Evidence where it exists. Candor where it doesn't.
Atlas separates measurement from interpretation, names the limits of what it can observe and withholds conclusions the evidence cannot support.
An audit that scores everything is easy to build and easy to sell. It is also how a business ends up told it has no reviews by a tool that never looked, or ranked 100/100 by a crawler that read one page.
Every rule below exists because the alternative produced a wrong number we would have had to stand behind. When Atlas cannot measure something, the report says so — and that sentence is the most load-bearing one we ship.
- 01A score we could not measure does not exist.
Every audit reads a limited slice of the world. When too little of a website could be read to support an overall number, the report says “not enough measured” — it does not average what remains and publish the result.
The same rule applies inside a report. A category with no usable evidence reports “not measurable”, which is a different statement from a low score. Zero is a claim. We only make it when we measured it.
- 02We say where our sight ends.
Our crawler reads the HTML a visitor’s browser is served. Content that loads by JavaScript, tags injected server-side, and accounts that exist but are not linked from the site are invisible to it.
So our reports say “not found in the pages we could read” — never “you don’t have this.” If a finding depends on something we could not see, the report names that limitation instead of papering over it.
- 03Measurement and interpretation are labeled apart.
Numbers in an Atlas report come from evidence collected during that run. Interpretation — what the numbers suggest, what we would do next — is labeled as interpretation and cited back to the measured reports it reads from.
You can always tell which sentence is a measurement and which is our judgment. Both matter. They are not the same thing, and a report that blurs them is selling confidence, not research.
- 04Claims need signatures. Ratings need nobody’s.
When Atlas drafts a page for a business, it will not print a certification, licence, or years-in-business claim unless the business has put its name to that exact wording — and the wording is checked again by a person before anything is approved.
Reviews, star ratings, and testimonials are refused outright. There is no version of an invented review that a signature makes honest, so the system will not generate one at all.
- 05Automation drafts. People decide.
Review requirements depend on the output and the agreed workflow. Public sample reasoning is not an approved recommendation. Client work needs an accountable review before it is treated as an execution instruction.
When the platform declines a request, that is the product working. The refusal states its reason in plain words, and the reason travels with the work, not in a log nobody reads.
When an Atlas report recommends something, the recommendation is standing on evidence you can inspect, gathered from your market at a stated moment, with its blind spots named.
A reported observation should be supported within its stated scope. An explanation or recommendation is a judgment to evaluate, not an observed fact. When evidence is missing or a finding is corrected, the next decision should reflect that.
Agree what success means. Then measure it.
These standards explain how findings are supported. The measurement plan defines the qualified outcome, source, baseline, cost basis and next decision.
See what the engine measures — and what it hands to people.
Atlas helps reveal the opportunity. Our people scope, approve and deliver every service — starting with a free market audit and a consultation, before any scope or payment.