Playbook — learned fittings
This is how HealPipe gets cheaper and faster overnight: the same structural break never pays for an LLM heal twice.
Memory-backed healing — student-proof differentiator
When HN renames the same class again, Stage 0 resolves it before coffee. Judges scoring self-heal depth should see fingerprint triggers and a learning loop — not just
bdata scraper heal on every hiccup.Stored transitions
0
unique hash-pair fixes
LLM heals skipped
0
via Stage 0 replay
Avg confidence
—
across playbook entries
Loading playbook…