Index changelog — what changed, and when

A ranking that changes without saying why is not evidence. This page records every change to the Local Digital Visibility Index — new cohorts, methodology revisions, scoring corrections and coverage improvements — so that a firm whose score moves can tell whether the firm changed or the ruler did.

Each dated open-data snapshot is tagged in the pipeline repository, so any figure published here can be recomputed from the data as it stood that quarter.

2026

14 September 2026 (release) — double glazing, and the first index that is not a town

Two new cohorts, 215 businesses, taking the index to 34 cohorts and 1,371 businesses across eleven trades. No existing score moved.

	businesses	seed share of its own SERP	median
Bristol	116	61.7%	50
Gloucestershire	99	63.3%	50

Those are the two strongest results in the index after Swindon veterinary practices. We expected the opposite.

We predicted this sector would fail, and published the prediction’s failure. Double glazing is a construction-family trade, and construction holds only 8.3–10.8% of its own local search results in all three towns measured — Checkatrade, the Federation of Master Builders and TrustATrader hold the rest. The reasonable inference was that a double glazing table would measure directories. It does not. The firms hold three fifths of their own first page.

Gloucestershire is a county because the trade is

The first cohorts were seeded the usual way: Google Business Profiles within 10km of a town centre. That works for a vet, whose customers come to the premises. It fails for a glazier, who goes to the customer.

The measurement said so plainly. In the Cheltenham cohort, local firms outside it held 73 of 120 first-page slots against the cohort’s 45; in Gloucester, 69 against 40. Checking the firms by name found them 11 to 14km out, carrying exactly the categories we had configured. The radius was wrong, not the sector.

Widening each town separately would have made two overlapping circles 11.5km apart — the failure that killed an earlier locksmith index. So the two town cohorts were replaced by one county cohort at 24km.

The four firms the town radii missed now sit at ranks 1, 2, 9 and 12. The published Cheltenham and Gloucester tables would have omitted the two best-performing double glazing firms in the county while reporting that the trade scored 33%.

Searching from two miles away returns different firms

The county cohort measures county, town and neighbourhood keywords separately, because they are not the same question. Comparing the top ten:

	local firms hold	directories hold
Town query, e.g. “double glazing Bristol”	75%	10%
Neighbourhood query, e.g. “double glazing Bishopston”	33%	17%

Across the three original cohorts, town and neighbourhood searches shared only 11–36% of their top ten. “Double glazing Cheltenham” is won by local firms; “double glazing Charlton Kings” is won by a county-wide SEO microsite, TrustATrader, Facebook and Yell. A business can sit near the top of this index and be invisible to someone two miles away who typed their own suburb.

One known defect: “Bishopston” is both a Bristol suburb and a village near Swansea, and a Swansea glazing site ranks in that neighbourhood’s results. That keyword will be replaced next quarter. Fishponds, Charlton Kings, Hatherley, Quedgeley and Hucclecote are unambiguous.

Structurally absent, not badly scored

Three firms hold first-page positions in Gloucestershire — one of them position 1 for two separate queries — and have no Google Business Profile within 40km. Every index is seeded from Google Business Profiles, so those firms cannot appear in it at all. That is a limit of the method, not a judgement about them, and the methodology now says so.

Conflict of interest

Pane Relief, a Bristol glazing business, is a PYC client. It appears in the Bristol index scored identically to everyone else, at rank 26 of 116. The alternative — excluding it — would be a worse answer, because a league table that quietly omits the compiler’s clients is not a league table.

What else changed

Adding 215 businesses re-ran every computed figure on the site, and fourteen hand-written titles and descriptions that quoted the old corpus were corrected in the same commit. Among them: businesses ranking for nothing at all moved from 526 of 1,156 to 636 of 1,371; never named by an AI assistant from 845 to 1,016; sites failing every technical check from 85 to 99. One had been wrong since 1 September — the score definition page still described the distribution across 271 businesses.

The facet navigation now reads “By area” rather than “By town”, because a county is not a town.

14 September 2026 (release) — enrichment reaches the twenty-one new cohorts

No score moved. All 1,156 business scores were compared before and after regeneration and every one is identical. Enrichment is unscored by design, so this release changes what the site can say, not what it ranks.

