Same data, different jobs

Kenshiki tests whether a claimed identity is statistically coherent with a real human life — across address history, carrier signals, device tenure, and stated facts — and delivers the result as a signed, replayable evidence packet rather than an unexplainable score. Credit origination is where we deploy it first, because that is where the regulatory demand for proof is sharpest today. But the computation does not know it is doing credit.

A real life leaves one coherent evidence trail. Read from one end, that trail rejects the fabricated applicant. Read from the other, it admits the genuine one that legacy systems wrongly refuse — the thin-file borrower, the night-shift worker, the new arrival. Every market below is the same trail, read for a different claim.

Same data, different jobs: one evidence layer rejects the fabricated life and protects the real one from wrongful denial.

Expansion without discipline is how companies fragment. Every market we consider is scored against the same five tests before it gets a slide, a line of code, or a sales call. A market that passes all five gets the same machine pointed at a different claim. A market that fails one gets a smaller bet — and below, we say which test it fails.

Each row is the same evidence trail, read for a different claim. Problem-size figures are third-party estimates, sourced in each market section below.

Status legend — Now: deployed wedge. Next: architecture-ready, active design work. Horizon: validated fit, longer procurement or partnership arc. Statuses describe our roadmap discipline, not customer relationships.

• ● Life-process claim : full fit
• ● Paired error costs : full fit
• ● Examiner demand : full fit
• ● Data residency : full fit
• ● Industrial rings : full fit

This is where we live today, because this is where the two errors are both priced and both examined. Estimated U.S. synthetic-identity fraud losses crossed $35 billion in 2023, lender exposure on cards, auto, retail, and personal loans sits at an all-time high, and the average charge-off per minted identity runs near $13,000. Meanwhile the same legacy stack that admits the synthetic quietly refuses the genuine thin file — the 24-year-old, the recent arrival, the cash-economy worker whose life is real but whose bureau file is short.

Same data, different jobs: the tail of a life’s evidence curve rejects the fabricated file; the onset of that curve admits the real newcomer. One computation, both errors, one packet an examiner can replay.

• ● Life-process claim : full fit
• ● Paired error costs : full fit
• ● Examiner demand : full fit
• ● Data residency : full fit
• ● Industrial rings : full fit

A deposit account is a claim that a life stands behind it. Mule networks — synthetic and rented identities opened in batches to move scam proceeds and stolen-check money — are the purest ring problem in finance. Under the Customer Identification Program rules, institutions must form and document a reasonable belief that they know who their customer is; the evidence packet is that documentation, natively. For credit unions there is a second claim hiding in plain sight: field-of-membership eligibility is a residency assertion regulators expect to be verified.

Same data, different jobs: the ring of accounts that share an impossible household is refused; the unbanked household opening its first account is welcomed with evidence instead of suspicion.

• ● Life-process claim : full fit
• ● Paired error costs : full fit
• ● Examiner demand : full fit
• ● Data residency : full fit
• ◐ Industrial rings : partial fit

In auto rate evasion, the lie is spatial — which makes it our cleanest single fit. Garaging misrepresentation claims the car sleeps in a cheap-premium ZIP while the policyholder’s life says otherwise. And the examiner demand just arrived: Colorado’s external-data regime and the NAIC’s model AI bulletin now require carriers to demonstrate that external data and algorithms treat policyholders fairly — a proof mandate with no incumbent answer. One honest mark against this market: garaging is mostly individual misrepresentation rather than industrial rings, so the fifth test scores partial.

Same data, different jobs: the misstated garage is corrected; the night-shift nurse driving at 3 a.m. is proven fairly priced, not flagged as anomalous.

• ● Life-process claim : full fit
• ● Paired error costs : full fit
• ◐ Examiner demand : partial fit
• ● Data residency : full fit
• ● Industrial rings : full fit

Carriers finance a $1,200 handset for every subscriber they approve — a credit decision wearing a phone plan. The deployment topology here is the most frictionless we have: the decisive evidence is the carrier’s own network exhaust, so the computation runs where the data already lives and nothing sensitive moves at all. The honest partial: telecom buyers answer to their own P&L more than to a bank-style examiner, so the proof mandate is softer — the value case leads with loss, not compliance.

Same data, different jobs: the synthetic subscriber is stopped before the device ships; the legitimate prepaid customer with three years of coherent history graduates to postpaid on evidence instead of a bureau file they never had.

• ● Life-process claim : full fit
• ● Paired error costs : full fit
• ● Examiner demand : full fit
• ● Data residency : full fit
• ● Industrial rings : full fit

This is the longest arc and the deepest fit. Pandemic unemployment insurance proved the failure mode at national scale. But the morally load-bearing half is the other job: the same era proved that blunt identity gates wrongfully lock out exactly the people benefits exist for. A system that abstains and escalates when evidence is thin — rather than silently rejecting — is the difference between program integrity and program cruelty.

Same data, different jobs: the interstate fraud ring is stopped; the eligible claimant with a disrupted paper trail is paid, with the agency holding replayable proof of both decisions.

• ● Life-process claim : full fit
• ● Paired error costs : full fit
• ● Examiner demand : full fit
• ◐ Data residency : partial fit
• ● Industrial rings : full fit

The newest claim type, and the starkest: is the person on payroll the person — and the place — they say they are? U.S. prosecutors have documented North Korean IT workers, operating through stolen identities and domestic laptop farms, employed across hundreds of American companies — 309 in a single case, generating $17 million for a sanctioned regime — with the broader scheme estimated by the Department of Justice to channel hundreds of millions of dollars a year to weapons programs.

A claimed Ohio residence against an actual pattern of life is our test, verbatim; sanctions exposure supplies the examiner. The honest partial: employer-side data alone is thinner than carrier or bureau channels, so this market matures as consortium and carrier partnerships do. The other job matters here too — clearing the genuine remote hire over a bounded historical window, under a documented purpose, instead of subjecting every employee to standing surveillance.

Same data, different jobs: the laptop farm is exposed by what a real life would have left behind; the legitimate remote worker is cleared without a camera ever pointing at them.

The template has a sixth test we apply silently and absolutely: the deployment boundary’s social contract. Our architecture verifies claims a person has made — on an application, a policy, a payroll record — over a bounded historical window, under a documented legal basis, with abstention as a first-class outcome. It does not follow people.

So we do not build individual tracking or real-time surveillance products, tools for immigration enforcement targeting, election or protest policing, or social scoring of any kind — for any customer, at any price. These refusals are not caution; they are what makes the rest of the map trustworthy.

An evidence layer that would testify to anything is not evidence.

The figures on this page are third-party estimates of annual problem size — the loss pools our buyers carry — not Kenshiki forecasts and not addressable software spend, which is a fraction of any loss pool. Sources differ in scope, geography, and method; we cite each beside the number it backs and welcome corrections. Figures last verified June 2026.