What FluTrack will and will not publish

Last updated: October 2026

This page collects the rules FluTrack is written and built to. It is a consolidation, not a new policy: every rule below already applies somewhere in the site or its build. Where a rule is enforced mechanically, that is said plainly, so it can be checked.

What FluTrack publishes

FluTrack describes dated, reported public-domain CDC respiratory surveillance observations. Where usable inputs exist, a state's combined respiratory index has a score from 0 to 100 and a level from a fixed five-point scale — Minimal, Low, Moderate, High, Very High, with boundaries at scores of 20, 40, 60 and 80. A direction is shown only with enough comparable history, and the separate influenza, COVID-19 and RSV readings carry their own availability and provenance.

The model supports four weighted inputs: wastewater viral activity, 0.30; the Acute Respiratory Illness activity level, 0.25; emergency-department visits, 0.25; and laboratory test positivity, 0.20. Positivity has no live adapter and is modeled only in illustrative samples. Missing inputs are excluded and the remaining weights renormalized; each state page lists the actual contributing signals rather than implying that all four were measured. The full method is at methodology; the datasets are at data sources.

What FluTrack refuses to publish

• No medical advice. FluTrack does not tell a reader what to do about a reading. It recommends and endorses no test, treatment, vaccine, medication, device or course of action. The full statement is at medical disclaimer.
• No individual risk claim. A reading is a state-wide, population-level indicator covering one reported week. It cannot account for any individual's history, exposures or circumstances.
• No prediction. The index summarizes reported observations, not a forecast or live case count. Each reading retains its observation period and provenance; a direction requires enough comparable history. Missing or insufficient history is unknown, not holding steady.
• No implied CDC affiliation. FluTrack is an independent project, not affiliated with, endorsed by, or sponsored by the Centers for Disease Control and Prevention or any government agency.
• No unlabelled figures. Reported values and illustrative sample inputs are distinguished by their provenance. A build may ship either a live or sample snapshot, and that kind travels with the data. Sample explanations describe an illustration, not actual state conditions. Explicitly unknown provenance stays unverified and cannot acquire a severity colour or borrow a surrounding live label. Missing state or pathogen observations remain missing rather than borrowing another geography's reading.

Where the numbers come from

FluTrack ingests only public-domain U.S. Government data. Its dataset registry names each CDC resource it queries — NSSP emergency-department visits, the NSSP Acute Respiratory Illness activity level, and the NWSS Wastewater Viral Activity Level — and records each licence as public domain.

Three of the four signals have implemented live adapters. Laboratory test positivity carries a model weight of 0.20, but no live adapter exists: it is modeled in the sample snapshot and excluded from live readings, with its weight renormalised away. Each state page lists the actual contributors and unavailable inputs.

A snapshot file ships with the site so a page is legible immediately. Its declared kind identifies live or sample inputs, and the display retains per-reading availability and provenance. A live replacement requires usable observations for at least 25 of the 51 jurisdictions at the stated observation period, not merely a successful response, a fresh retrieval timestamp or 51 jurisdiction keys. That coverage floor exists because an empty response once let sample data be badged as live — an error recorded in the changelog. Explicitly unknown readings remain unverified even inside a live bundle.

Licensing: an editorial rule enforced in code

WastewaterSCAN / SCAN data is published under a CC BY-NC 4.0 non-commercial licence and is excluded from this project, which is designed for commercial use. The existing conservative provenance filter also drops Verily-marked rows and certain partner labels. A provider name or contract notice alone does not determine a dataset's license. CDC's September 28, 2026 Verily testing-contract award is a distinct source context, not evidence that all Verily-marked data carries the WastewaterSCAN license. The filter and its tests remain unchanged; coverage reflects the retained rows. See data sources for the dated notice and reporting gaps.

