# AOC linkset backfill

Purpose
- Safely backfill advisory linksets and observations under Aggregation-Only rules.
- Preserve offline kit integrity and determinism during data migrations.

Inputs
- Deterministic NDJSON dataset (gzip) for linksets and observations.
- Target database and collections for advisory linksets and observations.
- Offline kit bundle mirroring the backfill dataset.

Preparation
- Run a dry-run import to validate schema and guardrails.
- Backup target collections before any import.
- Stage rollback scripts and confirm indexes are reproducible.
- Set ingestion flags for backfill windows (link-not-merge enabled, aggregation-only disabled if required for rehearsal).

Execution
- Import the NDJSON dataset with deterministic ordering.
- Record import metrics and structured logs.
- Run a determinism probe test that compares golden hashes.

Rollback
- Restore from backup and reapply deterministic indexes.
- Re-run determinism probes and confirm guard flags are reset.

Evidence to capture
- Backup hash or archive checksum.
- Import logs with counts and zero merge counters.
- Determinism test results and hashes.
- Offline kit bundle hash.

Dataset generation
- Export from staging with a pinned tenant and stable ordering.
- Verify determinism by hashing the NDJSON output twice; hashes must match.
- Publish a .sha256 alongside the dataset.

Offline posture
- Backfill datasets are mirrored into offline kits for air-gap verification.
- Exports and evidence are stored as content-addressed artifacts.