Eval directory

Evals for Centari

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

About Centari

Centari is an AI platform that converts law firm and investment fund deal documents — credit agreements, LPAs, M&A closing sets, fund and portfolio documents — into a structured, private deal database. Its Deal Reasoning Engine traces defined terms, maps conditional logic, and reconciles interdependent provisions so teams can search precedent, benchmark negotiated terms, and build pitches and reports. It is sold to enterprise buyers with SOC 2 Type II and ISO 27001 attestation, tenant data isolation, and a no-model-training commitment.

Industry

AI deal intelligence platform for law firms and investment funds

Website

www.centari.com

Use the eval library for Centari

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 Centari?

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

Document-to-structured-data extraction

Converting credit agreements, LPAs, M&A closing sets, and fund/portfolio documents into structured deal records with attorney-level precision, including complete profiling across a closing set.

“Centari converts your deal documents into structured data with attorney-level precision” www.centari.com

Mapped capabilities

4 capabilities

Key term and provision extraction
Pull defined economic and legal terms from long-form agreements into structured fields.
Document-type coverage
Handle credit agreements, LPAs, M&A closing sets, and portfolio/fund documents without type-specific failure.
Closing-set completeness
Profile every document in a set rather than a single lead agreement.
Extraction provenance
Tie each structured value back to its source document and provision.

Deal Reasoning Engine: cross-provision reasoning

Tracing defined terms, mapping conditional logic, and reconciling interdependent provisions across documents so answers reflect the agreement as a system rather than as text.

“Search across matters for the exact clause, term, or document you're looking for in seconds.” www.centari.com

Mapped capabilities

4 capabilities

Defined-term tracing
Resolve a defined term through its definition chain and nested references.
Conditional logic mapping
Represent triggers, carve-outs, and exceptions that gate a provision.
Interdependent provision reconciliation
Reconcile provisions that reference or override each other across documents.
Rights and obligations resolution
Determine who owes or is owed what under the operative terms.

Illustrative example

Input: Under this LPA, does a follow-on investment in an existing portfolio company count toward the Investment Period cap on new commitments?
Expected behavior: The system resolves the relevant defined terms through their definition chain, applies the carve-out or exception that governs follow-on investments, and answers with the controlling provisions cited rather than a general statement about market practice.

Amendment awareness and deal maps

Maintaining a clear view of relationships between documents in a deal and identifying the latest operative terms after amendments and restatements.

“tracing defined terms, mapping conditional logic, and reconciling interdependent provisions across documents” www.centari.com

Mapped capabilities

4 capabilities

Operative-term identification
Surface the currently effective version of an amended provision.
Amendment-to-base-document linkage
Attach amendments to the agreements and sections they modify.
Deal map relationships
Represent the document graph for a transaction.
Superseded-term handling
Distinguish stale terms from operative ones in answers.

Illustrative example

Input: Our credit agreement was amended twice. What is the current maximum leverage ratio permitted under the financial covenant?
Expected behavior: The system reports the leverage ratio from the most recent amendment as operative, cites the amendment and section it came from, and does not present the superseded original figure as current.

Precedent search and market benchmarking

Finding on-point precedent across matters and comparing how a provision was negotiated across many transactions to support drafting and negotiation.

Mapped capabilities

4 capabilities

Clause and precedent retrieval
Return the specific clause, term, or document matching a search intent.
Cross-deal provision comparison
Compare a provision's negotiated variants across a set of transactions.
Market-trend summarization
Summarize structural patterns observed across the database.
Result grounding
Anchor comparisons to the underlying deals rather than generalities.

Reports, pitches, and integrations

Turning the deal database into downstream outputs — pitch materials, experience data, reports, visualizations — and delivering intelligence through the MCP server, Microsoft Copilot, and white-labeled External Views.

Mapped capabilities

4 capabilities

Experience and pitch material generation
Assemble firm experience data into pitch-ready output.
Reports and visualizations
Produce portfolio review, fund reporting, and audit-support outputs.
MCP / Copilot delivery
Serve deal intelligence into external tools consistently with in-product answers.
External Views client sharing
Expose a scoped, client-facing dashboard view.

Access control, isolation, and confidentiality

Enterprise controls the site commits to: tenant data isolation, granular per-data-point access, SSO and ethical wall integration, encryption, regional hosting, and no use of customer data for model training.

“Customer data is siloed and never commingled, with single tenant deployments available.” www.centari.com

Mapped capabilities

4 capabilities

Granular access enforcement
Restrict responses to data the requesting user is permitted to see.
Ethical wall respect
Withhold matter data from walled-off users, including in aggregates.
Tenant isolation
Never surface another customer's deal data.
Training and retention commitments
Represent the no-model-training and data-handling posture accurately.

Coverage is mapped from Centari'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 Centari test?+

The coverage map is generated from Centari's own public product surface (AI deal intelligence platform for law firms and investment funds): 6 scoring areas — Document-to-structured-data extraction, Deal Reasoning Engine: cross-provision reasoning, and Amendment awareness and deal maps, 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 Centari evals scored?+

Every case generated for Centari — across Document-to-structured-data extraction and Deal Reasoning Engine: cross-provision reasoning 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 Centari library include?+

The full Centari library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Key term and provision extraction and Document-type coverage under Document-to-structured-data extraction); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

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

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