The 1 September release noted that Lighthouse and backlink data covered only the original eleven indices, and that the lab-versus-field analysis therefore described 66 businesses rather than the whole corpus. The enrichment pass has since run. It finished on 2 September and had not reached the site until today, because score.mjs was never re-run after the backfill and the published pages read from its output.

Lighthouse now covers 1,055 of 1,156 businesses and backlinks all 1,156. Chrome UX Report data remains sparse, as it must: a site needs enough real traffic for Google to report a distribution at all, which is itself a finding about businesses this size.

Lab versus field	1 September	14 September
Businesses with both	66	121
Median lab LCP	11.9s	12.6s
Median field LCP	1.7s	1.7s
Lab-to-field ratio	7.0×	7.4×
The two verdicts disagree	21 (32%)	44 (36%)

The direction of the disagreement held, which is the part that matters: 24 sites pass the lab test while real visitors fail Core Web Vitals, against 20 condemned by the lab whose real visitors are fine. A clean Lighthouse score remains the more misleading of the two errors.

Two hand-written page descriptions still quoted the old figures and were corrected in the same commit as the regeneration: the speed-score analysis and Is my PageSpeed score bad?. Every figure in the bodies of those pages is computed at build time and was already correct; only the frontmatter, which is written by hand, had fallen behind.

1 September 2026 (release) — twenty-one new indices: architects, physiotherapy, veterinary practices

The index goes from 11 cohorts and 271 businesses to 32 and 1,156, adding three trades across all seven towns already covered. No town is new; no methodology changed; no existing score moved.

Cohort sizes run from 8 to 80. The three new trades were chosen because a national directory does not own their search results the way one owns construction — the sector’s own firms hold a meaningful share of the first page, so a league table of them measures the firms rather than the directory.

Thirteen businesses were removed before publication. Every seed is built from Google’s own category-filtered listings, so all 918 collected businesses carry the sector’s category. But the sweep matches a category held anywhere on a profile, and thirteen had a primary category plainly outside the trade — a recruitment agency, a tile warehouse, a radiator shop and a kitchen showroom among the architects, a pest-control firm among the vets. They were dropped from the records and from the seeds, and the affected cohorts were re-scored. Six borderline cases were kept, including a playground-architecture practice and a chiropody chain, on the grounds that removing a real firm from a public ranking is its own kind of error.

What adding 885 businesses did to the published findings. Every analysis on this site is computed from the datasets when the page is built, so tripling the corpus re-ran all of them. Most held:

Finding	11 indices	32 indices
Median Digital Visibility Score	51	51
Never named by an AI assistant	73%	73%
Rank for nothing	48%	46%
Speed’s correlation with findability	−0.13	−0.09

Two moved enough to matter. Median Technical rose from 75 to 88, which strengthens rather than weakens the finding that invisibility is not explained by a bad website. And the share of variation left within a trade-and-town cohort rose from about two thirds to 86% — so trade and town together now explain 14% of the difference between businesses, down from a third. Adding three trades and 885 firms made the trade effect smaller, not larger.

What the new cohorts do not have yet. Enrichment beyond the free sources has not been run on them, so Lighthouse and backlink figures cover only the original eleven indices. Enrichment is unscored, so no ranking is affected, and the lab-versus-field analysis still describes 66 businesses rather than the whole corpus. It says so on the page.

A note on how this nearly went wrong. The registry of indices was being kept uncommitted precisely so the monthly measurement stayed scoped to reviewed work. That convention failed the moment an unrelated commit swept the file up: the 1 September cron read 32 indices where 11 were intended and had to be stopped by hand seven hours in. Indices now carry an explicit publish flag, and the scheduled measurement, the site regeneration and both backfills all act on that flag alone. An index that no one has reviewed is now inert by construction rather than by memory.

1 September 2026 (correction) — “NAP consistency” was a domain check

The Local presence pillar listed NAP consistency as one of its five components. NAP means name, address and phone. The code compared the domain on the Google listing against the domain of the site being measured, and looked at none of the three:

function napConsistency(item, business) {
  return domainsMatch(item.url, business.url);
}

No score changes. The component did what it did and every published figure came from it; what was wrong was the label, which claimed three signals while measuring one. It is now called a website-to-listing domain match, in the methodology and in the code.

This is the fifth claim/code mismatch found this quarter, after the AI pillar’s four engines, the Content pillar’s two null signals, the sector-blind credentials check and the miscounted review velocity. The pattern is consistent enough to be worth naming: every one was a label that outran its measurement, and none would have been caught by a test, because in each case the code did exactly what it was written to do.

