How the 2K+ firm comparison set is built.
This is a mock client-facing version of the data sources page: clear enough to build trust, honest enough about automation, and simple enough to support the pattern story without burying people in methodology.
The dataset tracks what a client can actually encounter.
Sources include public law firm websites, homepage screenshots, visible navigation, logo and brand assets when available, proof signals, social placement, calls to action, color behavior, and category-level language.
The point is not to manually adjudicate every firm one by one. The point is to use a large enough dataset to see repeated choices clearly: which colors dominate, which symbols recur, which proof formats become table stakes, and where a firm can still create memory.
Find the market
We collect public law firm websites across Colorado and adjacent legal categories so each report is grounded in a real comparison set, not a hand-picked mood board.
Capture visible signals
The profile layer records homepage screenshots, color families, logo motifs, hero imagery, proof formats, navigation patterns, calls to action, and visible social placement.
Normalize the messy parts
Automated scans turn public websites into comparable labels. Similar CTAs roll up together, colors are grouped into families, and visual patterns become searchable signals.
Use volume for pattern confidence
At this scale, the useful question is not whether every individual scrape is perfect. It is whether the same cues repeat often enough to reveal the market pattern.
Scraped data is useful because patterns survive small imperfections.
Web scraping is imperfect because websites are imperfect: assets move, copy changes, colors render differently, and some patterns need judgment. That is why the report uses the dataset for trend confidence, then keeps the claims directional and client-readable.
A single mislabeled firm would be a problem if the goal were a forensic inventory. It matters much less when the goal is trend evidence across 2K+ public sites. We are looking for repeated market behavior, not pretending the internet is a clean spreadsheet.
A small slice of the source layer.
The full profile database stays internal. This partial sample shows the kind of structured record behind each report without turning the dataset into a public directory.
Transparent enough to trust.
The report should make the source of the claims feel clear: public websites, repeatable profile fields, and market-level pattern counting.
Private enough to stay useful.
The full database, URLs, and categorization rules remain internal so the research layer stays durable and does not become a public scrapeable directory.