Mirror mode lets you see Prynt’s decisions on your own traffic before any of them reaches a user. While it is on,
every decision is answered allow, and the verdict Prynt would have returned (block or challenge) is kept next
to it. Scores and reasons stay real, so you can see exactly why.
Every Control account starts this way: the mirror turns on by itself and lasts the first 14 days. When the window ends, the mirror report lists what Prynt would have stopped, with counts, the most common reasons and recent examples. You judge the decisions first; then they apply.
Uber describes the same practice for its fraud rules engine, where a rule with the Evaluate status “is being run but won’t actually affect the user” (Uber Engineering, 2017). Mirror mode does that for every decision at once; evaluate mode does it for a single rule.
Mirror mode is part of Prynt Control.
1. Send identifications as usual
Nothing changes in your integration. Identify on the pages that matter and send your account id as linkedId, so the
report can tell you which accounts were involved (see Send your user id):
const prynt = await Prynt.load({ apiKey: 'pk_live_…', endpoint: 'https://api.pryntid.com', extendedResult: true });
const { requestId } = await prynt.identify({ linkedId: String(user.id), tag: { action: 'signup' } });
2. The mirror starts by itself
On a Control account, mirror mode starts automatically with the first request Prynt receives after Control is activated, and runs for 14 days. It starts automatically only once per account: if the mirror has ever been started, it is not started again on its own.
From the console (Mirror mode) an admin or the owner of the account can:
- stop it early, so decisions apply from that moment;
- restart it for 1 to 30 days, for example after changing your rules. Restarting opens a new window from today.
The console uses these endpoints, authenticated with your console session:
| Endpoint | What it does |
|---|---|
GET /admin/mirror | Status: available, active, startedAt, until, defaultDays, maxDays |
POST /admin/mirror with { "days": 14 } | Restart the window (admin role; 1–30 days) |
DELETE /admin/mirror | End it now; decisions apply again. The window is kept for the report |
GET /admin/mirror/report | What Prynt would have stopped during the window |
3. What your code receives
While the mirror is active, the identification response is answered allow and carries a mirror object, both at
the top level and inside decision:
{
"requestId": "req_…",
"visitorId": "vst_…",
"decision": {
"decision": "allow",
"riskScore": 82,
"reasons": ["bot", "datacenter"],
"mirror": { "active": true, "wouldBe": "block", "until": "2026-12-15T10:00:00Z" }
},
"mirror": { "active": true, "wouldBe": "block", "until": "2026-12-15T10:00:00Z" }
}
decision.decisionis what your code should apply. In mirror mode it is alwaysallow.mirror.wouldBeis the real verdict:allow,challengeorblock.riskScoreandreasonsare the real ones.
Write your integration against decision.decision from the first day. When the mirror ends, the same field starts
carrying the real verdict and your code does not change:
// On your server, after reading the event with your secret key
if (event.decision === 'block') return res.status(403).json({ error: 'unavailable' });
if (event.decision === 'challenge') return res.json({ next: 'verify' });
return proceed();
When you read the event later with your secret key (GET /v1/events/{requestId}), events recorded in mirror mode
carry shadowDecision, the verdict that was not applied:
{ "requestId": "req_…", "decision": "allow", "shadowDecision": "block", "riskScore": 82 }
Webhooks
Webhook payloads sent during the mirror include the same mirror object. Because the decision is allow, the
decision.block webhook never fires in mirror mode; use mirror.wouldBe in the visitor.identified payload if
you want to count would-be blocks on your side.
Edge connectors
The edge gate (GET /v1/edge/gate, used by the nginx, OpenResty and Workers connectors) answers 204 (allow)
during the mirror, and adds an X-Prynt-Mirror header with the verdict it would have returned (block or
challenge). Your edge keeps letting traffic through; your logs can record the header.
4. Read the report
Mirror mode in the console shows the report; GET /admin/mirror/report returns it:
{
"mirror": { "available": true, "active": false, "startedAt": "2026-12-01T10:00:00Z", "until": "2026-12-15T10:00:00Z",
"defaultDays": 14, "maxDays": 30 },
"window": { "from": "2026-12-01T10:00:00Z", "to": "2026-12-15T10:00:00Z" },
"counts": { "block": 63, "challenge": 412, "allow": 18240 },
"total": 18715,
"topReasons": [{ "reason": "datacenter", "count": 201 }, { "reason": "multiAccount", "count": 96 }],
"examples": [
{ "requestId": "req_…", "visitorId": "vst_…", "wouldBe": "block", "riskScore": 91,
"reasons": ["bot", "datacenter"], "url": "https://example.com/signup", "country": "BR",
"createdAt": "2026-12-14T22:41:05Z" }
]
}
Read it in this order:
- The counts. How many requests would have been blocked or challenged, out of the total.
- The top reasons. If one reason dominates, check that it describes abuse and not a normal part of your traffic (an office VPN, a partner’s servers).
- The examples. Open the would-be blocks. Label the ones Prynt got wrong as legit and the right ones as
fraud in the console or through
POST /v1/outcomes; those labels change the next decisions.
If something looks wrong, fix it before the window ends: add an allow-list entry, or adjust or remove the rule responsible. Restart the mirror if you want a fresh window to confirm the change.
When the mirror ends
At until, or when you stop it, decisions apply again: decision.decision carries the real verdict and the
decision.block webhook fires as usual. The report for the last window stays available.
Prynt never blocks on its own, in mirror mode or after it: it returns a decision and your code decides what happens.