Bidirectional model fit — escalate and downgrade
· One min read
Anchor already stopped too-hard work with a first-line SUGGEST-ESCALATE. Premium sessions still burned credits on too-easy work with only a soft “say so and ask.”
What shipped
| Direction | Token | Rule |
|---|---|---|
| Too hard | SUGGEST-ESCALATE: <target> — <reason> | mythos-core 11 |
| Too easy | SUGGEST-DOWNGRADE: <cheaper> — <reason> | mythos-core 10 |
| Good fit | (silence) | no model pitch |
orchestrate.py treats both as fit gates alongside SUGGEST-REROUTE for specialty mismatch (no retry burn; --insist proceeds) — see Dual-axis model fit for the specialty side. Interactive platforms (Claude, Grok, Chat, local) state the same standing check. Downgrade heuristics stay conservative — rename/format/boilerplate, not “any multi-file mid work.”
See Model fitness and Doctrine.