Lumis · public methodology v1.0 · 2026-05 · printed 2026-05-15

How we measure.
No black box.

Lumis publishes every weight, every source, every threshold. Not out of goodwill — out of discipline. A methodology you can't reproduce isn't a benchmark, it's an opinion. Transparency is our least copyable moat.

01 · The eight dimensions

Seven dimensions scored
at equal 14.29 %. One indicator.

The official score is the arithmetic mean of D1 to D7, each weighted 1/7. D8 (observable paid footprint) is reported outside the average — it's inferred, not directly measurable, so displayed as an indicator rather than a judgment.

Dim
Name
What we measure
Source / tool
Weight
D1
Technical performance
Core Web Vitals (LCP, INP, CLS), Lighthouse score, server response time, image optimization, JS-blocking.
Google PageSpeed Insights · Lighthouse 11 · WebPageTest
14.29 %
D2
SEO on-page
Schema.org markup, sitemap.xml, canonical tags, tree depth, title and meta-description quality, internal linking.
Proprietary crawler · schema.org validator
14.29 %
D3
Indexation
Indexed vs. crawled pages, presence in top-100 SERP on 50 sector prompts FR-CH, new-page indexation velocity.
Google Search Console API · SerpAPI · crawl-frequency analysis
14.29 %
D4
Multilingual coverage
FR/DE/IT presence, correct hreflang tags, detection of Helvetisms (septante, Velo, parkieren…) vs. FR-FR/DE-DE translations.
Crawler + proprietary Helvetism lexicon
14.29 %
D5
AI Visibility · SEO-LLMO
Aggregate of 4 sub-scores at equal 25 % each: AI readability · Google Maps & AI Overviews · Conversational AI · Future Readiness.
Panel of 150 prompts FR/DE/IT sent weekly to ChatGPT · Claude · Perplexity · Gemini
14.29 %
D6
Accessibility
WCAG 2.1 level AA. Contrast, ARIA structure, text alternatives, keyboard navigation, semantic hierarchy.
axe-core · Lighthouse a11y · manual sample audit
14.29 %
D7
Social presence & directories
NAP consistency across 15 CH directories (local.ch, search.ch, swissfirms.ch, comparis.ch, moneyhouse…), engaged LinkedIn/IG/FB traffic, Trustpilot score.
Proprietary scraper + Meta Graph API + LinkedIn API
14.29 %
D8
Paid footprint
Active ads detected in Google Ads Transparency, Meta Ads Library, LinkedIn Ads. Outside the average — observable indicator, not a judgment.
Public APIs of ad libraries
outside
02 · Public Scope Card · the measurement contract

Every anchor has its scoping sheet.

Without published perimeter, every benchmark is contestable. For each anchor, Lumis publishes a Scope Card stating exactly what's included, what's not, on which date, with which methodology version. The PDF the visitor receives also contains its own Scope Card for symmetric comparison.

Example · Migros Method v1.0
Benchmarked URLhttps://www.migros.ch
LevelCorporate retail group (FR / DE / IT)
Geographic perimeterSwiss properties only
Includedmigros.ch landing, migros.ch/fr, migros.ch/de, migros.ch/it, main sitemap, blog
ExcludedMigros Online, Galaxus / Digitec (separately scored), regional cooperatives (Geneva, Vaud, etc.), M-Industrie sites, mobile apps
Why this URLMost representative entry point a competing Swiss retailer would compare against for its own score
Crawl dateWeekly on Monday · JSON audit archived
Audit trailcrawl-migros-2026-05-13T08:34:12Z.json · reproducible · available on request
03 · Calibration round & JSON audit

Before publication.
After publication.

A methodology that isn't calibrated and doesn't archive its raw inputs is not defensible. Lumis does both.

Before publication

Mandatory calibration round

Full crawl of the 15 anchors in private beta. We look at the score distribution. If 13 out of 15 cluster in Tier S 90+, the scale is broken — we don't publish a benchmark where everyone's S. If everything clusters in B 60-74, same.

The S / A / B / C thresholds are tuned on the real sample, not set arbitrarily. That's two days of delta before public ship, and it's non-negotiable for defensibility.

After publication

JSON audit per crawl

Every produced score generates a dated, archived JSON artifact containing the raw inputs (Lighthouse JSON, parsed sitemap, hreflang count, Trustpilot snapshots, raw LLM responses for D5).

The Scope Card links this artifact. That's what makes the methodology defensible against contestation: here's exactly what we measured, here's the date, here's the code. Reproduce it if you want. Without that, you're defending numbers on your word alone.

04 · Versioning & methodological honesty

Method v1.0.
Evolving, not frozen.

The methodology is versioned like code. Each major evolution (weight change, dimension addition, anchor removal) bumps the version. Old versions remain accessible. No historical score change is silent.

v1.0 · Spring 2026

Launch

Eight dimensions, seven scored at equal 14.29 % each. D8 outside the average. Calibration round on 15 anchors. D5 sub-scores at equal 25 % each pending six months of empirical data.

v1.1 · Autumn 2026

Calibrated weighting

After 90 days of real comparisons, recalibration of D1-D7 weights if one dimension correlates more strongly with measured SME conversions. Recalibration of D5 sub-scores. Mandatory public justification.

v2.0 · 2027

Multi-anchor per segment

Addition of 2-3 anchors per segment (Big 4 accounting = KPMG + PwC + Deloitte; pharma = Roche + Novartis + Lonza). Enables sub-sector comparison. Public methodology for additions.

Every defensible methodology
starts with its publication.

No black box. No secret algorithm. No arbitrary thresholds. A defensible method is Lumis's promise to its clients — and the moat global tools cannot copy without publishing their own.

Request early access →