OrbitTv/docs/VERIFICATION-0.108.4.md

0.108.4: Stage D evidence reader (phase 1, read only), built on the installed 0.108.3

Paul approved phase 1 on 4 November 2026 at 23:01. This is 0.108.3 plus one diagnostic capability. It changes no schedule, pool or library item. The same change is on main.

What it does:

• Trigger: am start ... --ez orbit_read_evidence true. The next background round reads Emby for every film and series on any channel's list, plus every library title premiered in the last 60 days.
• Fields: each item's description, tags, genres, studios, up to 6 leads, up to 3 directors, official and community ratings, taglines, production countries and provider IDs.
• How it reads: 40 items per request, at most 4,000 items a run.
• Where it goes: each item is appended as one JSON line to no_backup/schedule/evidence-0.jsonl. An item already there is skipped, so a run can stop and carry on.
• Logging: one line of counts only: asked, missing, read, failed batches.
• Missing isn't negative: a field the provider doesn't have is stored as missing (null), never as an empty value that would read as negative evidence.

Checks

• 489 unit tests run, 0 failures; lint has no errors. EvidenceReaderTest adds 3 tests: it keeps what the provider gave, missing stays missing, and a read carries on where it stopped.

APK: outputs/revision-0.108.4, signed, versionCode 164. Not installed. It goes on in the background with Paul's OK, and is then run once with Orbit idle.

Installed and run (4 November 2026, 23:09)

Paul said "install and read".

• Snapshot first: copied to Codex/install-0.108.4-20261104-230911, and on the box to no_backup/schedule-rollback-0.108.3.
• Install: versionCode 164. No plan changed, and the tickets are identical.
• The evidence read: evidence asked=2438 missing=2438 read=2431 failedBatches=0. The file was copied off for the trial (docs/stage-d-trial-2-2026-11-04.md).