Phone numbers and postcodes are now captured from both the Google profile and the business’s own homepage, so a real comparison can be built. It is deliberately not scored yet — it will be validated against a full quarter of data before it is allowed to move anyone’s Local presence score, and it will be announced before it does. Neither value is published as structured data about any business.

Renaming the field carried its own risk, and it was handled rather than discovered later: every record collected before today carries the old key, so the scorer reads both. Verified by re-scoring all 271 businesses and confirming no score, rank, pillar score or percentile moved.

1 September 2026 — scorecard presentation; no scores changed

Four changes to how scorecards are published. No Digital Visibility Score, rank, pillar score or percentile moved — verified by diffing all 271 businesses before and after the re-score that this required.

Three retired scorecards. The three Cheltenham venues removed from the Gloucester restaurant index on 29 August (recorded further down this page) still had live scorecard pages. They were in the sitemap, indexable, and published a rank against a 47-business cohort that no longer exists, with scores measured before three later corrections. One showed 23/100 for a named restaurant on a page asserting every figure was reproducible from a dataset that no longer contained it. The pages are deleted and the URLs 301 to the index.

The underlying cause was that the generator only ever wrote scorecards and never noticed one had gone. It now warns when a page exists for a business that is no longer in the index. It warns rather than deletes: removing a page is an editorial decision with a redirect attached, and a script cannot know the right target.

Descriptions shortened. The search-result description ran to a median 171 characters with 248 of 271 over the ~160 that get shown, and what was being cut was Measured Q3-2026 — the freshness stamp, which is the one thing distinguishing this page from the business’s own site. Rewritten to lead with the score, the quarter and the rank: median 143, and 15 over the limit instead of 248.

Firms ranked nearby. Each scorecard now links the four firms closest to it in the ranking, beyond the two the pager already showed. Scorecards had a median of three inbound internal links and 21 had two or fewer; it is now a median of seven with none below four. A version linking the sector leader from every card was rejected — it gave the top firms 41 to 55 inbound links, which is the shape of a link scheme rather than a reader feature.

Businesses identified in structured data. Each scorecard now says which business it is about, using the Google place ID from its Google Business Profile match (243 of 271) and the Companies House number where a unique register match exists (69 of 271). Both are published in the open dataset as localMatch and context.companies so the identification is checkable.

No address, telephone number or rating is asserted about any business. The methodology explains why: those would be claims about what a business is, and this is a third party stating only where a business is authoritatively described.

30 August 2026 (methodology) — Content & trust stops being three boxes ticked

The Content & trust pillar read three fields: does the homepage link to an about page, a team page, and something naming a professional body. Three booleans, so the pillar could only ever return 0, 33, 67 or 100 — four values, for 10% of every published score.

Two further signals, indexed page count and content freshness, were listed in the methodology and were null on all 271 records. The pillar had never measured them.

That matters more than the coarseness. This is the strongest predictor of findability in the index — it correlates 0.53 with whether a business ranks at all, higher than any other pillar — and it was the most crudely measured thing we published.

What changed. The same single fetch of the homepage now yields six signals instead of three: the three original link checks, plus whether the site links to a blog or news section, how many visible words the homepage carries, and how long ago the page said it last changed (from JSON-LD dateModified, then the Last-Modified header). The pillar now takes 48 distinct values instead of 4.

Indexed page count was removed, not implemented. A real indexed-page count needs Search Console access to each business’s own property, which we do not and should not have. Counting pages a site publishes is a different quantity, and naming it “indexed” would have repeated the error instead of fixing it.

Nothing was re-purchased. The homepage fetch costs nothing, so this correction, like the credentials fix before it, was free.

Index	Content before	Content after	Overall before	Overall after
Gloucester accountants	33	54	52	53
Cheltenham law firms	67	77	63	64.5
Gloucester restaurants	16.5	26.5	48.5	50
Cheltenham construction	33	40	44	46
Bath estate agents	67	70	65	66
Swindon estate agents	100	80.5	69.5	70.5
The other five	—	small rises	—	0 to +1

Swindon is the clearest illustration. Every firm in that cohort ticked all three boxes and so every firm scored 100, which told a reader nothing. Measured on six signals they range across the band, and the cohort median falls to 80.5 — not because anyone got worse, but because a test that everybody passes is not a test.

