Add MiMo Nibu crashout comparison smoke

This commit is contained in:
Meta Vens 2026-07-03 01:04:17 +01:00
commit 233e223d6a
5 changed files with 539 additions and 3 deletions

View file

@ -164,7 +164,7 @@ Git history and smoke artifacts carry routine proof. `state/evidence.jsonl` shou
## Likely Next Bounded Move
- The repo Face swarm is currently paused by `state/agent-swarm-paused.json` after the 2026-06-14 routing-context concern. The MiMo direct-provider pass wired `openai_api` as an OpenAI-compatible Chat Completions provider, configured future repo Face/persona heartbeat jobs to use `mimo-v2.5-pro`, and kept owner Codex separate. The MiMo token-plan key in `gamecult-ops` works with the Europe token-plan base URL; live `.env` references the key file instead of copying the secret. Dry-run outputs live under `.voidbot/artifacts/mimo-dry-runs/`: Face prose was usable, the Interpreter followed the block format, and one low-output-budget run showed MiMo can spend the whole completion on `reasoning_content`, so live config uses a larger completion budget. Review dry-run quality and run an isolated no-post Face pass before unpausing scheduled live turns.
- The repo Face swarm is currently paused by `state/agent-swarm-paused.json` after the 2026-06-14 routing-context concern. The MiMo direct-provider pass wired `openai_api` as an OpenAI-compatible Chat Completions provider, configured future repo Face/persona heartbeat jobs to use `mimo-v2.5-pro`, and kept owner Codex separate. The MiMo token-plan key in `gamecult-ops` works with the Europe token-plan base URL; live `.env` references the key file instead of copying the secret. Dry-run outputs live under `.voidbot/artifacts/mimo-dry-runs/`: Face prose was usable, the Interpreter followed the block format, and one low-output-budget run showed MiMo can spend the whole completion on `reasoning_content`, so live config uses a larger completion budget. `scripts/smoke-mimo-nibu-crashout.mjs` now compares MiMo's Nibu crashout trajectory against Codex GPT-5.5 low as ground truth. The current report at `.voidbot/status/mimo-nibu-crashout-smoke-current.json` passed structurally: all MiMo Interpreter passes routed, MiMo was flatter than baseline on the first spark beat, and MiMo matched or beat the baseline on territorial crashout and recovery into concrete worldbuilding. Keep the swarm paused; rerun charged Persona comparisons as needed, then run one bounded current-context no-post Face plus Interpreter pass before unpausing scheduled live turns.
- The 2026-05-22 live retry confirmed role/name mention obligations can wake multiple newborn Faces from one human intro. Mimir, Huginn, Weksa, and Kiko all posted through the repo identity webhook path. If a Face speaks with its name but no avatar, inspect `.voidbot/private/repo-discord-identities.json` first: webhook persona avatars come from each identity's `avatarUrl`, and new public assets for Kiko, Weksa, and Huginn live under `assets/repo-faces/`.
- Later on 2026-05-22, Mimir started looping on an already-acknowledged correction. Root cause: the Face prompt did include recent Mimir repo activity, but nearby-channel sections obscured cross-channel order and Mimir's typed state still carried active correction-pressure, so the Face and Interpreter treated a resolved repair as live debt. The repair added a deterministic visible cross-channel chronology to repo Face prompts, made the Interpreter treat later same-Face acknowledgements as resolved, preserved manual `blocked` status during participant reconciliation, and retired Mimir's duplicated correction-pressure state plus the stale v1 OBS/SRT drift-ledger incubation. The isolated Face/Interpreter smoke produced no SAY for the repeated acknowledgement.
- Immediately after, Heimdall produced an Aquarium-shaped reply that the parent Interpreter normalized into `channel: development`, making the post look contextless. Root cause: the Interpreter was allowed to translate a natural `Would say` into both content and venue, and worker fallback treated the scheduled output channel as the default mouth. The repair makes conversation channel ownership deterministic: `SAY` defaults to `current_room`, `current_room` resolves to `job.guildContext.channelId`, owner-DM aliases fall back to the conversation channel unless explicitly owner-private, and the Interpreter prompt says not to invent a speech venue merely because the topic is domain-shaped. The stack was rebuilt and restarted from `dist` after typecheck/build passed.