ABOUT THIS PROJECT
UK Broadband Data Explorer methodology
How we combine measured broadband performance, reported network coverage, provider evidence and geographic research, including the quality controls, calculations and limitations behind the UK Broadband Data Explorer.
What the Explorer is designed to show
The fibredog.com UK Broadband Data Explorer brings several different kinds of broadband evidence into one consistent view. It is designed to show what qualifying real-world tests measured, what network coverage was reported, which providers were credibly observed and how the evidence changed across equivalent annual periods.
Download, upload, latency and distribution evidence from qualifying tests.
Ofcom records describing reported access to defined service levels.
Credible provider observations assessed in their geographic context.
Like-defined evidence compared across five annual periods.
No single source answers all four questions. The Explorer keeps measurement, availability, provider and geographic evidence distinct, then applies a registered method appropriate to each measure. A speed test is not treated as proof of availability, and reported coverage is not treated as proof of the speed experienced inside a home.
This methodology explains the evidence, analytical decisions and limitations needed to interpret the results. It does not disclose the exact thresholds, matching rules, calibration constants, abuse controls or watermark construction used to protect the service.
Evidence sources
Measured broadband performance
Measured-performance evidence is drawn from the M-Lab NDT Data Set. NDT records observed test performance rather than the advertised speed of a broadband package. Qualifying measurements support the Explorer’s download, upload, latency, distribution, time-of-day and provider-observation views.
M-Lab test data is available under the Creative Commons CC0 public-domain dedication. Source: The M-Lab NDT Data Set, 9 August 2021–8 August 2026, measurementlab.net/tests/ndt.
Reported fixed-broadband availability
Reported availability comes from Ofcom’s Connected Nations Spring 2026 fixed-broadband coverage data. These postcode-level records describe the proportion of premises reported as able to receive defined service levels. The latest Explorer period uses Ofcom’s January 2026 snapshot, including the revised postcode coverage release.
Equivalent Ofcom releases are aligned to the corresponding Explorer periods across the five-year series. Ofcom data is reused under the Open Government Licence. It remains Ofcom-reported availability and is not presented as Fibredog measurement.
Provider and geographic evidence
Provider observations are checked against a maintained evidence register of published geographic footprints. It prioritises first-party provider information, then sufficiently specific published evidence where an official footprint is unavailable. Footprints are corroborating evidence, not a universal address-level availability database.
Named-place searches are built from OS Open Names, supplemented by Fibredog’s reviewed mapping of place identities and aliases to outward postcode areas. OS Open Names is an Ordnance Survey open dataset used under the Open Government Licence.
Great Britain representative postcode points are derived from the May 2026 ONS Postcode Directory. Contains National Statistics data, Ordnance Survey data and Royal Mail data used under the Open Government Licence v3.0 and applicable third-party terms. Northern Ireland points use approximate GeoNames-derived open geographic data under CC BY 4.0.
Five equivalent evidence periods
The Explorer uses the same day and month boundaries for each annual measurement window. This prevents one displayed year from silently covering a longer or shorter interval than another. The interface shows the year in which the period ends.
Ofcom releases are snapshots rather than continuous feeds. Coverage evidence associated with an Explorer year represents the corresponding published network snapshot, not a daily estimate for every date in its measurement window.
From individual tests to area evidence
NDT is observational. People choose when to test, often because a connection feels slow or unreliable. Results can also be affected by Wi-Fi, home-network equipment, device capability, subscribed package, server path, congestion and temporary conditions.
Determining whether even one observation is suitable for analysis is not a simple matter of removing incomplete rows or values outside a fixed range. Each candidate record sits inside a much wider body of spatial, temporal, provider and sample evidence. Its usefulness depends not only on its own technical integrity, but on whether its claimed geography is plausible, whether the attributed provider can credibly be associated with that location, how the observation relates to neighbouring evidence and whether the surrounding pattern is sufficiently coherent to support an inference.
That assessment requires several analytical layers to agree. Provider identities must be normalised across changing labels and network relationships. Observations must be reconciled to canonical geography. Their spatial arrangement must be evaluated as a geometric structure rather than as isolated points, including whether an apparent cluster has credible form, density, continuity and relationship to the provider’s defensible network shape. Repetition, separation, local concentration, contradictory evidence and the strength of the available sample all alter the weight that can safely be placed on the result.
The process is deliberately conservative. A technically complete record can still be analytically weak; an unusual record can still be genuine; and a large collection can still be misleading if its geographic or behavioural structure is poor. No single test, provider footprint or sample threshold decides the outcome by itself. Qualification is the result of corroboration across interacting evidence layers and calibrated uncertainty.
Only after qualification are eligible records organised into consistent annual, geographic and provider evidence structures. The public data foundation contains precomputed area-period and area-provider-period summaries, allowing the same registered analytical definitions to be applied throughout the Explorer rather than reconstructed ad hoc in the visitor’s browser.
