F11: demote web_ui's per-gap keyframe request (measured on air, default off)

[dorkmo]

Implements the F11 candidate queued by the F10 acceptance (PR #89, RESULTS §5), following the same gate → measure on air → flip protocol as F9/F10.

Why

The gap-tolerant canvas requested a keyframe on every base_seq gap — including a single lost delta frame. In the F10 surgical window (the control data for this A/B), three gap events produced three requests and two granted keyframe trains, each a multi-frame cost at the 243 B/frame budget, with the 10 s throttle as the only damper. F10's 0x6C stale-tile path now detects and repairs exactly that damage tile-by-tile, making the per-gap keyframe largely redundant.

What changed

• CanvasUpdate carries structured causes (seq_gap, tile_error) so the policy layer never string-matches reason text; a gap and a tile error can co-occur and reason only keeps whichever fired first.
• web_ui suppresses the req_keyframe publish only for pure-gap requests under LIFETRAC_KF_ON_SEQ_GAP=0. Cold start, grid mismatch, and tile-decode errors always pass through. Each suppression logs a countable kf-on-gap suppressed (F11) line (at WARNING — uvicorn leaves the root logger at its default level, which silently dropped the first attempt at INFO; caught on the bench when three real suppressions produced zero log lines).
• Default flipped to OFF in the second commit, after the on-air A/B passed.

On-air A/B (bench-evidence/F11_kf_on_gap_2026-08-02/RESULTS.md)

Same operating point as the F10 acceptance; gate off. Six gap events — four induced (3 s rx stops) plus two natural single-frame air losses, the exact event class the control paid keyframes for — produced six suppressions and zero keyframe requests, with the canvas fully healthy: short-gap losses re-swept below the 30 s stale horizon (no repair even needed), gap stragglers repaired by 0x6C in 1–2 report periods. The ungated cold-start path was verified live: a canvas-cold web_ui restart mid-stream published "delta arrived before any keyframe" within 3 s and was granted a keyframe. Every 0x60 radiation in the session log is accounted for (2 stream-acquisition + 2 cold-start, none gap-driven).

Tests

10 new in test_kf_on_seq_gap.py: structured-cause semantics (gap / tile-error / both / cold start), gate-off suppression, gate-on passthrough, keep-paths under gate-off, and the default pinned to OFF with the evidence pointer. Full suite: 1103 passed / 2446 subtests.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR implements F11 by adding structured keyframe-request causes to the image canvas update path and using them in web_ui to suppress pure base-seq-gap keyframe requests by default (while preserving cold-start, grid-mismatch, and tile-decode-error keyframe behavior). It also adds on-air bench evidence artifacts and new tests to lock in the gating semantics and default.

Changes:

• Add structured seq_gap / tile_error flags to CanvasUpdate so policy can gate without parsing reason strings.
• In web_ui, suppress req_keyframe publishes for pure seq-gap requests when LIFETRAC_KF_ON_SEQ_GAP=0 (default), and emit a warning log line per suppression.
• Add bench evidence artifacts and a dedicated test suite covering structured-cause semantics and gate behavior, and update the controller TODO entry to DONE/verified.

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
LifeTrac-v25/DESIGN-CONTROLLER/TODO.md	Marks F11 as completed and records the on-air verification summary and evidence pointer.
LifeTrac-v25/DESIGN-CONTROLLER/bench-evidence/F11_kf_on_gap_2026-08-02/tap_base.log	Adds broker tap artifact showing observed tile_stale and req_keyframe activity during the A/B.
LifeTrac-v25/DESIGN-CONTROLLER/bench-evidence/F11_kf_on_gap_2026-08-02/RESULTS.md	Adds written on-air A/B results and decision to flip the default to OFF.
LifeTrac-v25/DESIGN-CONTROLLER/base_station/web_ui.py	Implements the gate and suppression logging for pure seq-gap keyframe requests.
LifeTrac-v25/DESIGN-CONTROLLER/base_station/tests/test_kf_on_seq_gap.py	Adds tests for structured causes, suppression/passthrough behavior, and the default setting.
LifeTrac-v25/DESIGN-CONTROLLER/base_station/image_pipeline/canvas.py	Adds structured cause flags to CanvasUpdate and sets them on seq gaps and tile transcode/apply errors.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.