Augur
The control-testing layer for

Your control test checked a sample.Augur tests every case — with traceable exceptions and a human signoff.

Illustrative control run

Watch every case get tested

A simulated merchant-onboarding (KYB) suite running across a synthetic Italian merchant population. Documents are classified and extracted, records reconciled, rules executed, and every failure or uncertain case is routed to a human reviewer — with each step written to the audit log.

Augur control run · live/KYB-IT-01 · Merchant onboarding evidence · Synthetic Italian merchant population
0000 / 1,200 cases000 exceptionst+00:00.0
Full eligible population · one cell per case
Run r-0042 · v0.3-draft
PassFailUnknownNot tested
Coverage0.0% processed · 0.0% tested · 0 not tested (reason logged)
↑ a 25-case manual sample, for scale (illustrative)
Pass
0
0.0% of processed
Fail
0
0.0% of processed
Unknown
0
0.0% of processed
Not tested
0
0.0% of processed
Pipeline trailstreaming
RUN START
Suite KYB-IT-01 v0.3-draft approved by risk owner · population 1,200 eligible merchants.
Case under inspectionM-IT-00858
Enoteca Morandini S.r.l.s.
synthetic merchant · prov. MI
INGESTstage 1/7
documentatto_costitutivo.pdf
classArticles of incorporation · 0.95
legal nameEnoteca Morandini S.r.l.s.
p.iva96397610459
reaMI-163410
langit → normalized schema
linkedonboarding · CRM · warehouse
C1 ···C2 ···C3 ···C4 ···C5 ···
Exception queue → human review0 open
No exceptions yet. Uncertain or failing cases will be routed here — never silently dropped.
Audit logappend-only
t+00:00.0 system run r-0042 opened · suite KYB-IT-01@v0.3-draft · manifest pinned
PASS
Control satisfied, with evidence linked.
FAIL
Control not satisfied. Routed to review.
UNKNOWN
Evidence was too uncertain to decide. Routed to review.
NOT TESTED
Couldn't be tested — data missing or out of scope. Reason recorded.
How a control run works

From control statement to audit packet

Augur doesn't give compliance opinions. It runs the control your team already defined, the way your team would — across every case, with a record of each decision.

01
Agree the control

Start from an approved control statement: policy owner, jurisdiction, version, eligible population and source systems.

02
Connect and discover

Read-only access to allowed fields and documents. Schemas are inventoried and lineage recorded before anything runs.

03
Classify and extract

Documents are classified; names, registration IDs, dates and ownership are extracted and translated, each with a confidence score.

04
Reconcile

Identifiers are normalized and the document subject is linked to merchant and beneficial-owner records across systems.

05
Execute rules

Deterministic, versioned checks return PASS, FAIL, UNKNOWN or NOT TESTED — with coverage flags, not a model's opinion.

06
Separate the noise

Control failures are kept apart from missing data, untested cases and extraction uncertainty.

07
Prioritize, never bury

Exceptions are ranked with a materiality policy the risk owner approves. Nothing is silently dropped.

08
Human review

Exceptions route to authorized reviewers. Decision, rationale, reviewer, date, policy version and citations are recorded.

09
Export the run

A reproducible control-run and audit package: population, exclusions, exceptions, dispositions and evidence links.

What a pilot produces

One suite. One dataset. A run you can defend.

We start with one market-specific merchant-onboarding suite, defined with your risk team, and compare results side-by-side with your current process.

Control-run packet · contents
01Population testedEvery eligible case in scope, not a sample — with the count of what was and wasn't testable.
02Coverage & quality exclusionsMissing evidence, out-of-scope cases and low-confidence extractions, each with a recorded reason.
03Exception inventoryFailures and unknowns, ranked by the materiality policy your risk owner approved.
04Reviewer dispositionsWho decided what, when, why, and under which policy version.
05Evidence linksEvery outcome traced back to the document page and record it came from.
06Exportable audit packetA reproducible run your second line and internal audit can re-examine.
Architecture

Compute goes to the data

Sensitive document and record processing is designed to run inside a client-controlled execution boundary, with minimal telemetry and residency and retention set by you.

Client-controlled execution boundary
01Connect & discover
02Evidence processing
03Identity & reconciliation
04Rule execution
05Statistical measurement
06Review layer
07Audit & controls
Human in the loop

A risk officer approves:

  • ✓The control suite and its version
  • ✓Tolerance and materiality policy
  • ✓Escalation paths
  • ✓Overrides
  • ✓Final certification of the run

Augur prepares the evidence. Your people make the call.

Design a pilot

Bring one control you sample today

We're starting with merchant-onboarding control testing: the evidence your team already checks by hand, but across the full eligible population instead of a small sample. We'd like to define one market-specific suite with your risk team and compare results side-by-side with your current process.