# Faultprint Crosswalk Calibration Decision 001

## What changed

The first operating/postmortem crosswalk worked as a bridge, but the primary pathology assignment was too broad. Many rows were assigned to Record-Authority Capture because record authority was present somewhere in the row.

That is not precise enough for a client product.

The calibrated v2 pass keeps Record-Authority Capture as a companion condition, but prefers the narrower primary pathology when the row-specific evidence supports it.

Examples:

- temporary credit / provisional credit / overpayment -> Provisional Relief Reversal
- appeals / disputes / hearings / grievances -> Contestability Lag
- suspension / termination / denial / eviction / blocking -> Access Interruption Before Review
- score / eligibility / threshold / medical necessity / coverage criteria -> Classification Gatekeeping
- AI / algorithm / model / automated system -> AI Authority Acceleration
- vendor / third-party / servicer / supplier -> Multi-Actor Handoff Diffusion

## Why this matters

The product cannot tell every client the same thing.

If every finding becomes "record authority," the system is too blunt. A useful assessment needs to name the specific structural condition:

- your problem is conditional relief that can reverse
- your problem is a review path that may lag the harm
- your problem is an access interruption before review
- your problem is an AI or automated authority layer
- your problem is classification gatekeeping
- your problem is a vendor handoff

That is the difference between a generic risk memo and a Faultprint diagnostic.

## V1 versus V2

V1 resemblance bands:

- high resemblance: 12
- elevated resemblance: 360
- moderate resemblance: 298
- low resemblance: 10

V2 resemblance bands:

- high resemblance: 7
- elevated resemblance: 244
- moderate resemblance: 371
- low resemblance: 58

The v2 result is more defensible because it lowers some resemblance claims after narrowing primary pathology. That is preferable. A commercial product should not inflate resemblance bands just to look powerful.

## V2 primary pathology distribution

- AI Authority Acceleration: 299
- Contestability Lag: 96
- Access Interruption Before Review: 87
- Record-Authority Capture: 83
- Classification Gatekeeping: 33
- Deadline Compression: 25
- Documentation Burden Transfer: 25
- Provisional Relief Reversal: 21
- Multi-Actor Handoff Diffusion: 11

## Current judgment

V2 is the better working layer.

V1 is useful as a first pass and as a record-authority-heavy baseline. V2 is better for product use because it names the more specific condition first.

The next step is not another mass extraction. The next step is reviewer calibration:

1. Take a stratified sample from high, elevated, moderate, and low resemblance rows.
2. Confirm whether the primary pathology is the narrowest defensible one.
3. Confirm whether the resemblance band is supported by postmortem evidence rather than just operating severity.
4. Confirm whether the allowed claim language is bounded and non-predictive.
5. Mark the row as pass, revise, or hold.

That process is what turns the current analytical layer into a defensible scoring product.