An honest note on what this did not achieve. The finer measurement barely improved the pillar’s predictive power: its correlation with Visibility moved from 0.519 to 0.526, and with AI naming from 0.386 to 0.382. The three crude booleans were already capturing most of the available signal. What the change fixes is that the pillar is now a real measurement that matches its published description, and that it can distinguish between businesses that previously scored identically. It is not a claim that we can now predict findability better than yesterday.

Two findings fell out of the new signals, and neither is flattering to the sector: the median homepage was last modified 935 days ago, and only 115 of 271 businesses publish a blog, news or insights section at all.

30 August 2026 (methodology) — review velocity is now scored within sector

Earlier today’s correction to review velocity left an open question, and this page said we would rather ask it in public than settle it quietly. This is the answer: the component is now scored against a business’s own cohort instead of against a fixed threshold.

The old rule gave full marks at ten new reviews in ninety days, the same bar for everyone. Ten a quarter is routine for a restaurant and exceptional for a solicitor, so the component was substantially measuring which trade a business is in. Full marks now go to a business at its own index’s 90th percentile, so “active” means active for that trade.

Where a cohort has no distribution to compare against, the component is excluded rather than normalised. In four indices — Cheltenham, Gloucester and Worcester construction, and Gloucester accountants — three quarters or more of the businesses receive no reviews at all. Normalising there would have been worse than the problem: scaling to the cohort maximum would score a builder with three reviews at 100 and forty-three builders at 0, inventing a hundred-point spread out of a nought-to-three range, and ranking by percentile would give the tied zeros a midrank near the 41st percentile, rewarding a business for having no reviews. The Local pillar for those indices is computed from its remaining components. A builder is not digitally weak for working in a trade whose customers do not leave Google reviews.

160 of 271 businesses moved up, 47 moved down and 64 are unchanged. The largest single rise is 5 points and the largest fall is 2. Local pillar medians:

Index	Local before	Local after	Overall before	Overall after
Gloucester accountants	63	79	49	52
Gloucester construction	63	78.5	46	48
Cheltenham construction	61	76	40.5	44
Worcester construction	61	76	44	47
Exeter solicitors	78.5	80.5	72.5	73
Bath estate agents	88	84	65	65
Bristol estate agents	84	80	59	59
Bristol dentists	86	83	59	58
Cheltenham law firms	78	77	64	63
Swindon estate agents	91	85	70.5	69.5
Gloucester restaurants	92	84	49.5	48.5

The gap between restaurant and construction Local scores falls from 31 points to 8. That gap was the artefact this change exists to remove.

This is the ruler changing, not the businesses. It is also the first change we have published that moves some scores down: a restaurant or estate agent that was scoring well partly because its trade collects reviews now scores slightly lower. Nobody got worse at anything between these two figures. If you cited a score from earlier today, the superseded dataset is preserved alongside the new one.

Why change it mid-quarter rather than at Q4. Because the alternative was publishing a quarter of rankings we already knew were measuring the wrong thing. The comparison this breaks is against figures that stood for a few hours; the comparison it protects is Q3 against Q4, which is the one that will matter in September.

30 August 2026 (correction) — review velocity was miscounted, 105 scores changed

The Local presence pillar includes review velocity: how many new reviews a business received in the last 90 days. That count was wrong for every business we had measured, and it was wrong in the direction that lowers a score.

The cause was one line. Timestamps arrive as 2024-03-15 10:23:45 +00:00 and were normalised with .replace(' ', 'T'), which replaces only the first space. The result kept a space before the offset, Date.parse returned NaN, every review was skipped as unparseable, and the function fell through to zero. A firm with 425 lifetime reviews was scored as having received none in ninety days.

Because the failure was silent and uniform, it did not look like a bug. It looked like a sector where nobody reviews anything.

Two things were fixed, not one: the parser now accepts every shape the source returns, and when no timestamp in a response parses, the function returns “unmeasured” rather than zero. Absent data is excluded from scoring everywhere else in this pipeline; this was the one place it was silently counted against the business.

Reviews were then re-fetched for all 266 businesses with a resolved Google profile — reviews only, rewriting only this one field, so the movement is attributable to this cause and nothing else. 264 now carry a real count; 7 remain unmeasured and are excluded from scoring rather than counted as zero.

