Lane C — Timing & sync (DEV-C1..C4): outbound WallClockRef epoch, node discipline, sync groups, soak

[aperim-agent]

CLAIMED — actively being implemented by the Claude devices/display program (started 2026-06-10). Do not pick up slices from this lane; coordination happens in comments here.

Spec: ADR-M010 + display-out brief presentation model + wall-clock-sync outbound-epoch section. Depends on Lane A events + Lane B node mode.

• DEV-C1 outbound epoch (WallClockRef per program, WS publication, RTCP SR, EXT-X-PROGRAM-DATE-TIME) — branch ship/dev-c1-epoch
• DEV-C2 node presentation discipline (vblank frame chooser, skew telemetry) — branch ship/dev-c2-node-discipline
• DEV-C3 sync groups (weakest-member tier, measure/test-pattern, drift alarms) — branch ship/dev-c3-sync-groups
• DEV-C4 clock-layer telemetry + acceptance soak harness — branch ship/dev-c4-soak

🤖 Generated with Claude Code

[aperim-agent]

DEV-C1 (outbound epoch) implementation + adversarial review + fixes complete; PR open. Review HIGH — the PTP leg published TAI into UTC surfaces (37 s error, stepping on reference flaps) — fixed with an explicit timescale contract (ptp_utc_offset_s + residual guard + ADR amendments). PDT live-validated on a real run. 2836 tests green.

[aperim-agent]

✅ DEV-C1 merged (PR #113). The outbound epoch is in (UTC timescale contract, WS timing.status, live-validated HLS PDT, RTCP SR seam, step→EXT-X-DISCONTINUITY). DEV-C2 (node presentation discipline) unblocks once B5 lands; C3 (sync groups apply/measure) can start.

[aperim-agent]

Status (verified vs main@e77bbd0): DEV-C1 epoch (#113) merged. Remaining: DEV-C2 (node presentation discipline — depends on DEV-B5), DEV-C3 timing half (weakest-member tier/measure/drift alarms; the sync-groups CRUD shipped in DEV-A3), DEV-C4 (soak harness) — queued.

[aperim-agent]

▶️ DEV-C2 (node presentation discipline + skew telemetry) starting now — DEV-B5 merged (81d93cc) so both deps (B5+C1) are in. C3 (sync groups apply/measure) follows C2; C4 (soak) closes the lane.

[aperim-agent]

DEV-C2 — node presentation discipline + skew telemetry: PR opened → #137

#137 (base feat/foundation-buildout).

Makes display-node presentation frame-accurate by pure pull-side frame choice against the wall clock — never engine feedback. The node decodes into a bounded 2–3-frame queue and, at each flip, presents the frame whose wall_at(pts) + link_offset is nearest the predicted next vblank (from KMS flip timestamps); repeat-if-early, drop-if-late. Consumes the DEV-C1 outbound presentation epoch + the B5-recorded NodeTiming.link_offset_ms (now wired end-to-end; the recorded-not-consumed docs are updated). Flip-timestamp skew telemetry exported (last_flip_skew_ns / max_flip_skew_abs_ns). Inv #1/#10 hold by construction — pure pull-side choice, engine-side publish stays wait-free, nothing feeds back.

• Epoch-source boundary shipped honestly: discipline runs against the SharedEpoch seam — for a self-contained node run the runner wires the run's local outbound epoch (the same cell the C1 timing-status task writes). A future DEV-B6 WS-client fills the same seam with the controller's epoch; nothing else changes. Did not block on B6.
• Free-run-on-loss proven: epoch set once, never refreshed → every disciplined present runs on the held map; no epoch ever → undisciplined latest-wins. The output never waits for timing.
• Exact-integer ns throughout (never float fps): period via one rescale, distance/half-period in i128.

Gates: fmt clean · full-workspace clippy -D warnings clean · display-kms (cli+output) 40 suites/0 failed · node_pipeline (ffmpeg,display-kms) 4/4 · display_present 19/19 · display_present_props 8/8. (Full cargo test --workspace re-running under heavy shared-box contention; the comprehensive clippy already proves the whole tree compiles, and st2110_assemble — the one base test that flaked on cross-worktree E0463 — passes solo 8/8.)

C3 (sync-group cross-node skew measurement) unblocks on merge. No merge performed — human is the final approver.

[aperim-agent]

✅ DEV-C2 merged — PR #137, merge commit f3cefc4. Pull-side wall-clock frame chooser (exact-integer, repeat-if-early/drop-if-late), flip-anchored vblank prediction, skew telemetry; consumes link_offset_ms; free-runs on epoch loss. Review APPROVE.

▶️ DEV-C3 (sync groups: apply/measure/test-pattern + drift alarms) starting now. C4 (soak) follows.

[aperim-agent]

▶️ DEV-C3 PR #144 opened — sync groups: weakest-member achieved tier (never over-claimed), Class-1 per-member offset apply (engine cadence untouched), test-pattern action (burnt-in counter+flash), drift-alarm hysteresis on the conflating broadcaster, SPA surfacing. Gates green (1164 control/events/config tests, web clean). Review + CI in progress. C4 (acceptance soak) is the last lane-C slice.

[aperim-agent]

✅ DEV-C3 merged — PR #144, merge commit 1e6e3e0. Sync groups: weakest-member achieved tier, Class-1 offset apply, test-pattern, drift-alarm hysteresis. Review APPROVE.

▶️ DEV-C4 (clock-layer telemetry + soak harness + deployment guidance) starting now — the scripted/documented/telemetry deliverable (servo offset/ppb export + pass thresholds, 2-node soak harness, inv-#1 chaos extension, ptp4l/chrony configs). Soak EXECUTION is hardware-gated (needs ≥2 real nodes; t630 + the incoming Pi 4) and rides the hardware-validation phase. C4 is the last lane-C build slice.

[aperim-agent]

▶️ DEV-C4 PR #149 opened — clock-servo telemetry (offset/ppb + audio servo gauges, ADR-M010 thresholds), the tested soak analyzer (multiview-telemetry::soak: p99 offset + invariant-#1 chaos cadence check), cargo xtask soak-report, scripts/soak-acceptance.sh (shellcheck-clean, --dry-run self-test), and deploy/sync-clock.md (ptp4l/chrony). Gates green (109 telemetry+xtask tests, clippy/fmt/deny/shellcheck/inclusive-language). Lane C build COMPLETE pending merge; the physical 24h/camera soak execution rides the hardware phase.

[aperim-agent]

✅ DEV-C4 merged — PR #149, merge commit a6e0e42. Clock-servo telemetry + tested soak analyzer + xtask soak-report + scripts/soak-acceptance.sh (--dry-run verified) + deploy/sync-clock.md. LANE C COMPLETE (C1–C4). The physical 24h/camera soak execution rides the hardware-validation phase. Note: independent review was infra-blocked (subagent outage) → self-reviewed + merged with the blocker named for retro-review.