Company Alpha
Report
Draftv0.4 · Generated Oct 6, 2026 · Prepared by M. OrtegaDemo · Sample investigation
Executive summary
Company Alpha's public statements on human review of credit decisions are not supported by observable evidence.
8 claims were examined against 142 evidence items. 3 are verified, 2 are contradicted and 3 remain unresolved. 4 require human review.
- Claims
- 8
- Verified
- 3
- Contradicted
- 2
- Unresolved
- 3
Verified findings
3- VerifiedCLM-002Customer chat assistant discloses that it is an AI system at the start of each session.Product documentation · 2 supporting, 0 contradicting
- VerifiedCLM-006Users can request human escalation at any point in an automated flow.Help center · 2 supporting, 0 contradicting
- VerifiedCLM-008Automated decisions can be explained to affected customers on request.Customer terms, §9 · 1 supporting, 0 contradicting
Contradictions
2Unresolved questions
8Claims not yet settled
- Insufficient evidenceCLM-003Model outputs are logged and retained for 24 months.Trust center statement · 1 supporting, 0 contradicting
- Partially verifiedCLM-005Fraud-detection model is re-evaluated against a fixed benchmark every quarter.Annual report, risk section · 2 supporting, 0 contradicting
- UnresolvedCLM-007Third-party models are assessed before production use.Procurement policy · 1 supporting, 0 contradicting
Open questions
- 01Which step, if any, reviews credit decisions before approval?
- 02How many automated approvals in the last quarter had a reviewer record?
- 03Which datasets were used to train the assistant, and how were they filtered?
- 04Where are model outputs stored, and who can read them?
- 05Which third-party models are in production and when were they assessed?
Evidence gaps
4- CLM-003Evidence strength 22%. More independent sources needed.
- EV-0005Trust center retention statement. Integrity check pending.
- EV-0006Procurement policy excerpt. Integrity check pending.
- EV-0009Model output containing customer-like data. Flagged for integrity review.
Required organizational contrast
4- CLM-001Explain the evidence that contradicts: “All automated credit decisions are reviewed by a human before final approval.”
- CLM-004Explain the evidence that contradicts: “Training data excludes customer-identifiable information.”
- CLM-003Provide records supporting: “Model outputs are logged and retained for 24 months.”
- CLM-007Provide records supporting: “Third-party models are assessed before production use.”
Internal verification recommended
Recommended once the organization authorizes access.
- Decision logs for credit approvals, last two quarters
- Reviewer assignment records for automated decisions
- Training data lineage for the customer assistant
Human review
5- EscalatedCLM-001All automated credit decisions are reviewed by a human before final approval.M. Ortega
- In reviewCLM-003Model outputs are logged and retained for 24 months.A. Okafor
- In reviewCLM-004Training data excludes customer-identifiable information.M. Ortega
- In reviewCLM-005Fraud-detection model is re-evaluated against a fixed benchmark every quarter.S. Marlowe
- Not assignedCLM-007Third-party models are assessed before production use.Not assigned
Evidence appendix
142- EV-0001Chat assistant session capture9f3a1c7e52…Intact
- EV-0002Help center article snapshot4be07d1a9c…Intact
- EV-0003Credit decision flow, timing analysisc81d5e0a7f…Intact
- EV-0004Annual report, risk section7a2e93c0b5…Intact
- EV-0005Trust center retention statemente50b8a42d7…Pending
- EV-0006Procurement policy excerpt1d94c7b3e8…Pending
- EV-0007AI policy, human review clauseb36f0d9a15…Intact
- EV-0008Escalation flow walkthrough6c1a8f3d02…Intact
- EV-0009Model output containing customer-like dataf20d6b9e4a…Flagged
- EV-0010Customer terms, §983e5a1c7f0…Intact
- EV-0011Approval notice, no review step0a7c4e91b3…Intact
- EV-0012Benchmark reference, vendor deckd9b2f5e07a…Intact
Showing 12 of 142 items. The full appendix is included in the exported report.