105 of 271 businesses have a different Digital Visibility Score. Every one of them went up, by at most 6 points. No business was harmed by the correction; some were under-scored and are now scored properly.

Index	Local before	Local after	Overall before	Overall after
Gloucester restaurants	76	92	46	49.5
Swindon estate agents	78	91	68.5	70.5
Bath estate agents	78	88	64	65
Bristol dentists	77	86	57	59
Bristol estate agents	79	84	57.5	59
Exeter solicitors	74.5	78.5	72	72.5
Cheltenham law firms	76	78	62.5	64
All four construction and accountancy indices	—	unchanged	—	unchanged

This movement is a correction to the ruler, not a change in the businesses. Nobody improved between these two figures.

What the correction exposed

The last row of that table is the important one. Construction and accountancy did not move at all, because their real review velocity is genuinely near zero: 43 of 52 Cheltenham builders and 36 of 41 Gloucester accountants received no Google reviews in the ninety-day window. Those are now verified zeros rather than parsing artefacts.

So the correction did not lift all sectors equally — it lifted the sectors that collect reviews, and left the ones that do not exactly where they were. The gap between restaurant and construction Local scores has gone from 15 points to 31.

That is a true measurement, and it is also a limitation of the pillar that we had not previously stated: review velocity is substantially a property of a trade, not of a business’s digital competence. Builders and accountants do not accumulate Google reviews the way restaurants and estate agents do, and the Local pillar now reflects that difference more strongly than before.

Comparisons within an index are unaffected — every business is measured against others in the same trade and town. Comparisons of the Local pillar between sectors should be treated with caution, and the methodology page now says so. Whether the component should be normalised within sector is an open question we would rather ask in public than settle quietly.

The superseded datasets are kept alongside the corrected ones, so anyone who cited a figure before today can still find exactly what they cited.

30 August 2026 (correction) — the AI pillar measures one assistant, not four

The methodology stated that AI search presence was measured across “AI Overviews, ChatGPT search, Perplexity and Gemini”. It never was. The pipeline has only ever queried Perplexity, model sonar, on five prompts per index.

No score changes: the code did what the code did, and every published figure was produced by it. What changes is the claim made about those figures, which was broader than the evidence behind them.

The page has been narrowed to the truth rather than the pipeline widened to the claim, because the second option would have re-measured the pillar and made this quarter incomparable with last. Read every AI presence score as evidence about one assistant on five prompts on the measurement date — a zero means “not named by Perplexity”, not “invisible to AI”.

29 August 2026 (correction) — three venues removed from Gloucester restaurants

Three entries in the Gloucester restaurant index were not in Gloucester. The seed sweep searches a radius around the town centre, which reached into Cheltenham: a Ramada by Wyndham, the restaurant inside that same hotel listed separately, and a hotel restaurant trading as Cheltenham. The cohort is now 44 businesses and its median moves from 47 to 46.

Two rules were tightened so the same sweep cannot repeat it. Hotel groups are now treated like other national chains — a branch restaurant lives on the group’s website, not its own — and the review step checks whether a venue’s own name places it in a different town from the index it is being added to.

Being ranked in the wrong town’s league table is a real harm to those three businesses, and it was our error rather than theirs.

29 August 2026 (later) — credentials check corrected, 43 scores changed

The Content & trust pillar checks whether a homepage links to credentials. That check matched only property-sector bodies — Propertymark, NAEA, ARLA, RICS — while being applied to every index, so construction, law, accountancy and dental cohorts scored near zero on a signal many of them actually pass. It is now specific to each sector: ICAEW and ACCA for accountants, the SRA and Law Society for solicitors, FMB, Gas Safe and NICEIC for construction, the GDC and CQC for dentists, and so on.

41 businesses had the check change, and 43 of 274 now have a different Digital Visibility Score. Sector medians moved as follows:

Index	Before	After
Exeter solicitors	69	72
Cheltenham law firms	61	62.5
Bristol dentists	56	57
Gloucester accountants	48	49
Gloucester restaurants	47	46
The other six	—	unchanged

This movement is a correction to the ruler, not a change in the businesses. No firm improved or declined between these two figures; a signal we were measuring badly is now measured properly. Treating it as performance would be exactly the misreading this page exists to prevent.

