ASHLAR HDMI DIAGNOSTIC PACKAGE

Start with FORUM_SUMMARY.txt for the short explanation, then read
DIAGNOSTIC_REPORT.md for the measurements, limitations and proposed tests.

IMPORTANT UPDATE
The tester subsequently reproduced the failure on Stable 12.2.1 as well.
Both Stable and the Intel-10b support test image reportedly work when automatic
refresh-rate adjustment is disabled and fail when it is enabled. The original
"Stable works, Nightly fails" assumption must not be treated as established.
Starting Kodi manually at the movie's refresh rate has not yet been reported.

CONTENTS
- FORUM_SUMMARY.txt: forum-friendly English summary.
- DIAGNOSTIC_REPORT.md: full current English diagnostic report.
- logs/: Kodi, kernel/i915, DRM/display snapshots and decoded EDID information.
- measurements/: timestamped SCDC readings, register snapshots, settings and
  observational JSON records from the original test session.
- audit/: pinned source receipts and offline function/register comparisons.
- tools/read_scdc.py: the read-only collector used for the SCDC measurements.
- MANIFEST.json: file inventory, source/export hashes and redaction counts.

HOW TO READ THE CAPTURES
The hardware captures are from 2 November 2026. The later autorefresh findings
are tester reports; no new matching Stable/autorefresh-off raw capture is
included. File prefixes identify separate experiments. A before/after kernel
dump includes older boot messages, so do not count repeated lines as new events.

SCDC timestamps ending in Z are UTC. Kodi timestamps in these captures are
Europe/Paris local time, UTC+1 on the test date. Kernel timestamps are seconds
since boot. Do not correlate them without converting the clock domains.

The logs are not a physical HDMI capture. A clean Kodi log or a sampled lock
flag does not establish that the television never lost the picture. CED values
are sampled register contents, not a measured bit-error rate or frame-drop count.

PRIVACY AND INTEGRITY
Exported copies mask network URLs/addresses, common access-token/credential
patterns, email addresses and Windows/home-directory user names. Raw originals
are retained locally and are not changed. This is targeted redaction, not a
claim of complete anonymisation: hardware model, BIOS, driver versions, EDID,
timings and playback diagnostics are intentionally retained. Review before
posting publicly. Source/export hashes are recorded separately in MANIFEST.json.
Invalid UTF-8 bytes in one kernel dump are represented as hexadecimal escapes
in the exported text; this conversion is also marked in the manifest.

No enrollment keys, SSH configuration, support credentials, screenshots, private
remote-control scripts, firmware/image binaries or full upstream source trees
are included. All explanatory material is in English. Technical log messages
and numeric measurements are preserved except for the listed redactions.

No machine was contacted, no firmware/image was installed, and no GitHub release
was changed while preparing this package.
