Answer the question six weeks later.
Fraud decisions are never argued about on the day. They are argued about when the chargeback arrives, when a customer complains about a hold, or when someone asks why the team let that one through.
Verification logs
AllLowMediumHighCriticalLast 24 hours- #104209:1478High risk
- #104109:0254Medium risk
- #104008:4724Low risk
- #103908:3112Low risk
- #103808:1291Critical risk
The decision is made in a minute and defended for a year.
Somebody looks at an order, decides it is fine, and ships it. Six weeks later the chargeback lands and nobody can reconstruct what they were looking at.
The same thing happens in reverse. A held order turns out to be a real customer, and the only record of why it was held is a Slack message that has scrolled away.
So every assessment is kept: the score, the findings, the reasoning, and the recommendation, against the order reference you supplied. A second run on the same order adds a row rather than replacing one.
The assessment, not the customer
The log is deliberately thin on personal data. It holds the conclusion and the evidence for it, and joins back to your system on the order reference you sent.
- Order reference
- The order ID you supplied. It is the join key back to your own system, and it is the only piece of the order payload that is written down.
- The assessment
- Score, level, every risk factor with its severity and category, the written reasoning, and the recommendations.
- Organization and time
- Which workspace the verification belongs to, and when it ran. Every run is its own row, whether it came from the dashboard, the API, or the Shopify app.
- Cost, separately
- What the verification consumed is recorded on the credit ledger and the AI usage record rather than on the log row, so billing and evidence stay independent of each other.
- Not held
- Customer names, emails, phone numbers, addresses, payment details, IP addresses, and user agents. These are used for the analysis and are not written to the database.
Four uses, in order of how often they come up
Most common
Representing a chargeback
Second
Tuning your own threshold
Third
Explaining a hold
Fourth
Watching an account over time
The dashboard reads the same data
Volume, band distribution, and credit consumption over time, so the question “is this working” has a number rather than an impression.
Orders scored
this month
Share in each band
against last month
Credits consumed
and what is left
Retention and access
How long are logs kept?
Can we export them?
What happens when we re-verify an order?
Who in our team can see them?
Do the logs contain customer personal data?
One score, four surfaces
Score a week of orders and read the log.
Five free credits is roughly twenty orders. Enough to see whether the scores agree with what you already know happened.
5 credits on signup · no card required