The pipeline
How a reading is built
Every reading on Truely follows the same path, from raw public records to a single plain-English line with a confidence attached. Nothing is generated from thin air — each step is deterministic and traceable back to a named source.
Official records
We start only with published UK government and regulator datasets — HM Land Registry, MHCLG, Ofsted, police.uk, the Environment Agency, DEFRA, Ofcom, Ordnance Survey and two dozen more — under the Open Government Licence. No scraped opinions, no proprietary black boxes.
Normalisation
Each signal is placed on a common scale — most as a percentile against every other covered postcode in England — so "£1,028 council tax" becomes "cheaper than 99% of the country". That's what makes one area comparable to another.
Cross-signal comparison
Signals are read together, not in isolation. Low crime beside improving schools beside rising broadband reads differently from low crime beside falling school rolls. The relationships between signals are where the interpretation lives.
Temporal analysis
Where we hold history — deprivation back to 2015, Band D council tax to 1993, crime over a rolling 24 months — we read the direction of travel, not just the snapshot. A place improving for eight years reads differently from one that just dipped.
Weighing & interpretation
Signals are weighed by relevance to the question being asked — a family lens leans on safety and schools, an investor lens on fundamentals and affordability. The weighting is documented, never hidden, and the same inputs are shown alongside the conclusion.
Confidence scoring
The final reading carries a confidence level set by how many signals were present, how recent they are, and whether they agree. A reading built on four corroborating, current datasets is labelled differently from one resting on a single older figure.
Provenance
Where every number comes from
Not all data is equal, and we don't pretend otherwise. Every figure on Truely falls into one of four tiers — and we tell you which, so an observed fact is never confused with a modelled estimate.
Observed
Published directly by an official source. The strongest tier — we are simply reporting the record.
Derived
Computed deterministically from observed data — a ranking, a percentile, a distance. Reproducible from the inputs.
Estimated
An informed approximation where no single official figure exists. Always shown with the evidence behind it and a confidence label.
Interpreted
A plain-English reading composed from the tiers above using documented rules. It explains; it never overrides the underlying data.
Confidence
What "high", "medium" and "low" actually mean
Confidence isn't decoration. It's set by a simple, consistent rule: how complete the evidence is, how recent it is, and whether the signals agree. Here's exactly what each level means.
A worked example
Reasoning, with the contradictions left in
Real places send mixed signals. A reading that smooths every contradiction into tidy certainty isn't intelligence — it's marketing. Here's how Truely holds supporting and conflicting evidence together, and lands on an honest read.
"Is this area getting better or worse?"
An inner-urban postcode, illustrative of the pattern.
Supporting signals
Conflicting signals
"Improving, with pressure. Schools, connectivity and serious-crime trends point to a strengthening area — but rising overall crime and worsening affordability suggest transitional growth rather than a settled, low-pressure neighbourhood. Weigh it as a place on the way up, not one that's already arrived."
Auditability
Every score, traceable to its inputs
There is no black box to take on faith. Each lens score is a transparent weighted composite, and every score carries its evidence: each input, its weight, its value, and how it reads against England. You can see exactly why a place scores the way it does — and so can an auditor, a lender's risk team, or a regulator.
GET /api/v1/intelligence/{pc} returns each lens's score, band, confidence and the full evidence array — every weighted input and how it reads vs England. Integrators can audit a score programmatically, not just trust a number.
The limits
What Truely does not do
An intelligence layer earns trust as much by what it refuses to claim as by what it shows. These lines are deliberate.
The full source list and refresh cadence live on the methodology page; live dataset status is on the status page. When the reasoning changes materially, we record it in the changelog.