CL Remap (advisory drafts)

SUPERVISOR CL Remap is the advisory-only flow for proposing code reassignments on an existing CL. Lives at Vessel → CL Remap.

Flow

1. Pick a vessel
2. Select a subset of components (by L4, by free text, or all)
3. Click Run remap — the system runs Kimi K2.6 in two tiers:
   • Tier 1 (max_tokens 32768, batch 120, parallel 5) — fast initial pass
   • Tier 2 (max_tokens 65536, batch 60, parallel 3) — deeper for FLAG_HUMAN drafts
4. Drafts land in cl_remap_* tables, marked FLAG_HUMAN for ambiguous proposals
5. Supervisor reviews, approves or edits
6. Approved drafts merge back into vessel_components

Status

The Kimi remap orchestrator (scripts/kimi-remap/) currently has a known bug — Tier-2 parser dies on Unterminated string in JSON at position N when batches truncate at max_tokens. The handoff snapshot for the EGOR LETOV run (234 FLAG_HUMAN drafts) is at audit-notes/kimi-run-egor-letov.json on PR #2.

Do not run a fresh remap until the parser is fixed — see incident response for the fix path. Use the existing drafts in the meantime.

Constraints

See AI architecture — Kimi constraints for the hard-won list (model name kimi-k2.6, temperature 0.6 / 1.0 only, etc.).