docs(rate-of-closure): restore the handoff doc to the current-state policy

SPEC.md

@@ -4959,6 +4959,7 @@ Active development with stable core, continuous tool expansion, and web API in p
 
 | Date | Version | Changes |
 | ---- | ------- | ------- |
+| 2026-08-17 | 1.17.18 | docs(rate-of-closure): bring `src/rate_of_closure/AGENT_HANDOFF.md` back under the `CLAUDE.md` handoff policy. The file had grown to 2,205 lines across 140 dated entries — the same append-only drift the root handoff already recorded and corrected for itself at 2,708 lines. Those entries move verbatim to `docs/agent_handoff_archive/2026-08_rate_of_closure_handoff_log.md`, matching the existing root-log archive convention, and the live document is rewritten as 103 lines of current state: what the tool is, where the PyQt6 and React surfaces and the `swing_sim` physics packages live, why PR #4466 cannot be merged by any strategy (measured — `-X theirs` gives 47 failures and 40 errors, `-X ours` gives 19 collection errors), the four files where the source branch is *older* than `main` and would silently revert shipped work, what remains of #4466 by area, and the local-environment traps that cost real debugging time (mypy 1.13 crashing on Python 3.13 for multi-file sets, the two-tier `tools_core` capability, and PowerShell rewrites normalising `SPEC.md` to CRLF). No source or test behaviour changes. |
 | 2026-08-16 | 1.17.15 | feat(swing-sim, rate-of-closure, #4103): land the `swing_sim.flight` slice of `consolidated/rate-closure-remainder-2026-08-13` with its React counterpart, so the Python and TypeScript sides of the ball-flight contract move together. Adds 21 `swing_sim/flight/**` modules (capability observation/evaluator, ground transfer and bounce execution, regional ground pipeline, surface simulation, spin-axis convention, cancellation) plus `spinAxisConvention.ts`, `capabilityFlightEvaluator.ts`, and the `capability_flight_evaluator_parity_v1` fixture. Unifies the spin-axis tilt convention on fade/right-positive: `spin_axis_tilt` becomes `positive_right` with `atan2(-omega_y,omega_z)` in both `result_catalog_data.py` and `ballFlightMetricContract.ts`, `deliveryDiagnostics` now reuses the shared `spinAxisTiltDeg` helper instead of an inlined formula, and `ball_flight_metrics_golden_v1.json` is regenerated for the new sign. Repairs a ground-crossing defect in `web/src/model/flight.ts`: the interpolation guard tested only the next point's height, so a descending launch — which starts at height 0 and is skipped on the first step by `t > dt` — interpolated from an already-below-ground point, producing a negative fraction and a trajectory time before zero. The metric contract then rejected it, surfacing a descending launch as a `RangeError` instead of the nonconverged result it is. The guard now also requires the previous point to be above ground. `tests/test_wind.py` deliberately keeps `main`'s 1e-9 parity tolerance; the source branch still carries the 1e-12 value that fails on Linux (see #4513). Verified: 1,163 `swing_sim` tests and 1,095 React tests pass, `tsc --noEmit` and `eslint` are clean. |
 | 2026-08-16 | 1.17.14 | fix(ci, tests): enforce the two-tier Python floor instead of letting the 3.10 lane run code that requires 3.11. The root distribution declares `requires-python = ">=3.11"` while ten sub-packages and Rust crates declare `>=3.10` and ship 3.10 wheels, so the 3.10 matrix lane is intentional — but it was running the whole suite, including root-package code. That produced failures that looked like defects and were not: a bare `tomllib` import aborting collection, and an `asyncio.wait_for` timeout in the p1am e-stop shutdown test whose cancellation semantics changed in 3.11. `conftest.py` now reads each package's own `requires-python` (regex-parsed, since `tomllib` is unavailable on the interpreter the guard must run on) and skips collection of anything above the running interpreter; it is a strict no-op on 3.11+, verified by identical collection counts. `CLAUDE.md` previously advertised a flat “Python 3.10+” that the root distribution rejects and now states the real two-tier contract. New `tests/test_python_version_contract.py` locks `requires-python`, the mypy target, the classifiers, the CI matrix, and `CLAUDE.md` together so the declarations cannot drift apart silently again. The `ci-standard` matrix drops to `["3.11", "3.12"]`: that job runs the root-package suite (`core_tests` is entirely `tests/**` and `src/shared/python/**`), so a 3.10 lane there collected nothing once the floor guard was correct. The ten sub-packages declaring >=3.10 are gated on 3.10 by their own maturin build + parity workflows, which is verified by a new contract test. |
 | 2026-08-16 | 1.17.13 | feat(swing-sim, #4103): land the `swing_sim.ground` skid/roll/bounce module as a self-contained slice of `consolidated/rate-closure-remainder-2026-08-13` against current `main`, rather than merging that branch wholesale. Adds 92 `swing_sim/ground/**` files, `swing_sim/canonical_numeric_json.py`, and the 10 shared Python/TypeScript ground golden fixtures under `src/rate_of_closure/web/src/model/__fixtures__/`. Ground's three other dependencies (`flight/result_metrics`, `solver/spatial_targets`, `solver/target_serialization`) were already byte-identical on `main`, so the module needed no other source changes. Repairs six tests that fail on the source branch itself: four `test_skid_roll_passivity` cases never passed `SurfaceRun`'s `active_surface` argument (added with regional-surface support) and raised `TypeError` before asserting any passivity property; `test_bounce_cancellation_is_typed_and_retains_request_identity` hardcoded a termination time and elapsed span that violate `RepeatedBounceResult`'s chronology invariants; and `ground/__init__.py` eagerly imported `to_ground_model_result`, leaking the explicitly unqualified compatibility adapter into the package namespace despite its exclusion from both `__all__` and the lazy-import map. 321 ground tests pass. |