jiang-agent-transcript-passlisted

# Jiang Agent Transcript Pass Use this for semantic processing of `content/workflow/tasks/<source-slug>/transcript-agent-packets.jsonl`. ## Model Policy Use `gpt-5.4` for normal transcript packet parsing. Scheduled production wakes should use low reasoning when supported. Request escalation for noisy ASR, ambiguous speaker interaction, dense Great Books passages, or geopolitical argument chains. Escalate to `gpt-5.5` when the packet contains unusual Jiang metaphors, possible contradictions with earlier sources, hard speaker ambiguity, or material likely to mutate the public lens. Do not use mini-class models for normal packet parsing; they are only for coordination or cheap comparison. The corpus is now a calibration anchor. Use existing semantic bundles, strong episode reads, and lens pages as comparison surfaces when judging whether a packet contains a recurring mechanism or a new source-specific pressure. ## Contract Process only the packet's `focus_refs`. Use `context_segments` only to interpret those focus refs. Return JSON matching: ```text ops/schemas/agent-transcript-pass.schema.json ``` Validate outputs with: ```bash node ops/scripts/validate-agent-pass.mjs content/workflow/proposals/<source-slug>/*.semantic.json ``` ## Speaker Handling Speaker labels are machine diarization hints. Do not trust them blindly. Use context to decide whether a span is: - Jiang monologue, - an audience/interviewer question, - Jiang answer, - a back-and-forth exchange, - quo