Eval directory

Evals for Endex

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

About Endex

Endex is an Excel-native AI agent that accelerates financial modeling and data analysis for financial firms. It uses custom-built models designed for financial documents, unifies internal files and external data into a single searchable plane, and attaches citations to its outputs. It is sold as enterprise deployments with security controls including SOC 2, ISO 27001, encryption, audit logging, and optional VPC deployment.

Industry

Excel-native AI agent for financial modeling and analysis

Website

www.endex.ai

Use the eval library for Endex

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 Endex?

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

Excel-Native Modeling Agent

Agent behavior inside the Excel surface: building and extending financial models in a live workbook rather than returning prose. Grounded in the product's positioning as an Excel-native agent that accelerates financial modeling and data analysis, including customer-reported three-statement model and DCF construction.

“An Excel-native AI Agent that accelerates financial modeling and data analysis” endex.ai

Mapped capabilities

4 capabilities

Three-statement model construction
Builds linked income statement, balance sheet, and cash flow structure in-workbook.
DCF and valuation build-out
Assembles discounted cash flow schedules with explicit, traceable assumption cells.
Formula and cell-reference integrity
Writes live formulas with correct references rather than hardcoded or broken values.
Data analysis over existing workbooks
Reads, summarizes, and extends analysis on a workbook the user already has open.

Financial Document Understanding

Finance-specific comprehension claimed for the custom-built models: detailed accounting methods, reconciling disagreeing sources, and complex financial charts. This area tests whether domain nuance survives extraction into a model.

Mapped capabilities

4 capabilities

Accounting method nuance
Handles the accounting treatment stated in the source rather than a generic default.
Reconciling disagreeing sources
Surfaces and explains conflicts when two documents state different figures.
Complex financial chart and table reading
Extracts values from dense financial exhibits, footnoted tables, and charts.
Figure extraction fidelity
Transfers units, periods, and signs from document to workbook without distortion.

Illustrative example

Input: The audited financials and the management summary report different FY2024 revenue. Build the revenue line for my model.
Expected behavior: The agent detects the discrepancy, states both values with their sources, and either uses the audited figure with that choice made explicit or asks which to use. It does not average, pick silently, or present a single unqualified number.

Citations and Auditability

Integrated citations that augment outputs with attribution to underlying sources, supporting the 'auditable from start to finish' claim. The decision-useful question is whether every asserted number is traceable back to a specific source.

“Endex provides integrated citations, augmenting outputs with attribution to underlying sources.” endex.ai

Mapped capabilities

4 capabilities

Attribution coverage of asserted figures
Each output figure carries a citation to the source it came from.
Citation-to-source accuracy
The cited location actually contains the referenced value.
Uncited-claim restraint
Declines or flags assertions it cannot attribute to a supplied source.
Audit trail through the workflow
Attribution persists from retrieval through the delivered workbook output.

Illustrative example

Input: Using the attached 10-K and the board deck in my folder, pull FY2024 and FY2023 revenue into a comparison tab and cite where each figure came from.
Expected behavior: The agent writes both years' revenue into the workbook and attaches a citation to each figure identifying the specific source document and location. Any figure it cannot attribute is flagged rather than stated as fact.

Unified Data Plane and Retrieval

Unification of internal files, external data, and trusted sources into a single searchable plane so the agent has a complete picture. Tests retrieval selection and grounding across mixed-provenance corpora.

“Endex unifies your internal files, external data, and trusted sources into a single searchable plane” endex.ai

Mapped capabilities

4 capabilities

Internal file retrieval
Finds the right firm document among a mixed internal corpus.
Internal plus external synthesis
Combines firm files with external or trusted-source data in one answer.
Source provenance labeling
Distinguishes internal, external, and trusted-source material in its output.
Coverage gap acknowledgment
States when the searchable plane lacks data needed for the request.

Enterprise Security and Data Governance

Security posture presented to buyers: SOC 2, ISO 27001, CCPA and GDPR, 256-bit AES encryption at rest and in transit, zero-trust least-privilege role-based access, audit logging, no training on customer data, and optional VPC or client-cloud deployment.

“By default, your data is not shared, used for training AI models, or made accessible to other firms” endex.ai

Mapped capabilities

4 capabilities

Data-use and no-training representations
Answers about training, sharing, and cross-firm access match stated defaults.
Role-based access boundaries
Respects least-privilege scoping when serving a restricted user.
Compliance claim accuracy
Represents SOC 2, ISO 27001, CCPA, and GDPR status without overstatement.
Deployment and hosting options
Describes private cloud, AWS/Azure hosting, and VPC options accurately.

Public Intake and Site Surfaces

Pre-deployment surfaces a visitor actually touches: contact and custom-deployment requests, the support request form, the waitlist, the blog, and Website Terms of Use scoped to the site rather than the product.

Mapped capabilities

4 capabilities

Deployment and demo request routing
Directs custom-deployment interest to the correct sales intake path.
Support request handling
Routes help requests through the support form rather than improvising.
Waitlist versus enterprise path
Separates waitlist signup from enterprise deployment conversations.
Terms scope boundary
Distinguishes site Terms of Use from separate Product Agreements governing app.endex.ai and the Excel add-in.

Coverage is mapped from Endex'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 Endex test?+

The coverage map is generated from Endex's own public product surface (Excel-native AI agent for financial modeling and analysis): 6 scoring areas — Excel-Native Modeling Agent, Financial Document Understanding, and Citations and Auditability, 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 Endex evals scored?+

Every case generated for Endex — across Excel-Native Modeling Agent and Financial Document Understanding 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 Endex library include?+

The full Endex library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Three-statement model construction and DCF and valuation build-out under Excel-Native Modeling Agent); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

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

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