The Sky Score API

Note: the DEFRA Lden raster tier has been live since 6 August 2026. It was bypassed between 3 and 6 August 2026 for the coverage reason below; that bypass was lifted once the consumer site was given the same measurements, so the two surfaces cannot disagree. Four things to know, stated plainly because some were understood only after the first was published:

• Coverage. DEFRA's aircraft contours reach only 10.4% of live London postcodes, and treating the unmapped remainder as quiet flattened the component.
• The dB-to-score mapping. Re-derived on 4 August after measuring the raster directly: it had been reusing a band table built for DEFRA's published reporting bands, which start at 55 dB, against raster values that start at 40 dB. That put 80.4% of every measurement we hold on a single perfect score. Fixed in code, staged behind the bypass.
• Vintage. DEFRA Round 4 was published in 2022 but maps 2021, which Round 4 documentation describes as "a highly anomalous situation" influenced by COVID travel restrictions. Traffic-suppressed contours understate a year of ordinary traffic, unevenly by airport. Measured at Heathrow, where the CAA publishes each year's annual contours: DEFRA's 2021 55 dB L_den contour covers 75.6 km², the CAA's model of 2025 (annual flights 2% below 2019's; ERCD Report 2601, Table 12) covers 142.4 km², nearly twice the area. They are different models, so that is an order of magnitude, not a correction, and we apply none. Re-anchoring needs Round 5, expected around 2027.
• Outdoor levels. L_den is assessed outdoors, at the most exposed facade. Indoor exposure depends on glazing and ventilation, which no public dataset records per property.

The geometry tier is documented in full in METHODOLOGY section 4.5. It is not a DEFRA reading and does not inherit the vintage caveat directly, but it is not independent of it either: since v3.8 each airport's distance scale is sized by that airport's measured Round 4 footprint.

How wrong is the estimate? Measured, not asserted. Where DEFRA published a contour we hold both numbers for the same postcode, so the estimator can be scored against DEFRA's value rather than argued about. DEFRA's maps are themselves modelled, not measured, but they are the official yardstick. Across all 35,441 London postcodes DEFRA's maps cover (methodology v5.5):

• Mean absolute error 1.03 points on the 0-10 scale; median 0.9, 90th percentile 2.2. It was 1.879 before v5.3, which derived the flight-path corridors from the UK AIP and fitted the corridor penalty, and 1.32 before v5.5, which made the airport term follow the runway rather than a circle. Both were fitted on half of these postcodes; on the held-out half the error is 1.09.
• 57.2% land within 1.0 point of the DEFRA value, 88.4% within 2.0.
• Signed bias −0.29: the estimate reads slightly LOUDER than DEFRA's maps, so where it is wrong it leans pessimistic about noise rather than optimistic. That is by rule since v5.5: the fit is not allowed to make any city read quieter than DEFRA, and in every city we cover it reads louder. For a noise product that is the safer direction to err, and it is stated here rather than left for you to discover.

Two things that figure is not. It describes postcodes near airports, because that is the only population where DEFRA published anything to compare against, not a global accuracy figure. And the bias is not purely estimator error: Round 4 maps a COVID-suppressed year, so some share of that −0.29 is real traffic reduction in the measurement rather than overstatement in the estimate. We do not apply a correction for it, because the split is not measurable from the data we hold. The same caveat reaches the fit itself: the corridor penalty's weight and the runway shape were fitted against this 2021 map, so they inherit the understatement, and we treat them as provisional until a normal-year yardstick is available.

Reproduce it yourself against our published dataset: python scripts/check_quiet_estimate_error.py.