Onboarding design partners now.Talk to us
truluce
Book a demo
For consumer lenders and card issuers

Underwriting decisions you can defend.

Every manual-review case analyzed, every statement cited, every decision on the record.

Try the live demo Book a demo
Watch the one-minute tour
Founded by risk and engineering leaders from
Made CardAmerican ExpressWells FargoMetaCitibank
Case #2043 · Charlotte, NCSynthetic example
The agent recommends
Reject
High confidence
Every statement cited
Open a line to see its evidence and what it means
1,800 miles in 11 minutesE6E7

The application came from Charlotte. Eleven minutes later, the ID check was finished from a hosting network in Panama City. Nobody covers 1,800 miles in 11 minutes, so someone else most likely finished it.

  • DocV session · DocV session networkHosting provider in Panama City, Panama; Socure flags it as riskyE6
  • Travel · Between the sessions1,800 miles apart, 11 minutes between themE7
Rule Impossible travelSESSION.TRAVEL v2High risk
Device on 3 other applicationsE8E9

The phone used for the ID check shows up on 3 other applications in the last 30 days, under other names, and 2 of them were rejected. One device behind many identities is a classic sign of a fraud ring.

  • Linked applications · Shared identifiers3 applications in 30 days share the DocV device; 2 rejectedE8
  • Device velocity · DocV deviceSeen with 4 other namesE9
Rule Other applications sharing identifiersLINKAGE.APPLICATIONS v4High risk
DocV passed, but doesn't explain itE4E3

The license and selfie passed and the identity matches records, so the person is real. A passed ID check still doesn't say who completed it, so it can't outweigh the travel and the shared device.

  • DocV · Document and selfieAccepted: North Carolina driver's licenseE4
  • Socure KYC · Name, date of birth, SSNMatch records (0.97)E3
Rule Document and selfie resultDOCV.RESULT v2Low strength, in favor
Referred for a fraud score of 820E1E2

Why the case is here: Sentilink rated the application 820 out of 999 for first-party fraud, meaning the applicant may not intend to repay. A high score sends a case to review; it doesn't decide it.

  • Application · Referral reasonHigh first-party fraud scoreE1
  • Sentilink · First-party fraud score820 of 999E2
Rule Vendor score consensusSCORES.CONSENSUS v1Medium risk
Your analyst decides. Try it.
Truluce means true light: every decision lit by its evidence.
Manual review today

Slow, inconsistent and hard to explain.

4+
dashboards per case
2
analysts, two answers
0
written record of why
How it works

From referral to decision, on one screen.

  1. 1
    Case referred
    From Taktile, Alloy or your API
  2. 2
    Evidence gathered
    The vendor data you already pay for
  3. 3
    Your rules run
    A versioned, signed playbook
  4. 4
    Memo written
    Every statement cites its evidence
  5. 5
    Analyst decides
    Sent back to your platform
Fewer false declines

It looks for the innocent explanation, too.

Fraud tools find reasons to decline. Truluce also finds the evidence that clears a good applicant, and cites it.

See all three in the demo
Referred: Low name-address correlation#2041 · Austin, TX

A license issued 82 days ago shows the application address: a recent move.

Accept, high confidence
Referred: License address differs#2044 · Columbus, OH

The old home sold 61 days ago, and the mortgage partner confirmed the new one.

Accept, high confidence
Referred: Proxy on the ID check#2046 · Seattle, WA

Apple's privacy relay: Safari, the same device, and a timezone that matches.

Accept, high confidence

Every statement points to its evidence.

Click a citation, see the source. Gaps are named, never filled in.

Memo

DocV was completed on a hosting network in Panama E6, 11 minutes after a Charlotte application E7, on a device shared with three other applications E8.

E8Linked applicationsYour application history
3 applications in 30 days share the DocV device; 2 were rejected.
A sentence with no evidence behind it never reaches the analyst.

Fraud patterns, drawn for the analyst.

Impossible travel, shared devices and linked applications, at a glance.

1,800 miles in 11 min
Impossible travel
Charlotte, NC
Application · 2:14 PM
Home internet
Panama City, Panama
DocV · 2:25 PM
Hosting network
Same DocV device
#2043This case
#1987Rejected
#2011Rejected
#2029In review

Prove a change before it goes live.

Replay a playbook on your past decisions, and see what each vendor is worth.

Try backtests in the demo
Replayed on 1,240 past decisions
Synthetic example
Agreement
86.1%90.4%
Missed fraud
73
False rejects
4.2%3.1%
Synthetic identity score90 → 92%
First-party fraud score86 → 89%
Address risk83 → 87%
Proxy or VPN71 → 90%
Low name correlation88 → 90%
Current playbookProposed
Playbook release
2026.11.12
Signed off
  • Email not tied to the applicantEMAIL.OWNER v1
  • Network posturev2 → v3
  • Linked applicationsv3 → v4
  • Fixed-line travel speed90 mph
Risk lead
Signed Nov 12 after a backtest

Your policy, versioned and signed.

Rules your Risk lead can read, change and sign off.

Built inside Made Card

We ran this review by hand first.

“I had to remember to do all of this and spend a lot of time doing it. This led us to our next big project: the AI agent.”

Sahasranaman RanganathanChief Risk Officer at Made Card, co-founder of Truluce
Read his field notes
Applications referred to review
50%15%

After a funnel analysis of every signal. So far, only false positives removed.

40

changes to the onboarding flow so far

6

steps in the identity waterfall

For bank partners and model risk

An AI reviewer your auditors can audit.

Versioned playbooks
Signed by a named Risk lead
Reproducible runs
Every memo rebuilt from its snapshot
No credit data
Kept out of fraud recommendations
Your cloud, if you want
Your model account, your data
Security and deployment
Audit log · Case #2043Synthetic example
  1. Memo written10:01:12
    Playbook 2026.11.12 · model and prompt versions recorded
  2. Opened by M. Chen10:02:40
    Analyst
  3. Rejected, agreeing with the agent10:07:05
    Comment on the record
  4. Sent back to Taktile10:07:06
    With the analyst's name
  5. Rebuilt from its snapshotAny time
    Same evidence, same rules, same versions
Integrations

Plugs into the stack you already run.

Cases in, decisions back
TaktileAlloyREST APIBatch upload
Identity and fraud data
SocureSentilinkTransUnionPersonaProveLexisNexis
truluce
Lookups
Google MapsProperty records
Outcomes and alerts
Card processorData warehouseSlack
All integrations

Become a design partner.

Run Truluce on your own past cases first.

Talk to the foundersTry the demo