Original analytical summaries remain preserved. Visitor-facing calibration is a separate interpretive layer, so a public presentation adjustment does not overwrite its source aggregate.
The visible charts are the end of the process, not the process itself. Reaching them requires a maintained body of geographic, provider, temporal and distributional evidence; resolution of conflicting signals; metric-specific aggregation; and repeated validation of the resulting patterns. Applying a few universal cut-offs to the source data would produce a materially different and substantially less defensible result.
The Chaos Filter
The Chaos Filter is Fibredog’s default evidence-quality system. It exists because large observational datasets contain a mixture of strong local evidence, ambiguous attribution, thin samples, geographic anomalies and results shaped by the circumstances in which people choose to test.
It combines several evidence families, each of which informs the interpretation of the others:
- record validity and completeness;
- canonical geography and place resolution;
- observed-cluster geometry and spatial relationships;
- provider identity and published-footprint corroboration;
- repeated evidence across an area or period;
- sample sufficiency and uncertainty signals; and
- calibrated presentation of measured-speed distributions.
These are not a linear checklist and cannot be reduced to one universal confidence score. The meaning of a spatial cluster depends on the scale and topology of the selected geography, the density and separation of its observations, the provider’s plausible network morphology, evidence in adjoining areas and the behaviour of the sample through time. Contradictory signals are retained as uncertainty to be resolved rather than silently forced into agreement.
One unusual test is therefore not automatically wrong, one dense cluster is not automatically representative and one broad provider claim is not strong enough to validate every observation inside it. The filter evaluates a multidimensional evidence structure in which the significance of one signal can change when the surrounding geography, provider evidence or repeated observations change.
Switching the filter off exposes less-corrected evidence, including sparse samples, provider observations outside a defensible footprint and the known downward bias of problem-driven testing. It is useful for transparency, but it is not the recommended headline view.
The operational thresholds, mathematical constants, spatial tests, calibration functions and decision architecture remain private. They form part of Fibredog’s proprietary quality-control and abuse-resistance system; publishing them would turn an explanatory methodology into a reconstruction guide and make the controls easier to reproduce or evade.
Measured speeds and distributions
The Explorer uses percentile summaries because they describe the shape of measured performance more honestly than a single average.
Ten per cent of qualifying measured downloads fall below this threshold.
Half of qualifying measured downloads fall below this value and half above it.
Ten per cent of qualifying measured downloads exceed this threshold.
These are measured-test percentiles, not guaranteed package speeds, universal household minimums or the theoretical capability of every line in an area.
The public view applies a consistent calibrated interpretation to qualifying download percentiles. It is designed to reduce distortion associated with problem-driven testing and constrained test conditions while retaining differences between places, periods and providers. Calibration does not convert observational testing into a random household survey.
Upload and latency use their corresponding qualifying records. Latency is shown in milliseconds and lower is better. Time-of-day change compares like-defined periods within the evidence; it does not claim that every household experiences the same evening effect.
Reported availability versus measured reality
The coverage comparison deliberately places two different questions side by side:
- Reported availability: what proportion of premises Ofcom data reports as able to receive a stated service level?
- Measured attainment: what proportion of qualifying tests met the corresponding measured threshold?
For 30 Mbps, 100 Mbps and 300 Mbps comparisons, measured attainment is based on counted qualifying download tests. The gigabit comparison uses a clearly identified estimate of the share reaching a high measured threshold, derived from the stored distribution evidence for each postcode-period before aggregation.
Ofcom percentages are combined using the number of contributing full postcodes recorded for the relevant availability measure. Measured percentages retain their own qualifying-test denominators. The two evidence families are calculated separately before a displayed difference is taken.
The difference is predominantly explained by take-up and package choice. Many premises with access to faster services remain on slower packages. The measured result can also be influenced, to a lesser extent, by Wi-Fi, devices, in-home wiring, congestion, test timing and other conditions. It should not be interpreted as a count of network faults or failed lines.
Observed provider evidence
A provider appears when credible qualifying measurement evidence is attributed to it under the selected geography, period and filter state. This requires more than matching a returned label. Known brands, historical names, wholesale relationships and network identities must be normalised without collapsing genuinely different operators into one another.
Geographic plausibility is assessed against the maintained footprint register and the wider evidence around the observation. Published footprint evidence rarely behaves like a clean boundary: it may describe towns, partial deployments, corridors, build phases or broad service regions at different dates and levels of precision. The Explorer therefore considers whether the observed spatial pattern is compatible with a defensible network shape, rather than merely asking whether a place name occurs in a provider’s publicity.
Provider performance summaries weight area-period values by the relevant qualifying count. Download uses qualifying download counts, upload uses qualifying upload counts, and latency uses qualifying measurement counts.
Observed is not the same as available. Presence means qualifying evidence was observed. Absence may simply mean that no usable test entered the sample. This is not a substitute for an address-level availability checker.
