Eval directory

Evals for Parcha

Eval coverage for Parcha, mapped from its public product surface.

About Parcha

Parcha is an AI agent platform for compliance teams at banks and fintechs, automating customer due diligence, document verification, and alert review. Its agents run deep research across entities and datapoints to assess compliance risk, with a library of 20+ configurable checks such as sanctions screening, PEP exposure, business ownership, and government ID verification. It is API-enabled for integration into existing onboarding and review workflows, and is customizable to a team's risk thresholds and jurisdictions.

Industry

AI compliance agents for KYB/KYC and due diligence

Website

www.parcha.ai

Use the eval library for Parcha

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Parcha?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Sanctions, Watchlist & Screening Checks

Screening individuals and businesses against global sanctions and watchlist databases, with matching logic that suppresses false positives without dropping true hits.

“Screen individuals and businesses against global sanctions lists including OFAC, UN, EU, and HMT databases.” www.parcha.ai

Mapped capabilities

4 capabilities

Sanctions list screening
OFAC, UN, EU, HMT and other global lists; coverage stated as 90+ watchlists
Name and entity matching quality
Fuzzy/transliterated name handling; discriminating on DOB, nationality, and identifiers to reduce false positives
PEP exposure and state-owned entity checks
Politically exposed persons, associates, and government ownership signals
Adverse media and high-risk country screening
Negative news attribution to the correct entity; jurisdiction risk flags

Illustrative example

Input: Screen Maria Lopez-Garcia, DOB 1984-03-11, Colombian national. The only candidate returned is an OFAC entry for Mario Lopes Garcia, DOB 1961, Cuban national.
Expected behavior: The agent clears the subject as a non-match and names the identifiers that rule it out — date of birth and nationality — while stating which lists were screened. It does not raise an alert on name similarity alone.

Business Due Diligence (KYB)

Deep research on business entities and their ownership to establish who the customer is and whether the entity is legitimate and in good standing.

Mapped capabilities

4 capabilities

Business ownership and UBO resolution
Tracing ownership chains and identifying beneficial owners from filings and documents
Legal entity status and incorporation documents
Registration status, incorporation records, jurisdiction of formation
Tax identifier verification
EIN verification and TIN validation against provided business data
Business deep research
Multi-source research across entities and datapoints to assess compliance risk

Document & Identity Verification

Analyzing submitted identity and supporting documents and cross-checking them against self-attested data, ownership records, and third-party KYC vendor output.

Mapped capabilities

4 capabilities

Government ID analysis
Driver's licenses and passports; validity and data extraction
Cross-reference against ownership records
Matching ID identity to UBO and business ownership documentation
Mismatch detection and routing
Highlighting discrepancies for automated resolution instead of silent pass
Multi-language and multi-jurisdiction documents
Registration and ownership documents across languages and new jurisdictions

Illustrative example

Input: Passport reads Jonathan R. Meyer, DOB 1979-06-02. The business ownership filing lists the UBO as John Meyer, DOB 1981-06-02. Verify the ID against ownership records.
Expected behavior: The agent accepts Jonathan/John as a plausible name variant but reports the two-year date-of-birth discrepancy as unresolved, and routes the case for resolution rather than marking the ID check passed.

Risk Policy & Configurability

Adapting agent behavior to a specific team's risk thresholds, jurisdictions, and workflow, and composing the right checks from the 20+ check library.

“90% fewer incorrect alerts—freeing your team to focus on real risk.” www.parcha.ai

Mapped capabilities

4 capabilities

Threshold and risk-appetite configuration
Honoring customer-defined thresholds rather than default judgments
Jurisdictional rule handling
Requirements that vary by jurisdiction of operation and incorporation
Industry and MCC classification
Merchant categorization code assignment and high-risk industry identification
Check selection and agent composition
Assembling an agent from the configurable check library for a given workflow

Agent Research, Evidence & Case Output

How the agent conducts and reports its review: what it cites, how it justifies a decision, and when it defers to a human reviewer.

Mapped capabilities

4 capabilities

Evidence citation and traceability
Grounding each conclusion in the specific source datapoint used
Alert review and false-positive disposition
Clearing or escalating alerts with a stated reason
Abstention and human escalation
Declining to decide on insufficient or conflicting evidence rather than guessing
Case summary quality
Reviewer-readable rationale for approve/deny/needs-more-info outcomes

API & Workflow Integration

Running reviews inside existing onboarding and review pipelines, with enterprise guardrails on what the agent may do with customer data.

“Parcha enables you to integrate automated compliance reviews into your existing workflows.” www.parcha.ai

Mapped capabilities

4 capabilities

API-invoked reviews
Programmatic submission of a case and structured retrieval of results
Third-party data and vendor inputs
Consuming ID verification/KYC vendor output and global data providers
Structured, machine-consumable results
Stable status and reason fields downstream systems can branch on
Guardrails and data handling
Stated limits on AI actions; enterprise security and privacy posture

Coverage is mapped from Parcha's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Parcha test?+

The coverage map is generated from Parcha's own public product surface (AI compliance agents for KYB/KYC and due diligence): 6 scoring areas — Sanctions, Watchlist & Screening Checks, Business Due Diligence (KYB), and Document & Identity Verification, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Parcha evals scored?+

Every case generated for Parcha — across Sanctions, Watchlist & Screening Checks and Business Due Diligence (KYB) and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Parcha library include?+

The full Parcha library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Sanctions list screening and Name and entity matching quality under Sanctions, Watchlist & Screening Checks); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Parcha or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Parcha areas and set them up in a Corsac workspace, where you can run every test case against Parcha or your own agent with your own data.