Run Ledger Contract

Authority-of-record for task_occurrences in /workspace/store/messages.db, the host's scheduled-run ledger (jbaruch/nanoclaw#1024), under coding-policy: stateful-artifacts.

Owner and schema

• Owner: the jbaruch/nanoclaw scheduler, through src/db-task-occurrences.ts and its state migrations; no container skill writes or migrates the table
• Schema of record: src/state-migrations/state-022-task-occurrences.ts and state-024-task-occurrences-schema-version.ts in jbaruch/nanoclaw
• One row per occurrence, keyed (task_id, scheduled_for); scheduled_for is the task's next_run at dispatch
• A cadence-registry task's task_id is cadence-registry::<group>::tessl__<skill>
• Every timestamp is UTC ISO with milliseconds and Z (2026-10-19T12:00:00.000Z)
• Every record carries schema_version 1; a shape change bumps it in a new state migration

Record fields

• completed is final; a retry restarts a failed, killed or unknown occurrence as running and counts attempts
• unknown means the host stopped before the run finished, never success

Readers

• jbaruch/nanoclaw-admin heartbeat precheck: skips a missed-task recovery the ledger covers
• jbaruch/nanoclaw-admin system-audit: reports unverified sends
• Each reader names the versions it accepts and counts only rows of those versions
• A row of another version is no usable prior state for that reader, and the reader says so
• A missing table or column, or a query error, makes the ledger unreadable: heartbeat decides on the follow_me cursors alone and says so on stderr; system-audit records an audit_errors entry
• Readers are read-only and never migrate

Rollout

• Owner and readers ship through separate pipelines: the host deploy and the admin plugin publish
• A version bump follows coding-policy: stateful-artifacts Cross-Pipeline Schema Bumps
• Every no-usable-state path only drops a suppression or adds an audit finding, never a wake-always or alert storm