Eval directory

Evals for Tower

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

About Tower

Tower is an AI-powered due diligence platform that gives each asset or transaction a dedicated workspace for collecting, organizing, and analyzing unstructured deal documents. It syncs record systems, communications, and disclosures into one place, bulk-organizes and renames data rooms, and answers questions across thousands of documents with structured, cited output. It is positioned for M&A and legal diligence workflows across financial, legal, corporate, customer/revenue, and employment matters.

Industry

AI due diligence software for M&A and legal teams

Website

www.withtower.com

Use the eval library for Tower

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 Tower?

6 scoring areas · 24 capabilities mapped · grounded in 4 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

Collection & Source Sync

Getting disconnected record systems, communications, and disclosures into one workspace per asset or transaction, including how outstanding diligence requests are tracked while material is still arriving.

“Every asset or transaction gets a dedicated workspace.” www.withtower.com

Mapped capabilities

4 capabilities

Integration-based sync of record systems
Pulling documents from connected systems into the workspace without dropping or silently truncating sets.
Communications and disclosure intake
Ingesting correspondence and disclosure material alongside formal deal documents.
Diligence request management
Tracking what was requested, what arrived, and what is still outstanding against the request list.
Workspace scoping per asset or transaction
Routing incoming material to the correct dedicated workspace rather than a shared pool.

Data Room Organization & Naming

Bulk restructuring and renaming of entire data rooms or client document sets under a customer-supplied convention, including the files that resist automatic identification.

“Organize and rename entire data rooms or sets of client documents with custom naming conventions and folder structures.” www.withtower.com

Mapped capabilities

4 capabilities

Custom naming convention application
Applying a specified filename template with the correct field values pulled from each document.
Folder structure and taxonomy assignment
Placing documents into the customer's folder scheme by document type.
Bulk operations at data-room scale
Consistency of naming and filing when the operation spans an entire data room, not a handful of files.
Unlabeled and ambiguous file handling
Behavior on opaque filenames (e.g. doc_final_v2.pdf) where fields cannot be recovered.

Illustrative example

Input: Rename every file in /Leases to 'Lease Agreement - {property address} - {counterparty} - {execution date}.pdf'. One of the files is a scan named doc_final_v2.pdf.
Expected behavior: Renames each file whose address, counterparty, and date are all recoverable from the document itself. Leaves the unidentifiable scan under its original name and flags it for manual review rather than emitting a name with invented or placeholder fields.

Cross-Document Extraction & Structured Output

Asking one question against thousands of documents and getting back an answer in the requested structure, with the citation trail that makes it checkable.

“Ask questions to thousands of documents at once and receive structured answers in your preferred format” www.withtower.com

Mapped capabilities

4 capabilities

Multi-document question answering at scale
Coverage and consistency of an extraction that must sweep the whole document set.
Output in the requested format
Honoring a specified table, schema, or field layout rather than returning prose.
Per-field citation and verifiability
Each extracted value traceable to the specific source document it came from.
Absence and non-answer handling
Marking a field as not found instead of filling it with a plausible guess.

Illustrative example

Input: Across every agreement in this data room, give me a table of document name, whether there is a change-of-control provision, and the cap on liability.
Expected behavior: Returns one row per agreement with the two fields populated and a citation to the source document behind each value. An agreement that states no cap is marked as uncapped; one where the clause is not found is marked not found, not guessed.

Contract Provision Review (Legal Matters)

The provision-level comparison work the product demonstrates: reading the same clause questions across different agreement types and reporting them side by side.

“Any violation of these Terms may result in immediate termination of your access rights” www.withtower.com

Mapped capabilities

4 capabilities

Change-of-control provisions
Detecting presence, and the trigger conditions, across agreement types.
Limitation and cap on liability
Extracting cap amounts and distinguishing an uncapped agreement from an unstated one.
Termination rights
Identifying termination provisions and who may exercise them.
Cross-agreement-type comparison
Applying one provision matrix consistently across service, lease, and vendor agreements.

Diligence Matter Domains

The non-legal matter areas Tower names as first-class workstreams, each with its own vocabulary and its own idea of what a red flag looks like.

Mapped capabilities

4 capabilities

Financial matters
Extraction and organization over financial deal documentation.
Customer and recurring revenue
Customer contracts and recurring revenue terms as a diligence workstream.
Corporate and shareholders
Corporate records, ownership, and shareholder documentation.
Employment matters
Employment agreements and related workforce documentation.

Confidentiality & Access Boundary

The data room is governed by access terms and a privacy policy, so the workspace boundary is a product surface: answers should stay inside the workspace they were asked in, and access should behave as granted or terminated.

“Sync disconnected record systems, communications, and disclosures into a unified workspace via intelligent integrations and smart diligence request management.” www.withtower.com

Mapped capabilities

4 capabilities

Workspace isolation of answers
Extractions cite and draw only from documents in the current workspace.
Data room access rights and termination
Behavior when a user's access has been revoked or was never granted.
Handling of personal data in disclosures
Treatment of personal data encountered inside deal documents.
Confidential material in outputs
Restraint when an output would carry confidential content outside its intended scope.

Coverage is mapped from Tower's public pages (4 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 Tower test?+

The coverage map is generated from Tower's own public product surface (AI due diligence software for M&A and legal teams): 6 scoring areas — Collection & Source Sync, Data Room Organization & Naming, and Cross-Document Extraction & Structured Output, 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 Tower evals scored?+

Every case generated for Tower — across Collection & Source Sync and Data Room Organization & Naming 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 Tower library include?+

The full Tower library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Integration-based sync of record systems and Communications and disclosure intake under Collection & Source Sync); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

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

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