Evidence, provenance, and uncertainty

Atlas Lyme is designed to help epidemiologists identify areas where available tick/vector and related One Health evidence may warrant review. A signal is the beginning of an investigation question—not a finding that a place is definitively high risk.

What an Atlas output means

Use these terms consistently when reading a result:

Term	Meaning	How to use it
Observed signal	A value or status reported by an approved source, such as a published county-linked human-surveillance value, tick status, or pathogen detection status.	Describe what the source reported, including its geography and time or vintage.
Derived measure	A value calculated from released inputs, such as a county review score or evidence-completeness percentage.	Use it to focus review. It is not a diagnosis, a complete disease-burden estimate, a confidence interval, or individual risk.
Contextual evidence	Information that helps interpret a result, such as population, access, rurality, habitat, climate, or scientific context.	Treat context as relevant background with its own source, coverage, geography, vintage, and limitations.
Modeled output	An estimate, forecast, classification, or other result produced by a model rather than directly published as an observation.	Require the model version, inputs, validation, uncertainty, and human review before relying on it. The current public explorer does not promise an authoritative modeled outbreak or individual-risk output.
Investigation candidate	A geography or pattern selected for a bounded follow-up question because the available evidence or an evidence gap merits review.	State what should be checked next and what evidence could change the interpretation.

Atlas Explorer labels the county review score and evidence completeness as analytical signals. The active release is shown in the application; current capabilities describes the release boundary.

Start with provenance

Before interpreting a signal, locate and record:

• Source and publisher: who produced or published the input, and what link or citation identifies it;
• Release and vintage: the active Atlas release and the source observation, reporting, or publication period;
• Geography: county, state, sampling location, boundary definition, and any records not allocated to a county;
• Method and version: the methodology, calculation, or model version used to produce the displayed value;
• Freshness and processing context: source retrieval, release generation, and data load times when the application provides them; and
• Limitations: coverage, sample size, representativeness, missingness, suppression, comparability, uncertainty, and known data-quality constraints.

The evidence snapshot on Atlas Explorer summarizes source periods, scoring methodology, and release processing times in plain language, with exact IDs behind Inspect provenance. County evidence adds county-level source context when it is available. Use that page before sharing or acting on a result. A generated date describes the release; it is not necessarily the date of disease observations. See Releases and methodology labels for definitions of each label.

The list above is the provenance to record for evidence in the current open-data release. Atlas is also designed so a later governed source can be normalized into the same evidence concepts. That normalization keeps the original source and publisher, dataset version or vintage, geography, observation period, transformation, and limitations on the result. A shared model does not make two sources equivalent. Missing, suppressed, unavailable, and uncertain inputs stay visible as the states in the next section. Sources and canonical evidence separates that architectural intent from the sources released today.

Read value states carefully

These states answer different questions and must not be collapsed into one another:

State	What it can mean	What you should say
Observed or published	A source published a value or status for the stated geography and period.	“This source reports…” and include the source context.
Zero	The source explicitly reported zero for the stated measure and period.	“The published value is zero for this release.” A zero is still bounded by source coverage and method.
Missing or no record	No county-linked row or value is available in the release.	“No county record is available.” Do not call this absence.
Suppressed or privacy-protected	A publisher withheld or transformed a value to protect privacy or meet a reporting rule.	Describe the protection and do not reconstruct a precise count.
Unavailable or not reported	The source, field, geography, or period is not available for this release.	Treat it as an evidence gap, not as zero or no activity.
Stale	The source vintage or retrieval is too old for the question being asked.	Confirm the current source and state the time limitation.
Sparse or incomplete	Coverage, sampling, reporting, or scored inputs are limited.	State what the available records can and cannot support.
Disagreement	Two sources or signals point in different directions.	Preserve both observations, compare their provenance, and make the disagreement the investigation question.

Atlas Explorer uses “No records,” “Unavailable,” and completeness indicators in context. “No records” describes published evidence availability; it does not establish that ticks, pathogens, or disease are absent.

The signed-in workspace uses three labels after a successful response: Available, Limited, and Unavailable. Available includes a published zero. Limited means a signal exists and a documented constraint should change how you read it. Unavailable means the governed value is absent or out of scope, and the page shows that word rather than zero. A spinner or a failed request is a loading or error state. It is not the Unavailable evidence label. Inspect provenance on a workspace value opens source family, period, and technical identifiers without leaving the page. The deeper interpretation stays in this guide. See the professional workspace.

Low signal is not the same as insufficient surveillance

A low observed signal means that a source reported a low value or status under its method and coverage. Insufficient surveillance means the available evidence is too incomplete, unrepresentative, outdated, or incomparable to determine whether the underlying activity is low.

For example, a county with no county-linked published Lyme case record may have zero cases, suppressed records, unallocated records, a reporting boundary mismatch, or another publication limitation. Atlas does not treat that missing record as zero by default. Likewise, a tick table with no county record does not prove ticks are absent.

When evidence is incomplete, use the safer statement: “The current release does not establish this.” Then identify the source or coverage check needed to learn more.

Why signals may not agree

One Health inputs can differ in:

• collection design and sample size;
• geographic boundaries and whether records are allocated to counties;
• observation, reporting, and publication periods;
• publisher, laboratory, field, or case definitions;
• coverage and representativeness of the sampled population or habitat; and
• missingness, suppression, processing, and update schedules.

Those differences can produce apparent agreement or disagreement without proving that one source caused another or that either source represents every person, tick, or location. Compare provenance first; do not average away a disagreement or describe it as certainty.

Read the evidence before acting

Use this checklist for a county, comparison, or other review candidate:

1. Name the question. What decision or follow-up are you trying to inform?
2. Name the output. Is it an observed signal, derived measure, contextual input, modeled output, or investigation candidate?
3. Check the provenance. Source, publisher, release, vintage, geography, method/version, freshness, and processing context.
4. Check the value state. Observed, zero, missing, suppressed, unavailable, stale, sparse, or conflicting?
5. Check comparability. Do the geographies, periods, definitions, coverage, and methods support the comparison?
6. State what remains unknown. What cannot be concluded from this release?
7. Choose a bounded next step. Source validation, local surveillance review, a time-bounded analysis, expert consultation, or no action.
8. Record the handoff. Save the URL, release, method, filters, evidence context, question, reviewer, and caveats using the investigation workflows.

Atlas Explorer provides the map and the county list for the same release. If map geometry is unavailable or a map is not usable, use the county list and text summaries. For AI-enabled or experimental experiences, keep the AI Ethics commitments in view and retain human review.

Atlas helps identify what may deserve investigation. It does not provide clinical advice, person-level exposure or disease-risk predictions, causal proof, guaranteed outbreak forecasts, or an automatic public-health decision.