01
Consent and intake
The lender defines the applicant claim and allowed evidence boundary.
Product proof
How Kenshiki turns source evidence into a reviewable credit record
This is the public product-proof view of the workflow: bounded questions, source-side checks, measured validation artifacts, policy branches, human attestation, and a record a lender can replay.
Guide paths
Start with the underwriting question, then open the deeper guide.
How It Works is the anchor for the product method. The guides below keep each buyer question in its own canonical page while showing where that question enters the Kenshiki workflow.
- Thin-file fit
Verifying thin-file borrowers
Use when the applicant is real and qualified, but the credit file has too little history to carry the decision alone.
Open guide - Identity risk
Detecting synthetic identity fraud
Use when a clean file may hide whether a real, continuous life stands behind the applicant at all.
Open guide - Governable evidence
Alternatives to credit scoring
Use when the question is how to move beyond opaque scores without replacing one black box with another.
Open guide - Model risk
SR 26-2 and model validation
Use when validators or examiners will ask how a third-party evidence input was validated — what the 2026 guidance expects, what it leaves out, and where inspectable evidence fits.
Open guide
Workflow
The product path is evidence first, policy second, reviewer always.
Kenshiki does not ask a lender to trust a black-box recommendation. It narrows the question, asks approved source environments for bounded answers, and preserves what happened before a reviewer records a disposition.
-
-
02
Identity gate
Consent-based eCBSV match/no-match runs before corroboration is trusted.
-
03
Source checks
Approved providers answer bounded questions without sending raw feeds.
-
04
Policy branch
Versioned rules decide whether evidence clears, refers, or stops the file.
-
05
Attested record
A reviewer records the disposition and preserves a replayable ledger entry.
Evidence axes
Each axis answers one part of whether the file and the life cohere.
The public product proof keeps the seven-axis story inspectable: what the signal helps separate, what boundary keeps it from becoming a raw-data dragnet, and where validation must measure dependence rather than assume independence.
-
Axis 1
Spatiotemporal presence
Asks a provider-side area-and-time question without acquiring raw location data.
- Helps separate
- A real but quiet file from an identity assembled for one transaction.
- Boundary
- Kenshiki does not buy, ingest, or store raw location trails.
-
Axis 2
Carrier-attested telecom
Checks whether line tenure and device signals cohere with the claimed identity.
- Helps separate
- Seasoned personal continuity from newly assembled contact infrastructure.
- Boundary
- Carrier signals corroborate continuity; they do not expose message content.
-
Axis 3
Address and residency
Checks whether address and residency signals support the claimed life pattern.
- Helps separate
- A real residence from a mailbox, borrowed address, or thin surface match.
- Boundary
- Residency corroboration supports context; it is not lifestyle scoring.
-
Axis 4
Income at source
Checks whether income evidence can be verified at an authoritative source.
- Helps separate
- A stale risk snapshot from a materially recovered income picture.
- Boundary
- Income evidence supports capacity review; it is not an automatic approval.
-
Axis 5
Open-banking cash flow
Checks whether account history supports the rhythm of the requested obligation.
- Helps separate
- A bounded hardship from an ongoing inability to support the obligation.
- Boundary
- Cash-flow fit avoids merchant-level storytelling and stays bounded to evidence.
-
Axis 6
Identity-element trace
Tests whether identity elements have a seasoned, continuous history.
- Helps separate
- A valid-looking profile from a lived identity with durable history.
- Boundary
- Identity continuity complements eCBSV; it does not replace consented verification.
-
Axis 7
Behavioral biometrics
A candidate signal for human-interaction and device-consistency checks.
- Helps separate
- Human application behavior from scripted or ring-linked activity.
- Boundary
- This is candidate scope and must remain status-labeled until approved.
Policy layer
The recommendation is not the index with labels attached.
After evidence is admitted, a versioned policy layer evaluates missingness, contradictions, gates, and coherence. A strong record can clear; an incomplete record can refer; any contradictory axis blocks approval and routes to review or stop.
Clear
Evidence supports the allowed question.
The branch can move forward when required gates clear and corroboration is strong enough for the lender policy.
Refer
Missingness or ambiguity needs review.
Material uncertainty remains visible so a reviewer can ask for more evidence or route to senior review.
Stop
A controlling contradiction fails.
A policy gate can stop the file when admitted evidence contradicts a required claim or identity boundary.
Attestation
The reviewer records a disposition; Kenshiki preserves the evidence trail.
The workflow ends with a human-attested record: what was asked, what was admitted, what branch applied, what the reviewer decided, and what reason should be available later.
The human remains the decision-maker. The system makes the evidence easier to open, explain, and replay.
Static map preview
A real life leaves context without turning context into surveillance.
The public page shows the map concept as a static explanation; the interactive map lives in the demo workspace. The full walkthrough — boundaries, synthetic-risk patterns, thin-file fit — is on Corroboration.
Static corroboration map preview
Green cells show a coherent source-side pattern. The isolated warning cell shows the kind of flat signal that needs review.
Home pattern
Residence continuity
Work routine
Source-side fit
Life rhythm
Context coherence
- Live pattern
- Multiple bounded source contexts cohere.
- Synthetic pattern
- Activity collapses to one unsupported claim.
FAQ
Common questions
What the product workflow decides, what it deliberately does not, and how the validation artifacts should be read.
- Does Kenshiki make the final credit decision?
- No. Kenshiki returns bounded evidence, policy branches, and reason context a lender can inspect. The lender keeps its credit policy and final decision authority.
- Is the Corroboration Confidence Index a credit score?
- No. The index summarizes admitted evidence strength. It is not a measure of creditworthiness and does not replace adverse-action, underwriting, or fair-lending review.
- What happens when evidence is missing or contradictory?
- The policy layer can refer the file for manual review, stop a controlling contradiction, or keep the missingness visible in the record instead of hiding it in a score.
- Are the corroboration axes statistically independent?
- Independence is measured, not assumed. The current validation artifact reports correlation and effective-axis-count estimates on the 60-case seed corpus as methodology evidence, not production validation.