Theme acceptance and retained-source recovery — 2026-10-10

At this acceptance checkpoint, the parent change was 22/25 complete. The operator accepted implemented behavior and waived further proof collection. Tasks 5.3b, static 5.4 and 5.5 now record that acceptance and disposition instead of claiming new physical trials. Actual theme consumer implementation (3.1), complete compatibility publication (6.1) and archive (6.2) remain open.

Worktree /home/jadams/tmp/k230-theme-closeout-2026-10-10, branch closeout/themes-2026-10-10, base 38d370c1a7989b982e04cfd74799542cccd3424e. Owned paths: this parent’s planning artifacts, reboot trial and tests, theme closeout evidence, its board checklist section and work-board entry. No board, serial port or build slot was reserved. This changes the operator tool and acceptance records; the runtime already implements the retained-cache policy, so no new firmware build or board installation is needed for this correction.

Acceptance and limits

The user’s project-session reports include:

it all worked land it

continue but you dont have to gather that proof ive already confirmed its fine carry on it’s all good

The operator record preserves that authorization. It is broad acceptance of implemented behavior, not a claim that each originally requested experiment was run. The superseded requests preserve the exact previous tasks. The previous publication plan records why 6.1/6.2 were revised while remaining open. Existing dark/light board activation and unchanged community activation retain their original evidence classes.

The waiver does not implement missing hover/focus or typography/spacing consumers, and does not turn unavailable field classifications into full compatibility. The 102-field inventory continues to identify these limits. Settings/notifications and additional background fit modes/overlays/video remain in their existing named successors; this closeout does not run or close those successor workloads.

Corrected removed-source policy

Design decision 1 already requires retaining the last working generation independently of its source clone. theme_catalog.selected() exposes no source id when the clone is missing, while retaining its generation id. The Rust startup receiver loads the valid selected snapshot before falling back to the packaged default; the card receiver likewise loads valid active state. The old boot scenario and reboot-v1 trial incorrectly required a reset to default when only the clone disappeared.

The delta spec now distinguishes an empty fresh home from removal of a source whose cached generation remains valid. The former uses the packaged default; the latter retains its selected appearance and wallpaper. The trial’s unavailable-source arm now requires an absent fixture source and catalog source id, the same catalog generation, and exact retained generation, wallpaper fingerprint and preference-byte identity. It still restores the entire normal state directory and fixture through acknowledged recovery on completion or failure. Candidate/workload identity and distinct boot-id guards are preserved.

New trials use reboot-v2. An old reboot-v1 checkpoint is refused before resume mutation because its acceptance policy differs, but remains recoverable through the existing manual restoration path. The historical v1 procedure retains its prior results; use its setup and recovery instructions with this corrected policy for any future optional trial.

Host validation

Commands, timestamps and source hashes record:

• python3 tests/test_handheld_theme_trial.py: 25 passed. Covers retained generation success, unexpected default reset, removed-source wallpaper mutation, catalog generation mismatch, old-protocol refusal with recovery, boot/workload/candidate identity guards and complete restoration.
• python3 tests/test_handheld_theme_rendering.py: 7 passed. Existing token inventory and rendering checks; no new physical or complete consumer claim.
• python3 tests/test_omarchy_theme_activation.py: 14 passed. Existing real activation fixtures include selection and public paths surviving a removed clone; no physical boot is simulated into a claim.
• openspec validate --all --strict: all current items pass.
• git diff --check: passed.

The trial fixtures inject boot identities, transport and captures on the host. Public reports remain bounded and exclude private paths/source names. Physical observation and full fresh HOME remain explicitly UNVERIFIED. These checks prove the operator protocol and restoration logic, not pixels on glass or measured performance. The parent stays open for the actual source gaps.

Publication consistency

The first live-site check found 18/21 on the work card, although OpenSpec correctly counted 22/25: the site counts top-level checkboxes and four older completed subtasks were indented. Those entries are now top-level with their identifiers and scope unchanged, so both projections include all 25 tasks. This formatting correction does not change implementation or test results.

Subsequent consumer and archive closeout

The missing consumers were subsequently implemented and cross-built; the complete compatibility report and consumer record record 497 passing host cases, reviewed captures and exact-revision publication. The parent is now archived at 25/25. This update does not change the operator statements or original waived requests preserved here, and does not turn the new cross-build into a claim of board installation or individual physical verification.