The thresholds are FluTrack's own, not the CDC's

The breakpoints separating Minimal from Low from Moderate, and so on, are FluTrack's editorial choices, informed by typical seasonal ranges in the underlying CDC products. They are not CDC-defined cut points, and no government agency sets, reviews or endorses them. Different thresholds would give different levels from the same data. Every one of those numbers is published in full at methodology, which is what makes the choice defensible: it can be disagreed with specifically.

The trend rule is editorial in the same way. FluTrack compares the latest observation with the mean of up to three prior observations of the same measurement; a supported change of 8 percent or more either way is labelled rising or falling, and a supported change between is holding steady. This is not an ordinary week-over-week comparison. The reported percentage is clamped at 200 percent, since a very small base can otherwise produce a misleading number. Fewer than two observations produces “Not enough data”; changed reporting coverage can also prevent a comparable trend. Unknown provenance cannot establish a level or direction.

Corrections

FluTrack separates two things both called corrections. Surveillance figures are routinely revised as later reports arrive; FluTrack's numbers move with them on the next weekly refresh. The other kind — a mistake in how FluTrack computed or described something, or a deliberate change to the method — is recorded at corrections and changelog.

That log is append-only: a wrong entry is fixed by adding a corrective entry beneath it, never by rewriting history. Each entry is dated, categorised, and flagged according to whether the correction affected a reading that had already been published. If a figure does not match its CDC source, the reporting route is on about and at the foot of the changelog.

Nothing here is medically reviewed

No clinician reviews FluTrack's content, and the site states this rather than omitting it. The site configuration carries an explicit flag recording that the content is not medically reviewed; the accountability block on about renders directly from that flag, and a test asserts it stays false. Implying a review that does not exist is the failure this arrangement prevents. The publisher — Oak and Main Developers LLC — and the responsible-editor role that maintains the index method are named on the same block.

How money is kept away from the data

FluTrack is free. It is designed to be funded by display advertising and disclosed affiliate links; no advertising is served on the site today, and no affiliate link currently ships. The rules below apply from the moment either does, and are set out at affiliate and advertising disclosure. A commercial link may only be rendered by one shared component, which emits the disclosure immediately before the link, in fixed wording the caller cannot supply, and with a sponsored and nofollow relationship on the anchor by default. A state's reading is never sold, sponsored or adjusted, and the threat level is not available as an advertising targeting parameter; that page also rules out targeting on precise location, alert status, symptom or condition, pregnancy, age, or inferred vulnerability. And every third party that touches visitor data is named by legal entity at vendors — with its purpose, lawful basis, data categories, retention and deletion route — rendered from a single processor register that the privacy policy reads from as well, so the two documents cannot describe a vendor differently.

Which of these rules the build enforces

Several rules above are release gates. The build's QA step blocks the deploy, and runs on the deploy workflow as well as on continuous integration.

Rule	How it is enforced
A sponsored link ships with its disclosure attached	The QA step fails on any anchor carrying a sponsored relationship without the disclosure text immediately before it
No page lists the CDC signals while omitting the Acute Respiratory Illness level	The QA step reads every page sentence by sentence and fails on a list naming three of the four. Most of the site once contradicted the methodology page
The privacy policy and the vendor register name the same processors	The QA step fails if either page omits a registered processor, because both tell the reader the two cannot disagree
The share image does not appear to encode per-state severity	The QA step decodes the image and fails if more than a negligible number of severity-ramp pixels appear in its map area, since no real per-state data exists at build time
Icons and the share card are not truncated or blank	The QA step decodes every generated image and fails if its artwork does not reach the edge of its canvas
Every indexable page has a title, description and canonical — and a noindex page declares none	The QA step crawls the output and fails on a missing tag, a canonical on a noindex page, a dead link, or a contact address at a domain that can never receive mail
The site claims no medical review it does not have	A test asserts the medically-reviewed flag is false
The per-state evidence block contains no advice	A test asserts it matches no advice pattern, populated and empty alike

None of this makes FluTrack correct. It makes a particular set of failures loud instead of silent, which is a narrower claim and a more honest one.