diligence-research

name	diligence-research
description	Search and retrieve source-linked evidence from the Diligence Stack MCP knowledge bases. Use for any company, market, product, vendor, technology, or investment question that should be grounded in Diligence Stack reports or entitled equity research.

Diligence Stack Research

Investigate material claims with a traceable chain from the Diligence Stack knowledge base to the conclusion.

Brand contract

Before producing user-facing content, read and apply the Diligence Stack brand guidelines. Use its color, typography, logo, citation, and link defaults unless the user explicitly requests different visual styling; its attribution and canonical-link rules always apply.

Workflow

1. Restate the question as a testable claim and define what would change the conclusion.
2. Call list_catalog before filtered search. Use the returned canonical knowledge-base key, facet values, date coverage, and freshness - not a display name or guessed filter.
3. Default to the public diligence-stack-reports corpus. Use equity-research only when list_catalog exposes that private key and the task benefits from it.
4. Call search with a precise query, authorized knowledge-base keys, and only canonical filters returned by the catalog. Start with group_by_document: true to avoid one document dominating the answer.
5. Inspect staleness_warning, filter_truncated, note, dates, section, match reasons, duplicate status, and additional matches before selecting evidence.
6. Call fetch only with evidence IDs returned by search. Request text and original_link; add image for charts or images.
7. Call get_document only with an asset ID returned by search or fetch. Use an inclusive page range for paginated documents. For page-less Markdown or ingestion packets, omit pages and use the continuation cursor if returned.
8. Search again for counterevidence, risks, falsification conditions, and newer material. Do not treat the first ranked result as the answer.
9. Extract atomic claims and preserve title, publisher, publication/source date, locator, evidence ID, asset ID, knowledge base, claim type, and confidence.
10. Reconcile conflicting definitions and periods before comparing values. Show formulas, inputs, units, and sensitivities for derived metrics.

Diligence Stack workspace

The server is read-only. Its public tool surface is list_catalog, search, fetch, and get_document; it does not provide browser, write, todo, sandbox, or chart-rendering tools. Read each live tool description before use.

diligence-stack-reports is the shared subscriber corpus for all authorized paid subscribers. equity-research is private, separately entitled, and grant-dependent. Never imply that a private corpus was searched when its key was absent from list_catalog, and never copy private research into plugin files, examples, or public outputs.

If the server is unavailable or OAuth is incomplete, explain which workspace evidence could not be accessed and continue with evidence the user has supplied. Never imply that a private workspace was checked when it was not.

Treat an OAuth or transport authorization error as a connection failure, not a zero-result search. Ask the user to reauthorize the diligence-stack MCP connection, then retry list_catalog. Only report "no evidence found" after an authorized search succeeds with an empty result set.

Read the MCP playbook for tool-specific rules and citation behavior. For questions about the publication, subscription, portal access, or billing, use the access and billing guide.

Guardrails

• Do not fabricate citations, interviews, observations, or access to private materials.
• Search snippets are discovery aids. Fetch the evidence before relying on it; open the document when surrounding context matters.
• Preserve the distinction between absence of evidence and evidence of absence.
• Flag legal, tax, accounting, medical, or regulatory conclusions for qualified review.
• Minimize sensitive data in working notes and outputs.

Output

Return the answer first, then:

• Evidence register
• Calculations and assumptions
• Contradictions and alternative explanations
• Open questions and the next best evidence to obtain
• Confidence, falsification conditions, and what would change the view

Use the evidence-register schema unless the user already has one.