CleanZone Field Brief
How the CleanZone Score Is Built (And Why It Matters)
A habitability score is only useful if you can see how it was made. We publish the full method: many metrics drawn from public datasets, weighted into composites, normalised against population cells, and thresholded against published standards. Nothing is hidden.
At a glance
- Raw metrics arrive in incompatible units — micrograms, decibels, becquerels, megabits — and cannot be added together until they're normalised onto one common scale.
- Normalisation (min–max or z-score against a chosen reference range) turns "8 µg/m³" and "42 dB" into comparable, unitless numbers, typically 0–100.
- Weighting is a value judgment, not a measurement. There is no objectively correct weight for noise versus air quality — only your own priorities.
- A single composite number can hide wildly different underlying realities. Two cells can both score 70 for opposite reasons.
- The point of exposing components — not just a headline score — is so you can re-weight the maths to match your own life, not ours.
Picture four readings sitting side by side: 8.4 micrograms of fine particulate per cubic metre, 48 decibels of night noise, 210 becquerels of radon per cubic metre, and 62 megabits of broadband. Which cell is "better"? You cannot answer that by adding the numbers — a microgram is not a decibel, and a becquerel is not a megabit. Any tool that hands you a single combined score has already made three decisions on your behalf: how to put those units on the same footing, how much each one should count, and how to combine them. This brief walks through that pipeline — normalise, weight, aggregate — so the decisions are visible rather than buried inside a number.
Step one: you can't sum incompatible units
Composite scoring exists because environmental data is stubbornly heterogeneous. Fine particulate matter is measured in micrograms per cubic metre. Night noise is measured in decibels — a logarithmic scale, not a linear one. Radon is measured in becquerels per cubic metre (or picocuries per litre in the US). Broadband speed is measured in megabits per second. None of these units share a common denominator, and a "48" in decibels means something completely different from a "48" in micrograms.
The fix is normalisation: converting each raw metric onto a shared, unitless scale — commonly 0–100 — before anything gets combined. The two standard approaches are min–max scaling (map the metric's plausible range onto 0–100, anchored to a reference range such as a published guideline) and z-scoring (express a value as standard deviations from a reference mean, then rescale). Both require picking a reference range, and that choice matters: anchor a particulate scale between the WHO annual guideline of 5 µg/m³ and the WHO interim target-1 of 35 µg/m³ (WHO Global Air Quality Guidelines, 2021), and a mid-range reading lands roughly mid-scale. Anchor it differently and the same raw value produces a different normalised score. Normalisation doesn't invent objectivity — it makes the comparison possible, and the anchor points used to do it are worth knowing.
Direction matters too. For pollutant-style metrics, a lower raw value is better, so the normalised scale runs in reverse — the lowest micrograms produce the highest score. For a metric like broadband speed, a higher raw value is better, so the mapping runs the other way. A normalisation step that doesn't flip direction correctly will quietly reward the wrong end of the scale.
Normalisation ranges are themselves a modelling choice, not a neutral fact. Anchoring a scale to the WHO guideline value versus a national regulatory limit will produce different normalised scores from the same raw reading — both defensible, both worth knowing which one you're looking at. A transparent tool tells you the anchor; an opaque one just gives you the number.
Step two: weighting reflects values, not truth
Once every metric lives on the same 0–100 scale, they still need combining — and that requires deciding how much each one matters. This is the step most single-number tools skip past silently, because it's the step with no objectively correct answer. A retiree who wants uninterrupted sleep might weight night noise at three times the importance of broadband speed. A remote-working family might do the reverse. Neither is wrong; they're weighting for different lives.
This is precisely why exposing the components matters more than polishing the final number. A score that lets you see "noise: 52, air: 61, water: 77, infrastructure: 70" — and lets you change how those four combine — respects that the weighting is yours to set. A score that only shows "65" has made that decision for you, invisibly, using someone else's priorities.
The same composite number can describe opposite realities. A cell scoring 65 because every component is a middling 65 is a genuinely different place from a cell scoring 65 because air and water are excellent while noise is poor. Averaging erases that distinction unless the components stay visible alongside the total.
Step three: aggregating without hiding the parts
The most common aggregation is a weighted arithmetic mean — multiply each normalised component by its weight and sum. It's simple and easy to audit, but it isn't the only option, and its behaviour is worth understanding: a weighted mean lets a very strong score in one component compensate for a very weak score in another. A geometric mean or a "weakest-link" minimum penalises a single bad component much harder, which suits someone who has a hard dealbreaker (say, radon) rather than a set of soft preferences. Which aggregation method is "right" depends on whether you want trade-offs to be possible at all.
None of that is a reason to distrust composite scores — it's a reason to want the components alongside them. A grid that exposes its underlying fields — for example pm25, noise, radon, flights, hardness, and g5_count, each fed by named public sources — lets you inspect what went into a number and re-combine it your own way, rather than trusting a single figure to already reflect your priorities.
Per-cell resolution changes the answer
Grid resolution is its own quiet decision. A very fine cell can be sparse — few nearby monitoring points to draw on, so the normalised value gets noisy. A very coarse cell smooths a city centre and its quiet rural fringe into one average, hiding the local variation that usually matters most to someone choosing a street rather than a country. There is no resolution that is correct in the abstract; there is only a resolution that matches the granularity of the underlying public data and the granularity of the decision being made. A per-cell score is a statement about the size of the area being summarised as much as it is a statement about the area itself.
Single opaque number
One figure, no visible anchor points, no visible weights. Fast to glance at, impossible to audit, and it silently encodes someone else's priorities as if they were yours.
Transparent components
Named fields, public reference points, and weights you can change. Slower to read at a glance, but you can see exactly why the number is what it is — and adjust it.
A composite score is a compression algorithm for your priorities — and every compression algorithm throws something away.
Public reference points used in normalisation
None of the figures above are CleanZone measurements; they illustrate the mechanism. The reference points that a normalisation step might realistically anchor to, however, are real and published:
None of these figures tell you what to weight most. WHO, the EU, and the EPA publish thresholds for identifying a problem in their own domain — not a ranking of which domain matters more to a given person's life. That ranking is the one step in the pipeline that only you can supply.
What a score can and can't do for you
A well-built composite score is a compression of a lot of dissimilar information into something you can scan quickly. That's genuinely useful — nobody wants to cross-reference six regulatory PDFs before choosing a street. But compression means information loss by design, and the two places it happens are exactly the two steps described above: the normalisation anchors, and the weights. See both, and a score becomes a starting point you can adjust. Hide both, and a score becomes a black box wearing the authority of "data-driven."
No single score replaces your own weighting. The point of laying out normalisation, weighting, and aggregation separately is so a number never has to be taken on faith — you can see the components, decide whether the weighting matches how you actually live, and recombine it if it doesn't.