MARCUS Documentation
Product documentation for using MARCUS in clinical and technical workflows.
MARCUS is a clinical retrieval and synthesis workspace that turns your institution's documents into citation-grounded answers, source briefings, and reusable project knowledge.
This documentation is written for people who are expert in medicine, operations, education, or policy, but not necessarily expert in software. You should be able to read these pages in the same way you would read a well-organized protocol manual: start with the practical workflow, then move into deeper reference and system behavior as needed.
What MARCUS Actually Does
MARCUS helps teams work with their own evidence corpus. In practical terms, that means:
- You create a project for one clinical service, policy set, pathway, or educational domain.
- You upload sources such as protocols, guidelines, pathways, policies, or reference documents.
- MARCUS breaks those files into searchable passages, analyzes them, and prepares supporting metadata.
- When you ask a question, MARCUS retrieves the most relevant passages, writes a response grounded in those passages, and shows citations back to the source material.
- The same corpus also powers briefings, authority signals, library views, knowledge surfaces, and comparison tools.
MARCUS is not designed to replace clinical judgment, local governance, or source review. Its job is to shorten the path from "I know the answer exists somewhere in our materials" to "I can find the relevant passage, understand the context, and verify it quickly."
Start Here
Getting Started
Build a clean first project, upload trusted documents, and ask a question with citations.
Guides
Learn operational habits for running MARCUS well after the basics are familiar.
Reference
Look up definitions, page behavior, API surfaces, and configuration controls.
Concepts
Understand how retrieval, enrichment, grounding, and security work in plain language.
Common Tasks
Create a project
Choose a scope that keeps retrieval focused and clinically interpretable.
Upload a document
Prepare the right file, understand ingestion, and confirm when it is ready.
Ask a question
Read the answer, citations, coverage cues, and follow-up paths safely.
Use the Library
Search, filter, compare, and inspect evidence across the full document set.
Use knowledge surfaces
Work with graphs, knowledge pages, themes, and evidence matrices when chat is not enough.
Troubleshoot weak answers
Diagnose low coverage, mixed projects, missing citations, and source-quality issues.
Recommended Reading Paths
Different readers come to MARCUS with different goals. These paths help you avoid reading the entire manual at once.
| If you are... | Read this first | Then read | Why this order works |
|---|---|---|---|
| A resident or clinician using MARCUS for the first time | /docs/getting-started | /docs/guides/reading-briefings and /docs/guides/troubleshooting-answers | You will learn the basic workflow first, then how to judge whether the evidence is actually usable. |
| A project owner or service lead setting up a corpus | /docs/getting-started/first-project | /docs/guides/managing-projects, /docs/guides/authority-levels, and /docs/guides/using-library | Corpus quality and scope are the main drivers of answer quality. |
| An educator or program lead reviewing knowledge coverage | /docs/guides/using-knowledge-surfaces | /docs/concepts/retrieval-system and /docs/concepts/evidence-grounding | This path explains how synthesis views relate to the evidence underneath them. |
| A technical reviewer, informatics lead, or developer | /docs/reference/glossary | /docs/reference/pages, /docs/reference/api, and /docs/concepts/security-model | This path gives you shared vocabulary, then implementation-level reference. |
How To Use These Docs Well
Use the documentation in the same sequence you would use the product itself:
- Learn the workflow.
- Learn how to evaluate output quality.
- Learn the system model only after the day-to-day behavior makes sense.
If you are trying to answer "what button should I click next," start in Getting Started or Guides.
If you are trying to answer "what does this term mean" or "which page owns this behavior," go to Reference.
If you are trying to answer "why did the system behave this way" or "what is happening under the hood," go to Concepts.
What A Good MARCUS Workflow Looks Like
A healthy MARCUS workflow usually looks like this:
- Create a narrowly scoped project.
- Upload a small number of current, trustworthy documents first.
- Wait for them to finish indexing.
- Review the source list and briefings to confirm the corpus looks right.
- Ask a concrete question.
- Open the citations before acting on any high-stakes answer.
- Improve the project when the answer quality is weak instead of repeatedly asking the same vague question.
That last point matters. When MARCUS underperforms, the problem is often not "the AI is bad." More often the issue is one of the following:
- The project contains mixed topics that should be separated.
- The relevant source was never uploaded.
- The uploaded file is a poor-quality scan or an outdated draft.
- The question is too broad for the answer to be clinically useful.
- The relevant answer exists, but only in a low-authority or narrow source.
What You Will Find In Each Section
| Section | Best for | What it covers |
|---|---|---|
| Getting Started | New users | The first project, first upload, first question, and first interpretation habits |
| Guides | Frequent users and project owners | Day-to-day operations, project hygiene, source interpretation, sharing, analytics, and troubleshooting |
| Reference | Support, informatics, and developers | Definitions, route inventory, endpoint families, and environment/configuration reference |
| Concepts | Reviewers and advanced users | Retrieval logic, enrichment, evidence grounding, security, and how the pieces fit together |
If You Only Read One Safety Rule
Treat MARCUS as a fast evidence navigator, not a substitute decision-maker. For any clinically important action, open the cited passages and confirm that the answer matches the source text, the patient context, and your local policy.