This page brings the 80-check x-ray to run on your own site, the five layers of the Cognitive Architecture of SEO with the failure mode of each one, the spreadsheet for tracking mentions in generated answers, and the references with access dates. The material is open and can be used in your own audit.
The tool below runs in your browser and stores your answers there. Nothing is sent to me, and nothing disappears if you close the tab. Level 1 of each symptom runs without any paid tool, just with the browser open.
Self-assessment
Eighty technical checks across six symptoms. Level 1 runs in the browser, with no paid tool. Answer what you can and see the score, with the action plan, in the last tab.
Score
0.0/30
Start with symptom 01
0 of 80 answered
The eighty checks came out of my audit work.
The symptoms are organized by the five layers of the Cognitive Architecture of SEO, an original model by Amanda Noronha.
Amanda Noronha®, 2026.
The five layers, and what each one answers
The x-ray is organized around a reading model, without prescribing an order of execution. There are five cumulative layers: a failure in the first one cancels the others, and solving the first three does not guarantee the last two.
Layer
The question it answers
The failure mode
1. Crawlability crawling, indexing, links
Does the system reach the content?
The content exists and the crawler does not reach it.
2. Structure HTML, schema, architecture
Can the content be retrieved in parts?
The crawler arrives and cannot tell what is the price, what is the author and what is the specification.
3. Interpretation vectors and semantic context
Does the system understand what it is about and who it belongs to?
The page is read and ends up far, as a vector, from the way the question was asked.
4. Memory entity and authority
Is the brand remembered when the answer generates no click?
The brand is understood on one page and not recognized as the same entity on the others.
5. Action decision and execution by agents
Does the person or the agent decide based on it?
The brand is cited in the answer and is not the final recommendation.
Where the check stops being verifiable
Thirty-five of the eighty checks cite the documentation that supports them. The source appears inside the item itself, in addition to the list at the end of the page, because whoever is checking needs it right there, at hand.
The other forty-five cite nothing, by declared decision.
They are of three types. Measurement method, such as running the same question on three platforms within the same window and recording the brand’s position. Vocabulary reading, such as comparing the page text with the way the question actually arrives. Editorial decision, such as the publishing cadence that sustains recurrence. None of the three has official documentation to cite, because none is a platform rule. They are craft.
Hanging a Google document on a check about editorial cadence would produce a decorative source. It would not support the check, and it would make the technical reader distrust the correct ones as well.
The distribution says something about the problem
Symptom
With source
Layer
01 Inconsistent entity
10 of 13
Memory
02 Data unreadable by agents
11 of 14
Structure
03 Absence across platforms
2 of 13
Memory
04 Unverifiable trust
11 of 14
Action
05 Vector distance from the audience
0 of 13
Interpretation
06 Non-recurring presence
1 of 13
Memory
Symptoms 01, 02 and 04 live in the memory, structure and action layers, precisely where platforms publish rules. Schema, authorship markup, rendering and crawl control have plenty of official documentation, and most of their checks cite a source.
Symptoms 03, 05 and 06 live where no platform publishes rules. How to measure presence in a generated answer, how close your page’s vector gets to the shape of the question, and how much repetition sustains memory are open questions. Thirty-six of the thirty-nine checks in these three symptoms have no source, and that gap is the map of the open frontier.
The prompt tracking spreadsheet
Symptoms 03 and 06, absence across platforms and non-recurring presence, depend on observing the generated answer over time. One round of the x-ray shows a snapshot of the day, and the spreadsheet shows whether your brand comes in, drops out or changes position week by week.
It does the same tracking a paid tool does, in five columns: one round per week, always with the same questions and on the same platforms. The scope and the date are what make the data citable later.
Prompt
Platform
Date
Brands cited
Your position
“best [categoria] for [uso]”
Platform 1
17/09
A, B, C
not cited
“best [categoria] for [uso]”
Platform 2
17/09
A, C, your brand
cited, 3rd
“[categoria] trusted in [região]”
Platform 1
18/09
your brand, B
recommended
Two rules that change the result. The first is to run it logged out, because a logged-in session carries history and history contaminates the answer. The second is to record the date, because the same platform answers differently a day apart, and without a date there is no way to know which of the two answers you measured.
Where the spreadsheet is: the link comes in the diagnosis PDF, which you generate in the Result tab of the x-ray. It opens as a copy in your Google Drive and brings the eighty checks, the weights, the sources and the score calculation, so you can repeat the diagnosis a few months from now, plus the prompt tracking tab in the format of the table above.
References
Each source was checked on the access date indicated.