MAP reconciles source against target down to the row, then issues a cryptographically signed evidence bundle. Same inputs, same hash — every time, on any machine — so an auditor can verify the result themselves. No AI in the verdict path.
Illustration of the determinism principle: identical inputs reproduce an identical hash. Edit the field and the hash recomputes; run it again and it matches. The product hashes canonicalised rows — not this string.
When you move billions of rows from Oracle into a lakehouse or a cloud warehouse, someone has to assure the regulator that nothing was dropped, truncated, re-typed, or silently changed. Spot checks and reconciliation spreadsheets rely on sampling and on trusting whoever ran them. They can't be reproduced by an outside party — and they don't survive the moment an auditor says "show me."
Row counts and a few sampled checks. Narrative, not evidence — and impossible to re-run identically.
A green dashboard you can't inspect, with a model or sampling sitting between the data and the verdict.
Every row reconciled and folded into one evidenceHash, signed with Ed25519. Hand it to your auditor; they re-check the signature, the hashes, and the verdict themselves — offline if they want to.
One signed artifact that stands on its own. Self-describing, self-verifying, and identical for identical inputs.
A deterministic PASS or FAIL with the exact counts behind it — matched, missing, extra, and differing rows.
Every difference located and classified against a 25-class corruption taxonomy — truncation, re-typing, encoding, and more.
The full declarative rule set that produced the verdict, captured in a versioned rulesHash so the criteria are part of the record.
An Ed25519 signature over the whole bundle, so any later tampering breaks verification immediately.
One fingerprint of the entire run, plus a byte-level artifactHash. Reproduce the run and the hash matches, bit for bit.
A browser tool and an offline CLI that re-check everything without sending data back to MAP. Independent verification, built in.
Point MAP at both ends, declare what "correct" means, and let it produce a record that holds up on its own.
Point MAP at source and target through envelope-encrypted connection profiles. Dialect quirks are handled so equal data reads as equal.
Declarative checks: column compare, referential integrity, uniqueness, cardinality, per-row assertions, and aggregate bounds.
A streaming merge-join compares both sides in a single pass at billion-row scale, in constant memory, deterministic by construction.
Download the signed bundle to WORM storage. Anyone — your auditor included — can re-verify it, in the browser or offline.
Anyone can diff two tables. The hard part is producing evidence an auditor will accept — which means the result has to be reproducible by someone who has no reason to trust you.
Partitioned, constant-memory execution carries a proven hash-identity guarantee: a partitioned run and a whole-table run produce the identical evidenceHash.
A streaming merge-join reconciler compares source and target in one pass, partitioned to run within fixed limits no matter how large the tables get.
Compare any supported source against any supported target — the canonical layer makes them comparable on equal terms.
Dialect quirks are normalised before comparison — binary handling,
UUID casing, CHAR padding — so data that is
equal reads as equal, and schema drift is detected rather than
silently passed over.
Built to drop into a regulated environment — in your own VPC or fully self-hosted — without bolting on governance afterward.
Every bundle is signed; tampering breaks verification on the spot.
Write-once, read-many. Records can't be quietly rewritten after the fact.
An immutable, linked history of every action — gaps and edits are detectable.
Plug into your identity provider; no separate password store to manage.
Five roles, with SCIM 2.0 provisioning to add and remove people automatically.
Connection secrets are envelope-encrypted at rest, never stored in the clear.
A CycloneDX software bill of materials, with dependency audits gating every build.
Deploy inside your own boundary; data never has to leave it.
Re-check any bundle from the command line, with nothing sent back to MAP.
Compliance packs
Independent verification is the whole point. Drop an evidence bundle into the browser verifier or the offline CLI and it re-checks the signature, the hashes, and the verdict — without sending anything back to MAP. If it verifies on a machine that has never spoken to us, the evidence stands on its own.
We're onboarding a small number of regulated institutions. If you're moving a regulated dataset and need evidence your auditor will accept, let's talk.