cosmetic reformat of tests/architecture/test_sidekick_external_imports_3316.py had
  enlisted an unrelated repo-wide import-boundary guard into this PR's blocking test set, and
  that guard fails against main's own tree — the src/shared/python/sidekick/ tree here is
  byte-identical to main, and the guard is a pure AST walk over src/ comparing dotted module
  names, so it is neither path- nor platform-dependent. main's exact bytes are restored so the
  file leaves the changed set. The guard's own failure is tracked as separate follow-up work.

Content integrity

• 1341 files changed; 670 under src/rate_of_closure, 111 under tests/rate_of_closure.
• All 848 changed Python files parse (AST); zero syntax errors, no truncation. The only empty
  files are 5 __init__.py package markers.
• Of 218 modified Python files, only 2 lost a symbol present on main, and both are the
  documented module-budget extraction rather than deletion — the
  src/rotation_converter/ui/pyqt6/main_window.py helpers now live in plot_helpers.py, and
  motion_tabs.py helpers in motion_helpers.py. Both new modules define and export the moved
  names.

Gate evidence, pinned toolchain

Reproduced against CI's own selections — ruff==0.14.10, mypy==1.13.0, Python 3.12, numpy
pinned <2.4 as requirements.txt requires:

• ruff check — clean (731 gate files)
• ruff format --check — clean (731 gate files)
• mypy — Success: no issues found in 465 source files
• CI's own 216 changed test files, run with CI's flags — 2465 passed, 11 skipped, 0 failed
• Both the 60% Coverage Policy Gate and the Sidekick Per-File Coverage Gate provably skip on
  this branch: coverage_gate_required only turns true when a tracked_packages path changes,
  and this branch touches none.

Known non-blocking failure

file-size-budget fails: three new GUI modules exceed the 500-LOC budget and are not in
scripts/monolith_baseline.txt — torque_profile_panel.py (612),
regional_ground_execution_workspace.py (572), rate_of_closure/ui/pyqt6/main_window.py (541).
It is not a required check, and the campaign handoffs already record the 612-line file as known
debt, so it is reported rather than refactored inside a consolidation.

Already-closed PRs this carries

All 32 are closed unmerged, so their work exists only on these consolidation branches.

From #4409 — Ground Study (21): #4332, #4335, #4350, #4351, #4352, #4353, #4354, #4355,
#4356, #4357, #4359, #4360, #4361, #4365, #4366, #4367, #4370, #4372, #4373, #4374, #4375

From #4410 — Rate of Closure (11): #4339, #4342, #4358, #4362, #4363, #4364, #4368, #4371,
#4376, #4388, #4402

Issues referenced

All open, and each is annotated with this PR. None are closed by this merge — the campaign
handoffs list remaining work under every one of them, and this PR deliberately carries no
closing keywords.

• Ground variation: bind a qualified execution job and browser-capable executor #4369 — Ground variation: qualified execution job and browser-capable execution
• Ground Model: total-distance metrics, targets, solvers, and variation #4273 — Ground Model: total-distance metrics, targets, solvers, and variation
• Ground Model: skid-to-roll dynamics on slopes and regional surfaces #4271 — Ground Model: skid-to-roll dynamics on slopes and regional surfaces
• Ground Model: validated 3D impact impulse and repeated bounce #4270 — Ground Model: validated 3D impact impulse and repeated bounce
• Ground Model: matched PyQt6 and React controls and 3D playback #4274 — Ground Model: matched PyQt6 and React controls and 3D playback
• Rate UI: track clubhead camera and add canonical 3D snap views #4284 — Rate UI: track clubhead camera and canonical 3D snap views
• Rate of Closure: deterministic web asset manifest and static-inspection runtime #4378 — Rate of Closure: deterministic web asset manifest and static-inspection surface
• [Ball Flight] Add a spatial target editor and 3D target geometry #4192 — [Ball Flight] Spatial target editor and 3D target geometry

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e9f01d99c1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Relax the wind golden tolerance

On Python 3.12.13 with pytest 9.0.3, the shear_turbulence case fails here: Python returns 9.786440272809793, while the cross-client fixture expects 9.7864402728063, a normal 3.49e-12 cross-runtime sin/libm difference that exceeds this 1e-12 tolerance. The changed-test lane in .github/workflows/ci-standard.yml collects this src/**/tests/test_*.py file and runs it in the Python matrix, so the commit cannot pass that lane unless the comparison allows the observed floating-point drift.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Validate timesteps in release builds

When the release-built PyO3 or WASM step API receives dt <= 0 or a non-finite timestep, this debug_assert! is compiled out, so zero silently returns an unchanged state, negative values integrate backward, and NaN propagates through the result instead of returning the advertised error. simulate calls the same function repeatedly, and the new workflow explicitly builds the wheel with --release; make this a runtime precondition that returns Err rather than a debug-only invariant.

AGENTS.md reference: AGENTS.md:L129-L133

Useful? React with 👍 / 👎.