# Substantive Testing > Run substantive attribute testing on audit samples. Use this when the user wants to create a test type, define attributes, upload supporting documents, execute tests against transaction samples, and export results to a local workbook. Use when this capability is needed. - Skill: `tomevault-io/substantive-testing` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/substantive-testing` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/substantive-testing/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Security - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/substantive-testing --- # Substantive Attribute Testing Guide the user through Tellen's substantive attribute testing workflow. This skill orchestrates the full lifecycle: creating test types, defining attributes, uploading documents, executing tests, and exporting results. ## Overview Substantive attribute testing validates individual transactions (samples) against defined audit attributes. Each attribute checks a specific assertion (e.g., "invoice date falls within the audit period", "vendor name matches the PO"). The Tellen AI agent reads uploaded supporting documents and evaluates each attribute, providing PASS/FAIL/EXCEPTION results with evidence citations. ## Workflow ### Step 1 — Choose or Create a Test Type Ask the user what they're testing. Common test types include: - **Disbursements / Accounts Payable** — Testing vendor payments against invoices, POs, and approvals - **Revenue / Accounts Receivable** — Testing sales transactions against contracts and shipping docs - **Payroll** — Testing payroll transactions against timesheets, rates, and approvals - **Journal Entries** — Testing manual journal entries for authorization and support - **Fixed Assets** — Testing asset additions/disposals against supporting documentation Use the Tellen API to check for existing test types or create a new one. **Current approach**: Direct the user to create the test type in the Tellen UI at **Substantive Testing → New Test Type**, then return here to continue with document upload and execution. ### Step 2 — Define Attributes Each attribute is a testable assertion. Help the user think through what they need to verify. Example attributes for a disbursements test: 1. **Invoice Date** — Invoice date falls within the audit period 2. **Amount Agreement** — Invoice amount agrees to the payment amount 3. **Vendor Match** — Vendor on the invoice matches the vendor on the PO 4. **Authorization** — Payment was authorized per the approval matrix 5. **GL Coding** — Transaction was coded to the correct GL account **Current approach**: Attributes are defined in the Tellen UI within the test type. Guide the user through the attribute definition process there. ### Step 3 — Upload Supporting Documents Use the `process_document` MCP tool to upload the user's supporting documents. These are the source materials the AI will reference when testing each sample. For each document the user provides: 1. Read the file from the user's local filesystem 2. Base64-encode the file contents 3. Call `process_document` with the file name, base64 content, and MIME type 4. Note the returned `file_id` — this links the document to the workspace Common document types: - Invoices (PDF) - Purchase orders (PDF, DOCX) - Receiving reports (PDF, XLSX) - Approval emails (PDF) - Bank statements (PDF) - Contracts (PDF, DOCX) Wait for each document to finish processing before proceeding. ### Step 4 — Execute the Substantive Test **Current approach**: Once documents are uploaded, direct the user to execute the test in the Tellen UI: 1. Go to the test type they created 2. Select the samples to test 3. Click **Run Test** to kick off the AI evaluation 4. Monitor progress in real-time as each attribute is tested The AI agent will: - Search uploaded documents for evidence relevant to each attribute - Verify evidence against the attribute's assertion - Return PASS, FAIL, or EXCEPTION with cited evidence and rationale ### Step 5 — Export Results **Current approach**: After tests complete, the user can export from the Tellen UI: 1. Go to the completed test 2. Click **Export Results** 3. Download the workbook (XLSX format) Help the user understand the export format: - One row per sample × attribute combination - Columns: Sample ID, Attribute, Result (Pass/Fail/Exception), Evidence, Rationale - Exception details include what evidence was missing ## Tips - **Document quality matters** — Cleaner, higher-resolution documents yield better AI analysis - **Be specific with attributes** — Vague attributes like "looks correct" produce unreliable results - **Combine with search** — Use `search_files` to spot-check specific items across uploaded documents - **Standards backup** — Use `research_us_gaap` if you need to verify the accounting standard behind an attribute --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/tellenai) — claim your Tome and manage your conversions.