A regulatory enquiry required an answer stronger than ‘the totals look right.’ I designed three independent tests across full ledger history and made every result reproducible.

Platform needed structural proof, not a row count.

Question: did stored balances match ledger history, and would direct balance edits, deleted movements, or out-of-path credits be visible? All queries had to be provably read-only.

Three tests that fail for different reasons.

1. Conservation identityCompare sum of movements with sum of stored balance snapshots.
2. Per-account reconciliationCompare every account’s stored balance with closing balance of its final ledger row.
3. Chain continuityVerify each row’s opening balance equals previous row’s closing balance.

First run found 354,646 breaks. Test was wrong.

I ordered movements by creation timestamp. That timestamp recorded request start, not commit order; observed skew exceeded one day. Ordering by insertion sequence reduced apparent breaks from 354,646 to zero.

Correction became standing rule: creation timestamps must not order or time-bound financial movements in reconciliation or forensic reports.

Publishing corrected reasoning mattered more than protecting first result.

Every independent test closed.

1. Conservation difference: 0.00Full movement history matched stored balances.
2. Per-account mismatches: 0No negative balances and no balances without backing ledger row.
3. Chain breaks: 0No evidence of balances altered outside normal transaction path.