Interpretation
One status per episode: what happened, or that it cannot be said.
TEI · Trustworthy Environmental Intelligence
Environmental data, interpreted only as far as the evidence goes.
TEI turns certified PM2.5 monitoring data into interpretations you can audit. When the evidence is too thin, it says so and records why.
The problem
Measurement is solved. Interpretation is not.
Monitoring stations measure particulate matter with high precision. The hard part is saying what caused an episode, how confident that conclusion is, and when the data cannot support one.
Dashboards report, forecasts predict and experts review. None produces an interpretation that can be audited, reproduced and challenged on its own terms. TEI is built for that gap.
The method
One rule: agree, or abstain.
The Environmental Concordance Framework (ECF) is a deterministic engine. For each episode it checks four independent signals: the PM2.5 level, consistent weather, persistence of three hours or more, and the pattern across monitoring sites.
When that bar is not met, TEI abstains and records the reason. Abstaining is not a failure: it is more defensible than a confident guess.
Reliability score for an episode E
R(E) = C(E) × D(N) × (1 − L(E))
- C(E)
- Signal concordance: the share of the four evidence signals that agree
- D(N)
- Network density, 1 − 1/N: with two monitors, reliability is capped at Moderate
- L(E)
- Known limitations: registered failure modes that apply to the episode
Scores run from 0 to 1, and R(E) can never exceed C(E). Below a set threshold, TEI abstains.
First results · Delaware County, Pennsylvania · 2022 to 2025
116 episodes. Each gets exactly one status.
Regional transport
Both monitors rose together: a regional air mass, with no source named.
Inter-site divergence
One monitor rose more than the other. Two monitors cannot confirm a local source, so reliability stays Moderate.
Abstained
Evidence too weak or incomplete: no interpretation, and the reason is recorded.
What every run produces
Four outputs, each traceable to its source.
Reliability score
A number from 0 to 1 saying how strongly the evidence supports it.
Evidence chain
The exact signals, sites and weather behind each conclusion.
Event registry
A permanent record of every episode, 34 fields each, reproducible from source data.
The evidence it uses
Certified data in. No modeled guesses.
EPA AQS PM2.5
Hourly Federal Reference Method readings, quality-assured and certified by the EPA.
ERA5 reanalysis
Hourly wind, temperature, humidity, pressure and rainfall at Philadelphia airport, used as physical context.
EPA AirNow
Near-real-time readings for current conditions. Never used for interpretation.
A research initiative
Validate it, extend it, build on it.
Researchers, public agencies, universities and health organizations can reproduce the method, test it on their own regions and improve it with us.
TEI interpretations are research outputs. They are not regulatory determinations, health advisories or legal findings.