# Faultprint Counterexample Curation 001

## Purpose

This layer reviews counter-signal candidates for the five strongest pathology modules. It does not claim that every candidate is a true counterexample. It separates stronger counterexample candidates from weaker caveats, absence rows, and low-resemblance signals.

## Summary

|pathology|review rows|strong candidates|control signals|absence signals|low-resemblance signals|weak signals|
|---|---|---|---|---|---|---|
|Contestability Lag|206|4|69|120|3|10|
|AI Authority Acceleration|42|4|19|4|7|8|
|Record-Authority Capture|23|2|5|1|1|14|
|Access Interruption Before Review|12|0|5|2|0|5|
|Documentation Burden Transfer|11|1|1|0|0|9|

## Contestability Lag

|row|tier|institution|sector|anchor|
|---|---|---|---|---|
|OCL-0022|strong_counterexample_candidate|Workday Responsible AI|HR technology|Workday describes alternative procedures including human review and optionality for data subjects.|
|OCL-0028|strong_counterexample_candidate|UnitedHealthcare AI and Prior Authorization|Health insurance|The page describes appeal review by people who were not involved in the original decision.|
|OCL-0080|strong_counterexample_candidate|Cigna Healthcare|Health insurance|Cigna describes a three-step appeal/review process and says external independent review may be available for certain coverage decisions.|
|OCL-0362|strong_counterexample_candidate|Axon Draft One|Public safety / police reporting AI|Axon states officers review edit and sign off on Draft One reports before submitting them into Axon Records or a third-party RMS for supervisor review.|
|OCL-0010|control_counter_signal|Lemonade|Insurance|The blog says AI Jim assesses claim nature severity emergency state fraud likelihood and whether the claim is handled automatically or by a human.|
|OCL-0027|control_counter_signal|UnitedHealthcare AI and Prior Authorization|Health insurance|The page describes coverage determinations prior authorization denials appeals and timing rules.|
|OCL-0073|control_counter_signal|Aetna|Health insurance|Aetna states members may request a coverage decision and may appeal if the answer is no.|
|OCL-0074|control_counter_signal|Aetna|Health insurance|Aetna states certain coverage denials can be reviewed by independent physician reviewers after plan appeal process exhaustion.|

## AI Authority Acceleration

|row|tier|institution|sector|anchor|
|---|---|---|---|---|
|OCL-0003|strong_counterexample_candidate|Fannie Mae|Mortgage underwriting|Lenders may use automated or manual underwriting; DU Validation Service is optional and borrower participation may not be required.|
|OCL-0021|strong_counterexample_candidate|Workday Responsible AI|HR technology|Workday states humans remain final decision makers for critical decisions.|
|OCL-0094|strong_counterexample_candidate|HireVue|Employment assessment|HireVue states AI assessments do not replace recruiters and instead help recruiters and talent teams assess candidates.|
|OCL-0238|strong_counterexample_candidate|Workday|AI governance / HR technology|Workday states humans remain final decision-makers for critical decisions and describes human-in-the-loop design philosophy.|
|OCL-0092|control_counter_signal|HireVue|Employment assessment|HireVue states assessments provide a structured way to screen many candidates faster and at lower cost than traditional interviews.|
|OCL-0222|control_counter_signal|Microsoft|AI governance|The standard states Microsoft AI systems include capabilities that support informed human oversight and control.|
|OCL-0231|control_counter_signal|ServiceNow|AI governance|ServiceNow states human-centered AI helps users understand when/how to use AI and that users are always in control.|
|OCL-0271|control_counter_signal|Zest AI|AI lending / credit underwriting|Zest AI describes automating underwriting with AI to make smarter credit decisions faster while giving applicants a fair shot.|

