Agenda Intelligence MD
A deterministic evidence-packet linter for claim-backed AI output.
It checks whether a claim is supported by the evidence you supply: broken references, mismatched quotations, unmatched numbers, missing evidence, and other evidence-pack integrity problems. It is designed for local workflows, agent pipelines, and CI jobs, and it does not require a model API key.
This project is not a fact-checker for real-world truth, not a source authenticator, and not permission to act. It validates the evidence relationship in the packet you provide.
Why this exists
When AI systems produce claims, it is often difficult to tell whether the source text actually supports the statement. Agenda Intelligence MD gives you a deterministic check on the evidence trail instead of trusting a model output at face value.
Use it to:
- validate AI-generated claims before publishing or acting on them
- review a RAG answer with inline citations and retrieved snippets
- lint evidence packets in CI or local development
- review supplier or document handoff packages with explicit evidence gaps
- attach output verification to agents and workflows
What this is
The core checker validates the supplied packet:
- claims resolve to source IDs that actually exist
- declared quotes match the provided source text
- numbers and wording are consistent with the supplied evidence
- evidence gaps and missing support are surfaced as findings
- packet status is reported in a deterministic way
What this is not
Agenda Intelligence MD does not:
- establish factual truth in the world
- verify source authenticity or provenance
- certify that a document is trustworthy or authorized
- grant permission to act or relay a decision
- replace human review for consequential actions
Choose your starting point
First run
Python 3.9 or later. Install the published package without cloning the repository:
Run a self-contained synthetic packet in Python:
Expected output:
This example needs no account, model API key, or network after installation. It checks the supplied evidence relationship, not source authenticity or factual truth.
To run the repository's CLI examples from a source checkout:
For the published package, use:
The bundled synthetic packet reports packet_status=packet_complete and factuality=not_assessed. A stale or inaccurate source can still pass. Add --strict when packet findings should fail a check instead of just reporting them.
For document-based review, start with the local-file review guide and example manifest:
For a RAG answer with inline citations and retrieved chunk texts, use the Output Verification adapter:
The original fictional answer routes to revision; the corrected answer routes to human review. Both run without a wallet, model API key, or network.
The evidence-packet contract
These checks use supplied text only. They do not retrieve missing sources or establish semantic entailment. Lexical overlap can miss paraphrases and accept misleadingly similar wording.
See docs/evaluation.md for tested behavior and limitations.
Interfaces
Compatibility profiles and adapters
The core checker is deterministic, stateless, and usable without a model API key. Optional generation and document adapters have separate dependencies.
Hosted Output Verification
Connect a Streamable HTTP MCP client to:
Refresh tools/list and call free_trial_status with {}. For an answer and source excerpts, choose review_answer or the optional open_answer_review form. For existing claim/evidence objects, use free_trial with your own input matching that tool's schema.
The local server exposes a 32-tool MCP catalog via stdio. The hosted worker exposes a different subset depending on the configured profile. Always discover tools from the endpoint you actually connect to.
Prepare a supplier document handover
For broader document archives, use the free Document Archive Review tool. It helps you:
- prepare a supplier or company evidence checklist
- assign owners for missing items
- attach authorized documents and source excerpts
- review extraction warnings and page/paragraph locators
- export a reviewer PDF, ZIP backup, or JSON backup
This hosted workflow is separate from the local Python review command. Extraction does not authenticate documents, and complete coverage still requires human review.
See the workspace guide and the service-engagement pilot.
Repository layout
src/ — core library and claim/evidence checksexamples/ — runnable example packets and integrationsdocs/ — references, quickstarts, deployment, and integration guidesapi/ — API-oriented surfaces and interfacesclients/ — client-specific adapters and examplescontracts/ — protocol and schema definitionsschemas/ — Schemas for requests, evidence packets, and responsesanalysis-bank/ — memory cards and retrieval guidancedeploy/ — deployment and infra helperstests/ and test/ — validation and regression checks
Status and examples
Documentation and development
Read:
Common verification command:
Run make verify-local when changing Worker, discovery, runtime, or validation-guard code.
License
This project is distributed under the MIT License.
The bundled official ACP schema retains its Apache-2.0 license and NOTICE.
Final note
Agenda Intelligence MD is best thought of as an evidence contract and preflight checker: a way to make AI-supported claims inspectable, before they become operational decisions.
If you want the fastest path in, start with the local Python install and the first-run example above.