Small-provider samples can still be informative, but the interface labels limited samples and suppresses evidence too sparse to support a useful default comparison. Footprints, identities and classifications may be revised when stronger evidence becomes available.
Postcode areas, named places and map points
The Explorer’s core public geography is the outward postcode area, meaning the first part of a UK postcode. A visitor can select one area, combine several areas or choose a reviewed named place that resolves to a documented set of outward postcode areas.
Place names are not simple synonyms for administrative boundaries. They can overlap, contain aliases and include postcode areas shared with neighbouring places. The resolver canonicalises relevant place records and removes duplicate postcode areas before a query runs.
The map uses one representative point for each outward postcode area. For Great Britain, that point is based on the coordinate-wise median of available unit-postcode centroids. Median coordinates reduce the influence of distant outliers but do not define a postal boundary or the geometric centre of a legal area.
Map circles describe evidence associated with the outward postcode area. Their location is for orientation and comparison; it does not imply that every test occurred at the marker or that the circle represents network coverage.
How multi-area results are calculated
Different measures require different aggregation rules. The Explorer registers the rule with the metric rather than applying one universal average to unlike quantities.
- Multi-area headline speed, percentile, upload, latency and time-of-day figures are arithmetic averages of the selected outward-postcode-area values in the unfiltered area view.
- Provider-filtered performance figures are weighted by the relevant qualifying-test counts.
- Ofcom availability figures are weighted by the contributing full-postcode counts supplied for that measure.
- Qualifying-test counts are summed where the display explicitly reports a count.
- Provider totals count distinct providers satisfying the current selection and evidence rules.
This preserves each metric’s intended meaning. It also means a multi-area headline is not necessarily the same as recalculating a percentile from every underlying test pooled into one national-scale sample.
Five-year comparisons
The time-machine view applies the same metric definition to each available annual period. Missing years remain visible rather than being joined across as though evidence existed, and years covering only part of the selected geography remain identifiable.
Area-level headline measures use the arithmetic average of available area values for that period. Availability retains its full-postcode weighting. With a provider filter active, performance uses the corresponding qualifying-test weights.
Absolute changes, percentage changes and percentage-point changes are different quantities and are labelled separately. A change between the first and latest available period describes the selected evidence; it does not establish its cause.
Measurement volume, represented providers, package take-up and testing behaviour can all change between years. Those changes remain part of the context in which a trend should be read.
Sample strength and uncertainty
The Explorer reports qualifying measurement counts where they materially affect interpretation and uses plain-language sample labels to distinguish stronger evidence from limited samples.
A high count improves the evidence base but does not eliminate selection bias. A small count does not make every observation invalid, but it reduces confidence that the summary represents the wider area or provider. Sample labels are interpretive aids, not formal margins of error.
The system also considers how evidence is distributed geographically and across repeated observations. A concentrated cluster can tell a different story from the same number spread across an entire selected area.
Figures are rounded for display. Calculations use stored unrounded values where available, so differences reconstructed from rounded labels may vary slightly from the displayed result.
PawPrint and evidence integrity
Explorer outputs use Fibredog PawPrint, a forensic data-watermarking system designed to help identify mass extraction, systematic scraping and unauthorised reuse.
PawPrint is designed not to change the substantive meaning or ordinary interpretation of legitimate published results. It is not intended to prevent quotation, academic discussion, journalism or normal use of screenshots and embeds permitted by the Explorer terms.
The watermark construction, issuance signals and detection rules are not public. Publishing them would weaken the integrity control they provide.
Limitations, corrections and updates
The Explorer is a carefully controlled synthesis of available evidence, not a census of every UK broadband line. In particular:
- M-Lab NDT measurements are observational and are not a random sample of premises.
- Wi-Fi, devices, home networks, package choice, congestion and test conditions affect measured performance.
- Calibration reduces known distortion but cannot remove every source of bias.
- Ofcom coverage describes reported availability at a source snapshot; it does not prove take-up or in-home performance.
- A provider observation is evidence of presence, not a complete availability statement.
- Public provider footprints vary in detail and can lag deployment, withdrawal or rebranding.
- Named places and outward postcode areas do not share exact boundaries.
- Representative map points are not boundary polygons or address locations.
- Multi-area summaries can conceal substantial variation within the selected geography.
- Modelled quantities depend on their stated assumptions and available distribution summaries.
- Historical comparisons can be affected by changing sample volumes, provider representation and source coverage.
Fibredog reviews source releases, provider evidence, place mappings and analytical behaviour as the Explorer develops. We may correct source handling, provider identities, mappings, classifications or presentation when stronger evidence becomes available.
Questions about a result or this methodology can be sent through the fibredog.com contact page with the selected area, period and relevant output identified.
Return to the Explorer and build a detailed picture of broadband in your selected area.