CONCEPT Cited by 1 source

Weekly integrity reconciliation¶

Definition¶

Weekly integrity reconciliation is the pattern of running a lower-frequency, out-of-band audit that checks whether the invariants a fast-path automation is supposed to maintain are actually holding, and surfaces any violations for human review. It's the structural complement to a continuous bot — the bot keeps the system moving forward, the reconciliation checks whether anything got skipped, silently dropped, or drifted.

The shape is domain-agnostic but appears especially clearly in fork-sync automation, where the fast path is "cherry-pick every merged PR from OSS to private" and the failure modes are all of the form a PR exists but the cherry-pick didn't happen.

The pattern¶

Two independent loops run on the same underlying data:

Fast path (minutes–hourly): the primary automation that does the work. Optimised for latency and forward progress.
Slow path (weekly): audits the invariants. Optimised for completeness — full sweep, expensive queries, no time pressure.

When they disagree, the audit wins. The audit reports the disagreement (typically to a dedicated issue or dashboard) and a human resolves it.

Why not just make the fast path correct?¶

Because the fast-path's failure modes are often things it literally can't see:

Missing signals it depends on — e.g. a label it expects the human to apply but that was never added. The fast path looks at every PR it can see; it can't inspect PRs that should have had a label but didn't.
Bypasses of the automation — e.g. someone merged directly to a branch without going through the fast path's workflow.
Drift from external state — e.g. GitHub allowed an action the fast path didn't request.
Lossy or rate-limited APIs — incremental cursors can skip items under edge conditions.

A second, full-sweep pass sidesteps all of these because it reads the ground truth and doesn't rely on the fast path's view of reality.

Canonical instantiation: Vitess cherry-pick bot¶

From the PlanetScale post: "we built integrity checks directly into the bot, which runs weekly. These checks provide a summary of any discrepancies, allowing for manual intervention when necessary." (Source: sources/2026-04-21-planetscale-automating-cherry-picks-between-oss-and-private-forks)

Two weekly checks:

Upstream-in-sync-with-OSS audits:

Open cherry-pick PRs against upstream — never merged, stalled.
OSS main PRs that were not cherry-picked into upstream — the hourly bot missed them.
PRs merged directly into latest-x.0 instead of being backported from upstream — bypass of the workflow.

Latest-branches-consistent audits:

Open backport PRs against latest-x.0 branches — stalled.
PRs merged into latest that are not backports — bypass.
PRs backported to latest-x.0 but not to higher-numbered latest branches where they also should be — incomplete propagation.

The audit results are posted to a dedicated GitHub issue. Human operators triage from there.

Why weekly and not continuous?¶

Scheduling the audit less frequently than the fast path is a deliberate choice:

Cost: audits are full sweeps. A full sweep every hour is expensive in API calls, compute, and analyst attention.
Noise: audit violations should be rare. Reporting them more often than they occur just adds signal-to-noise overhead.
Natural action cycle: humans read the issue at the start of the week and address what's pending; a weekly report aligns with that rhythm.

Generalises beyond fork-sync¶

The same shape appears in many corners of the wiki:

Anti-entropy in concepts/anti-entropy — periodic reconciliation between replicas in distributed stores.
Consistency checkers in patterns/consistency-checkers — Dropbox Magic Pocket, Snowflake, etc. run out-of-band checks that scan the entire data plane for drift between metadata and chunks.
Repo health monitoring in patterns/repo-health-monitoring — same idea applied to code repositories.
Dual-write migration validation — patterns/dual-write-migration often pairs with a reconciliation job that compares old and new systems for drift.

Weekly fork-sync reconciliation is just this shape applied to the very specific invariant "every OSS PR has a matching private cherry-pick PR".

Seen in¶

sources/2026-04-21-planetscale-automating-cherry-picks-between-oss-and-private-forks — PlanetScale's Vitess cherry-pick bot runs two weekly reconciliation checks and posts the results to a dedicated GitHub issue for human triage.