Skip to content

Releases: philyarrow/local-digital-visibility-index

Release list

Q3-2026 — 271 businesses, 11 South West indices

Q3-2026 — 271 businesses, 11 South West indices Latest
Latest

Choose a tag to compare

View all tags
@philyarrow philyarrow released this 30 Aug 14:03
q3-2026
3e20a87

The first full quarterly snapshot: 271 businesses across 11 indices in Bath, Bristol, Cheltenham, Exeter, Gloucester, Swindon and Worcester, each scored 0–100 on a Digital Visibility Score built from six weighted pillars.

The published analysis lives at hub.pyc.agency/indices. This release exists so the figures can be checked rather than taken on trust.

Downloads

File What it is
ldvi-q3-2026-all-businesses.csv Every business, one row — score and all six pillar scores
ldvi-q3-2026.zip Per-index JSON and CSV, the methodology, the licence and the citation file

What the quarter found

  • 131 of 271 businesses rank for nothing at all in their own sector's twelve-keyword basket. On speed, technical foundation and content they are indistinguishable from everyone else.
  • Speed does not predict findability — it correlates −0.13 with being found, while carrying 20% of the scoring weight. Published as an audit of our own weights rather than quietly fixed.
  • Median homepage last modified 935 days ago; only 115 of 271 publish a blog, news or insights section.
  • Median Digital Visibility Score is 51. Nothing scores above 90; three businesses reach the 80s.

Corrections in this release

This snapshot includes three same-quarter corrections, all dated and explained in the changelog:

  • Review velocity was silently miscounted for the whole quarter — a timestamp parse returned NaN, so every business scored zero on it. Fixed and re-measured; 105 scores changed, all upward.
  • Review velocity is now scored within sector. A fixed threshold was largely measuring which trade a business was in, so builders were penalised for a trade whose customers do not leave Google reviews.
  • The Content & trust pillar was three booleans and claimed to measure two things it never did. Now six real signals; indexedPageCount was removed rather than faked.

Superseded datasets are kept alongside the corrected ones, so any figure cited before a correction remains verifiable.

Reproducing it

Every score is recomputable from this data using the pipeline in this repository. The scoring is plain arithmetic — ratios, weighted sums, a clamp — so no published figure requires trusting a model.

collect.mjs -> score.mjs -> backfill-enrichment.mjs --with-paid -> generate.mjs

Data is CC BY 4.0; the code is MIT.

Assets 4
Loading