# U7 review and diagnostics evidence

Implemented session review with four synchronized three-axis Swift Charts timelines, shared cursor/domain, preserved gap/phase/marker/cue overlays, full-data sampling summaries, per-transition metrics, complete-leg distributions, filter recomputation and external derived export. Diagnostics records stationary and controlled drill phase events with optional rider calibration and all three capture modes.

Isolated portable snapshot: `/tmp/rideform-u7-core`; command `DEVELOPER_DIR=/Applications/Xcode-beta.app/Contents/Developer swift test --disable-sandbox --package-path /tmp/rideform-u7-core`; log `/tmp/rideform-u7-core.log`: 36 tests, zero failures. Four new cases verify known bias/SD/RMS/drift, movement exclusion, unavailable/insufficient data, independent raw-only streams without calibration, and display extrema/gap edges. Initial RED was compiler complexity in two test fixture expressions, corrected by explicit typed intermediate values. This is worker evidence; the main coordinator owns authoritative current-source runs.

SessionClock race regression: original independent TSAN harness `/tmp/rideform-u5-clock-race.swift` and log reproduced an array deallocation abort (exit 134). Mapping validation/reset and all mapping reads now share a lock. Equivalent 100,000 concurrent validation/read replay `/tmp/rideform-u7-clock-race.swift`, compiled with `swiftc -module-cache-path /tmp/rideform-u7-module-cache -sanitize=thread`, completed without diagnostics (`/tmp/rideform-u7-clock-race.log`). App SessionClockTests adds concurrent-copy coverage. This is a distinct U5 reliability correction carried with U7 integration.

Privacy manifest declares both 35F9.1 and 8FFB.1; API purpose is elapsed in-app timing and exported timestamps for in-app capture/events. See sensor-api-decisions.md. Compiled manifest verification belongs to the authoritative app build.

Pending coordinator evidence: authoritative app/UI build/test, rendered chart/UI review and export configuration equality. Pending human evidence: iPhone 30-second stationary/noise/drift trials, three matched fused/combined mode comparisons, controlled slow/rapid movement/sign checks, and roughly 50 garage transitions. Synthetic data is never hardware evidence.

Coordinator initial compile findings corrected: SwiftUI's `Transition` type conflicted with the core model, resolved by `RideFormCore.Transition`; optional derivative-window display uses zero when filtering has no window. Full review fixture is enabled only by simulator arguments `--uitest-motion --uitest-review`, saves 601 measured fixture rows with a known gap, three-axis signals, an outbound leg and two events. UI tests cover all four timelines, filter recomputation/reopening, early-stop uncalibrated diagnostics, and a real 30-second software timer followed by all 12 instructions (15 saved events). New app integration tests cover all three diagnostic modes through the simulator adapter and exact exported analysis/configuration equality with unchanged master identity. Current authoritative results are recorded by the coordinator; these test sources alone are not a claim of passing rendered or hardware verification.

## Final coordinator verification

See `calibration-desktop-update.md` for the final 41-test core and 29-test app runs, independent analysis verification, completed source review, chart workflow verification and build 2 installation. Earlier pending software checks above describe the initial handoff, not the final state. Hardware acceptance remains pending.