Last updated 10 September 2026
This page is not ready to be relied upon
This page describes what is actually in place. The second half describes what is not, which is the half that is usually missing.
Every table that holds a customer's data has row-level security enabled and forced in the database, and the application connects as a role that cannot bypass it. That distinction matters: a database superuser steps around every such policy silently, so the application refuses to start in production if it finds itself connected as one.
This is enforcement by the database, not by the application code. A query that forgot its tenant filter returns nothing rather than somebody else's ledger.
Sending something to a government system cannot be taken back, so those operations are written down before they are attempted, carry a fingerprint of exactly what was sent, and cannot be retried into a duplicate. An operation whose outcome is unknown stays unknown until somebody establishes what happened, rather than being optimistically retried.
Changes are recorded with who made them and when. The audit trail is written outside the transaction it describes, so an action that succeeded is never rolled back by an audit failure, and an action that failed never leaves a record claiming it happened.
The database is backed up by the hosting platform on its schedule. Restores are exercised regularly against a scratch database, and the exercise checks that the restored books still balance rather than only that the file loaded.
We are not going to imply otherwise by staying quiet about it:
If you have found a security problem, write to the address on the contact page with enough detail to reproduce it. We will confirm we have received it, and we will not take any action against somebody who reports a problem in good faith and does not access or alter other people's data while finding it.