Methodology
How evidence is made.
A Service Evidence record is not a testimonial and not a claim a business wrote about itself. It is a compiled reading of that business's own operational records, produced under stated rules, approved by the business, versioned, and published with its basis. This page explains what that involves and what it does not.
What the evidence is derived from
Service Evidence analyses a provider's completed-work records — the export of their own job or project system. Before any analysis, the export is read in the provider's browser and recognised customer details (names, addresses, email addresses, phone numbers, account details) are removed. The provider reviews the de-identified version before it is saved. The raw file is never uploaded.
Normalisation
Records are read under a service domain: a set of dimensions (the kind of work, the system or product, the problem, the client type, the place) with defined values and synonyms. A row is classified into those values; a value the domain does not recognise stays unrecognised rather than being guessed. This is what makes one provider's evidence comparable with another's, and what makes a count mean the same thing every time it is published.
Explicit rules, and exclusions that matter
- Only completed work counts. Cancelled, unknown or in-progress records are excluded and the exclusion is recorded.
- Every published fact has a minimum number of supporting records. Below the threshold it is not published.
- A fact resting on exactly the same records as a more specific one is not published twice.
- Stale evidence — work last observed too long ago — is excluded from publication.
- A more specific combination (a brand and a system and a problem) is published only when the records support it directly.
- Nothing is manufactured to fill a gap. A provider with little history gets little evidence.
Aggregation and de-identification
What is published is aggregate: counts, shares, periods and price ranges over sets of completed jobs. No customer is identifiable in a published record, no job row is published, and no note text leaves the provider's workspace. Internally, each published fact keeps a digest of the records behind it so that the basis can be audited without exposing them.
Historical pricing
Where a provider approves it and the records support it, a record may show historical price information: the median and middle half of recorded amounts for a defined kind of completed work over a stated period, with the sample size. This is history, as recorded. It is not a quote, an estimate for a particular job, an offer or a guarantee, and it is labelled that way wherever it appears.
Demand does not rewrite capability
Service Evidence also reads what customers are looking for — search and trend data — to decide which evidence matters commercially. That influences which facts are worth surfacing. It never changes what the operational history shows. Operational truth stays fixed; demand decides emphasis.
Approval, versions and updates
Nothing is published without the provider's recorded approval of the exact content: the scope, the categories, whether pricing is included, and a fingerprint of the content they saw. A publication is a numbered version. An update produces a new version with its own approval; earlier versions are kept, not overwritten. A provider can take a record down at any time.
For people and for machines
Records are written for a customer to read, and structured so that search engines and AI assistants can read them too: stable addresses, evidence present in the page itself, structured data that describes the business and its published evidence, and a visible link to the provider's official website. The aim is better information for a provider decision, whoever or whatever is making it.
What we do not claim
- Not an independent audit of every job. The analysis establishes what the records consistently show.
- Not a verification or certification of the business.
- Not a guarantee of correctness of the underlying export.
- Not a guaranteed ranking, recommendation or outcome in any search engine or AI assistant. We observe what those systems do; we do not control them.
Every published record states its own period, basis and version. If a record looks wrong, use the contact page and quote the record's address.