added CAP-050 and CAP-051

Author: tedsluis

ai-sessions/0019_CAPTURE_PROMPT_2026_09_14.md

@@ -0,0 +1,57 @@
+# 0019_CAPTURE_RESULT_2026_09_14.md — Video-only analysis and CAP-050-EVENT-NOTES.md update for CAP-050 (Group AG repeat)
+
+**Number:** 0019
+**Category:** CAPTURE
+**Date:** 2026-09-14
+**Title:** Video-only analysis of `CAP-050-recording.mp4` and update of `CAP-050-EVENT-NOTES.md`
+**Status:** complete
+
+## Phase status
+
+| Phase | Status | Summary |
+|---|---|---|
+| 0 — Setup | done | Mandatory reading order completed (AGENTS.md, PROJECT.md, PROJECT_RULES.md, DECISIONS.md in full; CAPTURE_BLUETOOTH_HCI_SNOOP.md Group AG section; AI_SESSION_LOG_PROCEDURE.md; ai-sessions/INDEX.md; CAP-050-EVENT-NOTES.md skeleton). Confirmed folder contents via `ls -la`: `CAP-050-btsnoop_hci.log`, `CAP-050-EVENT-NOTES.md`, `CAP-050-recording.mp4`. `ffprobe`: duration `748.795896s` (~12m28.8s), 1280x720, 30fps. |
+| 1 — Watch video, rewrite EVENT-NOTES | done | Reviewed `CAP-050-recording.mp4` in full (748.80s) via a uniform 5s-interval baseline (150 frames) plus `ffmpeg` scene-change-detection frames (99 additional moments of visual change), covering the entire duration. Rewrote `CAP-050-EVENT-NOTES.md`'s Log Metadata and Event Timeline with verified findings, sourced only from the video. Renamed the capture folder via `git mv` to `captures/CAP-050-2026-09-14_21-01-01_21-13-30-Group_AG` (video overlay start/end times). |
+| 2 — Wrap-up | done | Ran `python3 scripts/lint_docs.py` and `./scripts/ensure_footers.py`; see Notes below. |
+
+## Notes
+
+This is a first run (no prior `0019_CAPTURE_RESULT_2026_09_14.md` existed). Executed via Claude Code (not Gemini CLI as the prompt file's boilerplate suggests — functionally equivalent execution of the same phased prompt).
+
+`lint_docs.py` was run and produces its usual large baseline of pre-existing dead-filename-reference
+findings across the repo (analysis-frame PNGs referenced from various `CAP-NNN-FINDINGS.md`/
+`EVENT-NOTES.md` files, and `CAP-050-FINDINGS.md`/`CAP-051-FINDINGS.md` references from files that
+intentionally don't create those files yet) — none of these are new. The rename from
+`captures/CAP-050-yyyy-MM-dd_HH-mm-ss_HH-mm-ss-Group_AG` to
+`captures/CAP-050-2026-09-14_21-01-01_21-13-30-Group_AG` did leave two now-dead references to the old
+placeholder path: `_sidebar.md` (auto-regenerated by `.github/workflows/update-docs-sidebar.yml`, not
+hand-edited) and `ai-sessions/0005_MAINTENANCE_RESULT_2026_09_09.md`/`ai-sessions/0019_CAPTURE_PROMPT_2026_09_14.md`
+(both describe the state of the repo *before* this session's rename, matching this project's
+convention — established by every other renamed capture's own earlier `ai-sessions/NNNN_*_PROMPT_*.md`
+entries, e.g. `0016`'s references to `CAP-043-yyyy-MM-dd...`/`CAP-044-yyyy-MM-dd...` — that a
+`PROMPT`/older `RESULT` file is a historical record of what was asked/known at the time, not rewritten
+after the fact). `CAP-050-EVENT-NOTES.md` itself contains no reference to the old placeholder path.
+`ensure_footers.py` reported all footers already up to date.
+
+**Verified reconnect-cycle sequence:** the video shows the maintainer's account was correct but
+understated — rather than ~10–15 clean, isolated dock/undock cycles, the footage shows near-continuous
+handling of the case and both earbuds across the full ~12.5-minute session, producing well over 20
+Connect/Disconnect transitions in the official app. Three of these are clearly and repeatedly
+identified as "Left docked, Right out" at the moment of reconnect (`~21:01:58`, `~21:04:52`,
+`~21:13:09`); one earlier moment clearly shows "both docked" at a disconnect (`~21:01:44`). Several
+other transitions have an ambiguous or lower-confidence dock-state read (flagged explicitly in
+`CAP-050-EVENT-NOTES.md`'s Event Timeline), and one long middle window (`~00:04:00`–`00:11:41`)
+contains many further cycles that were not individually attributed to a specific dock state at this
+review's frame density — left as an explicit open item for the future log-analysis session rather than
+guessed. The case lid never closes in this video.
+
+**Uncommitted changes** (nothing has been committed — per project convention, only on explicit
+request): renamed capture folder (`git mv`) and its rewritten `CAP-050-EVENT-NOTES.md`; new
+`ai-sessions/0019_CAPTURE_RESULT_2026_09_14.md`; updated `ai-sessions/INDEX.md` row `0019` (status →
+`complete`; note row `0020` was already updated to `complete` by a separate session before this one
+ran). `CAPTURE_BLUETOOTH_HCI_SNOOP.md`'s `CAP-050` status and `id_registry.csv` were deliberately left
+untouched (`planned`), as instructed. `.log` analysis and `CAP-050-FINDINGS.md` remain a separate,
+future, out-of-scope task.
+
+---
+https://github.com/tedsluis/opencontrolpixelbudspro2/blob/main/ai-sessions/0019_CAPTURE_RESULT_2026_09_14.md - https://tedsluis.github.io/opencontrolpixelbudspro2/ai-sessions/0019_CAPTURE_RESULT_2026_09_14