[ebigunso]

Task_2 of the scene plan (stacked on #118).

• One public Scene: the time; participants, each with an optional identity key, name and description in any combination; a setting by key, in words, or both; flat custom values the application owns. It is stored once, on the episode, exactly as given: nothing is resolved, normalized, completed or guessed (ADR-D-0029). A key strengthens the perceived words and never replaces them. Words create no notion and no link.
• The scene's words are indexed with their episode: the embedded text is the summary followed by a labelled Setting: line and one With: line per participant. A description given at recall can therefore reach the scene, and a deployment that perceives only in words still gets situated recall. With no words the embedded text is byte-identical to before; keys and custom values are never embedded.
• A missing time is never guessed. remember fixes it to the instant the plan is prepared, so a replay is identical; a caller-built episode that states no scene is rejected.
• Absence is not a claim: no participants recorded does not mean the character was alone, and there is no completeness flag.
• source_conversation_id becomes the setting key, the participant and start-time hints become the scene, and the correction guard compares original_setting_key. Notions involved but not present, thread affiliations, the end time, and an observation's own time and speaker are distinct facts and stay as they were.
• The scene carries no activity yet: nothing in this plan reads one, and an always-empty field would read as a claim. It arrives with its cue in the routes plan.

Evidence: internal Tier D review approved with no code findings and judged each added test the sole observer of a distinct outcome; the Tier A review approved the public shape for the retrieval side and for v0.3 after one round (the activity removed, one With: line per participant, blank keyless participants rejected). fmt, check, clippy with warnings denied and the full suite pass for the worker and independently for the reviewer: 386 reported, 382 exercised, 4 opt-in service cases left to CI.

🤖 Generated with Claude Code