## Record-Authority Capture

|row|tier|institution|sector|anchor|
|---|---|---|---|---|
|OCL-0039|strong_counterexample_candidate|State Farm Drive Safe & Save|Insurance telematics|State Farm states the discount may vary based on actual driving but will not add a surcharge to the policy.|
|OCL-0098|strong_counterexample_candidate|The Work Number / Equifax|Employment and income verification|The page states consumers can see what data is on The Work Number and see verifier names from the last 24 months.|
|OCL-0014|control_counter_signal|GEICO DriveEasy|Insurance telematics|GEICO states the app collects GPS Bluetooth sensor and driving-related data during trips.|
|OCL-0033|control_counter_signal|Progressive Snapshot|Insurance telematics|Progressive states users can check Snapshot progress trip logs driving tips and related information online or in the app.|
|OCL-0037|control_counter_signal|State Farm Drive Safe & Save|Insurance telematics|State Farm describes route maps and driving events including acceleration braking cornering and phone distraction.|
|OCL-0053|control_counter_signal|Root Insurance|Insurance telematics|Root public filing states a two-to-four week test drive gathers and analyzes smartphone sensor data measuring braking, consistency, turning, time of day, driver attentiveness, a...|
|OCL-0068|control_counter_signal|USAA SafePilot|Insurance telematics|USAA states the app will record phone-handling infractions even if it is not the policyholder using the phone.|
|OCL-0044|low_resemblance_counter_signal|Nationwide SmartRide|Insurance telematics|Nationwide states SmartRide measures driving behavior to reward discounts and not raise rates.|

## Access Interruption Before Review

|row|tier|institution|sector|anchor|
|---|---|---|---|---|
|OCL-0121|control_counter_signal|ID.me|Identity verification|ID.me instructs users to take a video selfie and start liveness detection during verification.|
|OCL-0144|control_counter_signal|California EDD / state UI agencies|Unemployment insurance|New Jersey states workers applying for UI must verify identity through ID.me before receiving unemployment benefits.|
|OCL-0264|control_counter_signal|Upstart|AI lending / credit underwriting|Upstart states it has methodology to explain AI model outcomes including approval denial and pricing outcomes and generate more explainable Adverse Action Notices.|
|OCL-0652|control_counter_signal|Transcript Withholding and Access|Education records / postsecondary institutions|Department fact sheet states regulations require colleges to clearly communicate how much financial aid students will receive.|
|OCL-0658|control_counter_signal|School Discipline Due Process|K-12 education / student discipline|Education Law Center states its toolkit helps families and advocates respond to suspension expulsion or alternative education placement efforts.|
|OCL-0125|absence_counter_signal|ID.me|Identity verification|The reviewed corpus does not include denial samples demographic error rates agency access outcomes appeal results or complaint-to-correction records.|
|OCL-0215|absence_counter_signal|Internal Revenue Service|Tax return identity verification|The reviewed corpus does not include taxpayer-specific false flags verification outcomes refund delay durations appeal/correction records or complaint-to-correction evidence.|
|OCL-0122|weak_counter_signal|ID.me|Identity verification|ID.me describes its Wallet as helping individuals discover and access benefits and services through verified identity.|

## Documentation Burden Transfer

|row|tier|institution|sector|anchor|
|---|---|---|---|---|
|OCL-0457|strong_counterexample_candidate|UiPath Agentic Automation|Enterprise agentic automation|UiPath states human-in-the-loop brings people in to validate and provide ultimate confirmation of key decisions.|
|OCL-0286|control_counter_signal|Microsoft / Nuance|Healthcare documentation AI|Microsoft materials describe DAX Copilot as AI-based note creation and Dragon Copilot as streamlining clinical documentation.|
|OCL-0054|weak_counter_signal|Root Insurance|Insurance telematics|Root states day-to-day driving behavior is the most important part of how Root calculates rates.|
|OCL-0124|weak_counter_signal|ID.me|Identity verification|ID.me says users may need to allow camera permission to start liveness detection.|
|OCL-0152|weak_counter_signal|FBI CJIS|Criminal history records|FBI states requests should include available proof or documentation such as court records or expungement orders.|
|OCL-0154|weak_counter_signal|FBI CJIS|Criminal history records|The handout states officials must provide written notification that fingerprints will be used to check FBI criminal-history records.|
|OCL-0493|weak_counter_signal|Tesla Full Self-Driving Supervised|Driver assistance / consumer vehicles|Tesla states the driver must remain attentive and be prepared to take immediate action, warning serious injury or death can result.|
|OCL-0506|weak_counter_signal|YouTube Recommendation System|Video platform recommendations|YouTube states its recommendation system aims to identify the most relevant content for each user at any given moment.|

## Use Rule

Only strong counterexample candidates should be used to challenge or weaken a pathology module. Absence rows and low-resemblance rows are useful for calibration, but they are not affirmative counterexamples without human review.