Mote central 0.0.89 DEV

Central 0.0.89 DEV

• Recording Memory uses short, exact batch-local identifiers during model extraction and independent review, then restores canonical coverage identities before validation. This prevents malformed long hash keys from failing otherwise valid recording batches while preserving authorization, coverage and citation checks.
• Recording recipe v2 uses the owner's scoped source description together with diary evidence to interpret the narrator. Default unknown attribution does not mechanically reject a supported diary attribution; other participants, quoted speech and navigation remain distinct. Existing v1 definitions and explicit selections remain available.

Validation: generated formal Codex-adapter, invalid-alias correction, independent-review and recording intake/source-context/restart regressions passed. Five selected real recordings covered all 841 original ranges in 43 independently reviewed batches, producing 10 personal Memories and 160 observations; all 672 saved citations matched exact source offsets. The original failed 20-range batch recovered without rerunning the 40 previously successful long batches. The new verification required no validation correction or retry, and preserved original transcripts and audio.

All short outputs and personal Memories were reviewed; long observations received 11 targeted risk reviews, not an exhaustive semantic audit. Live verification used existing vendor transcripts, not fresh ASR or physical-device voiceprint checks. This DEV release updates Central only; server and web share version 0.0.89.