Skip to content

Supersession Audit

Per-sheet walkthrough — L1 Reconciliation Dashboard.

What the sheet shows

Every logical row whose append-only entry column has more than one version. Each rewrite carries a supersedes reason from L1's v1 vocabulary:

  • Inflight — the row was rewritten while still in flight (typical: status flip, late metadata enrichment).
  • BundleAssignment — the row was rewritten when an AggregatingRail picked it up and assigned a bundle_id.
  • TechnicalCorrection — the row was rewritten because the original posting was wrong (amount fix, account swap, etc).

Reads from the BASE tables (spec_example_transactions, spec_example_daily_balances), not the Current matviews — by definition Current hides the prior entries we want to audit here.

Screenshot

Supersession Audit

When to use it

Diagnostic deep-dive. High TechnicalCorrection volume signals a feed problem upstream; high Inflight is normal in a busy bundling cadence. The audit is also useful for cross-system reconciliation (does our rewrite count match the producer system's known correction count?).

Visuals

  • Logical Keys with Supersession (KPI) — count of distinct transaction_id values whose append-only entry column has more than one row.
  • Transactions Audit (Table) — every entry of every superseded logical transaction, ordered by (transaction_id, entry) so the audit trail reads top-down per logical row. The supersedes column on the higher-entry row tells you why it exists.
  • Daily Balances Audit (Table) — same shape on the (account_id, business_day_start) composite key.

Drills

None outbound. Supersession Audit is itself the diagnostic deep-dive endpoint; the analyst's next step is typically a copy-paste of the logical key into a producer-system case file.

Filters

  • Supersedes Reason — narrow the transactions table to one cause class (Inflight / BundleAssignment / TechnicalCorrection). The daily-balances table doesn't get a paired filter — supersession on daily_balances is rare enough that a second control would be noise.

No date filter — audits want the full historical horizon.