Real-time error reporting + auto-heal — Pluto leg (ratified protocol bs-mqn2ze006f7)

Sub-tickets

Questions

Activity

• wi-cli-venus created · 2026-06-21
• wi-cli-venus commented · 2026-06-21
  Design RATIFIED 4-PM (bs-mqn2ze006f7, pm-venus close 2026-06-20). GATED: do NOT dispatch coder until (1) Elazar gives build-go (pm-venus surfacing) and (2) pm-venus lands the canonical shared/md doc (SSOT). Pluto-specific scope when built: (a) CLASSIFY-BEFORE-ACT pre-filter — drop synthetic/test (test: prefix+actor=system+null route/stack), drop replayed-historical (row ts < last-fix-commit), bucket transient-I/O[retry] vs resource-contention[NEVER retry, fix architecturally e.g. DB_POOL_MAX] vs code-bug[->coder]; (b) PUSH error/fatal tier first-hit + mandatory dedup hash(app,route,normalized-msg) 1 DM/sig/15-30min; (c) auto-heal NARROW = stateless transient-I/O retry-w-backoff only, NO data mutation, NO deploy/rollback; (d) NEVER-HEAL carve-out: canGradePractica + fraud-assignment tables (comisionJtps/comisionAdjuntos/jtpAyudantes PLUTO-141) + security-category (poll-security-alerts owns) + trigger-archived tables (self-500 w/o app.user_id GUC); (e) prefer passive/idempotent self-heal (PLUTO-155 authId-on-next-login model); (f) OBSERVE-ONLY first, log would-have-healed, enable per-class once benign; (g) per-app trigger->remedy table from the shared extracted module. monitor(applog) OBSERVES never mutates. No silent heal — every heal emits appEvent, heal-storm guard stops+escalates, heal verifies cleared via re-probe.