21 June 2026 · 6 min read

Valid averages and the 75% rule: when a reading counts, and when it doesn't

Enforcing the Clean Air Regulations 2014 — Part 2 of 5. A series for regulators and compliance staff, drawn from the DOE CEMS Guidelines (Version 8, 2025), which we helped develop.

In Part 1 we set out the two-part compliance test: no daily average above the ELV, and no half-hourly average above twice the ELV. But that test only means something if the averages going into it are valid. An exceedance, the guidelines are careful to say, is an exceedance “as indicated by valid measurement.” So before you can ask “did this average breach the limit?”, you have to ask a prior question: does this average even count?

That question is answered from the bottom up.

Averages are built, not measured

A CEMS doesn’t record a “half-hour” directly. It records fast, then rolls the data up through three tiers — and each tier has its own validity gate:

How a valid average is built · minute → half-hour → day

1-minute averagevalid data for ≥ 45 sec

Half-hour average≥ 22 of 30 valid minutes

Daily averagefrom valid half-hours, over operating time

Compliance testvs ELV / 2× ELV

● A tier only counts if the tier below it cleared its 75% gate

Validity flows upward. A half-hour built on too few valid minutes is itself invalid — and never reaches the compliance test.

Tier 1 — the valid one-minute average

The smallest building block of a half-hour is the one-minute average. Under §2.3.2 of the guidelines, a one-minute average is valid only if it contains valid data readings representing any 45 seconds of the preceding minute.

Forty-five seconds out of sixty is 75%. If the analyser was warming up, mid-calibration, or simply not returning good data for more than a quarter of that minute, the minute doesn’t qualify — and it can’t be used as a building block above it.

Tier 2 — the valid half-hour, and the 22-of-30 rule

Half-hourly averages are the unit the law actually judges (alongside the daily figure), so this is the tier that matters most for enforcement. The sub-average period for a half-hour is the one-minute average, and under §2.3.1:

A valid half-hour average must contain at least 22 sub-average data within the half-hour period (75%).

Thirty minutes, of which at least 22 must be valid one-minute averages. Twenty-two of thirty is, again, 75%. Clear the bar and the half-hour stands as a number the regulator can act on. Fall short — 21 valid minutes or fewer — and the half-hour is invalid: it isn’t a low reading, a zero, or a pass. It simply isn’t a measurement.

Here 24 of 30 minutes are valid — above the 22 threshold — so the half-hour is valid. At 21 or fewer, the half-hour would be discarded as downtime, not scored as a reading.

Tier 3 — the daily average

The daily figure is derived, not measured. Under Regulation 17(4), for each calendar day the owner or occupier calculates the daily mean value relating to the daily operating time, from the half-hourly mean values. In other words, the day is the average of the valid half-hours across the hours the plant actually ran.

This is where validity does its quiet, decisive work. Invalid half-hours are excluded from the calculation as downtime — the guidelines define downtime as periods of operation yielding invalid or no data (calibration, maintenance, malfunction, audits, out-of-control periods). They are not folded in as zeros, which would artificially drag a daily average down and hide a genuinely high day. The daily mean is built only from the half-hours that actually count.

A quick worked example

Why does excluding invalid half-hours matter so much more than it sounds? Take a facility with a daily ELV of 150 mg/m³ that runs a full day — 48 half-hours. Suppose a calibration and a short fault leave 10 half-hours invalid, and the 38 valid half-hours average 140 mg/m³.

Done correctly, the daily mean is built from the valid half-hours only: 140 mg/m³ — a day running uncomfortably close to its limit, exactly as it should appear.

Now imagine those 10 invalid half-hours were wrongly counted as zeros instead of excluded. The “daily average” becomes (38 × 140 + 10 × 0) ÷ 48 ≈ 111 mg/m³ — a comfortable-looking day that never happened. Zero-filling doesn’t just blur the picture; it actively hides a facility operating near its ELV. That single modelling choice is the difference between catching a problem and missing it, which is why the guidelines treat invalid periods as downtime to set aside, not data to average in.

Why this matters for enforcement

Three practical consequences follow, and they’re worth holding onto when reviewing a year of data:

A breach must rest on a valid average. A spike inside an invalid half-hour is not, by itself, a reportable exceedance — though it is a strong signal that something needs explaining.
Invalid data is a story, not a gap to ignore. Persistent downtime can mean a CEMS that isn’t being maintained, isn’t holding calibration, or is being run out of its valid range. Low data availability is itself a compliance concern, even when the valid readings that remain look clean.
The 75% gates are deliberate. They tolerate brief, normal interruptions while refusing to let a facility certify a half-hour — or a day — on too little real data.

That is the foundation the rest of enforcement stands on: a clear, mechanical rule for which numbers are real. With validity settled, the next question is what an operator must do the moment a valid exceedance — or a CEMS failure — occurs.

Assessing CEMS data and want to read validity the way the guidelines intend? Talk to us — we work both sides of this, helping operators and regulators interpret the same DOE CEMS Guidelines we helped write.

This article is general guidance, not legal advice. For the precise legal position, refer to the Environmental Quality (Clean Air) Regulations 2014, the EQA 1974, and the DOE CEMS Guidelines.

← All insights