AI-enabled decision intelligence

Atlas Lyme is an Epidemiologist Decision Intelligence Platform. The product direction is AI-enabled: it is intended to help people organize, compare, synthesize, and explain governed One Health evidence so they can decide what deserves investigation.

AI-enabled does not mean that Atlas replaces epidemiologists or turns a score into a prediction. The preferred public-health framing is:

signals → evidence → hypotheses → anomaly candidates → investigation priorities

Each step must remain interpretable, bounded by the evidence, clear about uncertainty, and accountable to human review.

Capability status

Capability	Status	What you can rely on
Transparent follow-up-priority scoring	Available now	The current score is a deterministic, adjustable calculation over released inputs. It helps focus review; it is not a predictive model, diagnosis, causal claim, disease-burden estimate, or individual-risk output.
Public evidence and provenance context	Available now	The Explorer provides released signals, source/release/method context, limitations, and a table path for reviewing results. Start with evidence, provenance, and uncertainty.
Atlas Assistant: reviewed literature	Early access, feature-gated	/assistant can ask the Atlas API about reviewed PubMed and PMC Open Access literature when enabled. Answers include cited evidence and explicit unavailable-evidence outcomes. This does not query live Atlas geography or scores, provide medical advice, or replace human review.
Ask Atlas in the professional workspace	Experimental/demo	Atlas Explorer and Compare open the dedicated Ask Atlas page. When literature chat is enabled, it uses that same reviewed-literature service and leaves navigation with you. When it is disabled, the Ask Atlas page says Ask Atlas is unavailable and the other workspace pages stay usable. The request is a message plus optional conversation history. There is no Structured mode and no Both mode.
Ask Atlas page at /askatlas	Experimental/demo	The dedicated page uses the reviewed-literature service. Saved chats stay in this browser. It does not offer Structured or Both mode, and it does not store chats on Atlas servers.
Broader Atlas evidence assistant	Planned	Mixed Atlas data and knowledge-graph questions need separate governed work and release review. This is not a released capability.
Anomaly candidates and interpretable modeling	Planned	Future outputs may help identify unusual patterns or explain modeled context, but no current public result should be described that way without a released model, inputs, validation, uncertainty, and review record.
Literature monitoring or agent assistance	Planned	Future monitoring or agent workflows must use approved sources and explicit release controls. Released MCP evidence tools are anonymous reads of governed public evidence. They do not run an agent. The September 2026 smoke-only development receipt is historical. See API, MCP, and access.

The status labels are intentionally explicit. A planned architecture, a design exploration, or a local demo is not evidence that a capability is available in the public release.

What the current release does

Today, Atlas helps you:

• explore public county and state patterns;
• review human disease, tick/vector, pathogen, environmental, population, and related context when released;
• inspect the source, vintage, geography, release, method, completeness, and limitations behind a result;
• compare supported geographies descriptively; and
• turn a review-priority signal or evidence gap into a bounded investigation question.

See current capabilities and investigation workflows for the released behavior and task steps.

What human review means

Use an AI-assisted or analytical output as a prompt for review. A responsible reviewer should:

1. check the source and provenance;
2. verify that the geography, period, coverage, and method fit the question;
3. distinguish observed evidence from a derived score, contextual evidence, or modeled output;
4. state missing evidence and uncertainty rather than filling gaps with assumptions; and
5. decide whether source validation, local surveillance review, additional analysis, expert consultation, or no action is appropriate.

If an output cannot show the evidence and limits needed for that review, treat it as unavailable. Do not infer certainty from fluent language, a rank, a map color, or an AI-generated summary.

What Atlas AI is not for

Atlas does not:

• diagnose Lyme disease or provide clinical or treatment advice;
• predict an individual's exposure, infection, or disease risk;
• guarantee that an outbreak will or will not occur;
• prove that a vector, pathogen, environment, or population factor caused an outcome;
• convert missing surveillance into evidence of absence or zero; or
• make an automatic public-health decision or replace epidemiologists, analysts, local knowledge, or responsible authorities.

The AI Ethics commitments remain the public page for AI status, boundaries, and open decisions. Signed-in readers can open AI / Responsible Use from the professional workspace footer for the same topic inside the workspace. That page uses capability-maturity labels and links here for methods. Privacy and data-use commitments belong on the Privacy page; this guide does not add to either policy source.