Only the homepage check was re-run. Re-collecting every pillar would have fixed the defect and simultaneously re-measured five other signals, so the resulting movement could not be attributed to the correction rather than to sites genuinely changing. Nothing was re-purchased; the fix cost nothing.

The superseded datasets are kept alongside the corrected ones as q3-2026-superseded-2026-08-29.json in each index folder, so anyone who cited a figure before today can still find exactly what they cited.

29 August 2026 — four new indices, and a corrected credentials check

Four cohorts added, taking the index from 79 businesses to 219 across eleven indices:

• Gloucester accountants — 41 firms, median 48
• Cheltenham construction — 52 firms, median 40.5
• Worcester construction — 55 firms, median 44
• Gloucester restaurants — collected this release

Seed lists are now generated from Google Business listings and then reviewed by hand rather than assembled manually. Roughly a third of each raw sweep is unusable — national franchise and chain location pages, businesses returned by radius but based in another town, and firms in an adjacent trade — and is removed before anything is collected. Cohorts are not truncated to a target size: the sweep is ordered by review count, so trimming the tail would drop the smallest and newest firms first and bias every published median upward.

A scoring input was wrong, and it is worth stating plainly. The Content & trust pillar checks whether a homepage links to credentials. That check matched only property-sector bodies — Propertymark, NAEA, ARLA, RICS — while being applied to every index. Construction, law, accountancy and dental cohorts therefore scored close to zero on it, not because those firms show no credentials but because the detector could not see the ones they show. Credential terms are now specific to each sector. Scores in this release still reflect the old check; they correct themselves at the next collection, and the movement will be logged here rather than passed off as businesses improving.

Three sources are now recorded alongside every measurement and carry no weight in any score: a first-party crawl that obeys robots.txt, Chrome UX Report field data where an origin has enough traffic to appear, and Companies House registry data. Each index page now states what the public register says its cohort actually is, which is often broader than the index title — six firms trading as “Accountancy” in Gloucester are registered as bookkeepers.

27 August 2026 — Q3-2026 release: all six pillars, seven indices

Measured 27 August 2026. Seven indices, 79 businesses, all six pillars live for the first time: Speed & Core Web Vitals (20%), Technical (20%), Local presence (20%), Visibility (15%), AI search presence (15%), Content & trust (10%).

• Local presence coverage improved from 7 of 18 firms to 17 of 18 in the Bristol estate agent cohort, after three successive matching passes. One Cheltenham firm remains unmatched and is excluded from that pillar rather than scored at zero.
• Weights are renormalised per business across the pillars actually measured for that business, and the effective weights are published in each dataset. A missing pillar never advantages a firm.
• Data published under CC BY 4.0 — reuse permitted, including commercially and in AI answers, with a named credit.

Structural changes shipped the same day, with no effect on any score:

• Every pillar row on all 79 scorecards now links to the knowledge-base article explaining what that pillar measures, and each scorecard carries a Where to get this fixed section naming that firm’s lowest-scoring pillars among those we publish remediation guidance for.
• The glossary was re-framed. It had been presented as a standalone catalogue of statistical methods, which misdescribed its relationship to the index. Scoring uses plain arithmetic by design — ratios, weighted sums, a clamp — so that any published figure can be recomputed without trusting a model. The glossary is the companion toolkit for analysing published data, not the thing that produces it, and now says so.
• Structured data consolidated onto a single author and publisher entity. Generated index pages had been declaring an organisation that no other page on the site declared.
• Measurement date corrected. measuredAt had been derived from the time the scorer ran rather than the time the data was collected, so re-running the scorer over unchanged data moved the stated measurement date. It is now taken from the collection timestamps, including the index-level geo-grid and intent sidecars.
• One knowledge-base article and four other pages linked to a blog post that was never migrated from WordPress. Repointed.

22 June 2026 — six further South West indices (v0)

Bath, Swindon, Exeter, Cheltenham, Gloucester and Bristol dentists published as Q2-2026 v0 — a partial release measuring fewer than six pillars, with the measured subset and renormalised weights stated on each page.

20 June 2026 — index launched

The methodology, the index hub and the collection pipeline published. The first cohort, Bristol estate agents, followed on 22 June as Q2-2026 v0.

How corrections work

If a figure here is wrong, it gets fixed and the change is logged on this page with the date. Request a correction with the business name and what you believe is inaccurate.

Scores are never adjusted on request. The methodology is applied identically to every business in a cohort, including businesses that ask.