Self-audit
The audit, run on this site.
This is the diagnostic audit in the form a client receives it, applied to auspexseo.com on 9 September 2026. It is published because there are no client audits to show yet, and because the form carries most of the argument. Every figure below is a reading taken from the live site, with the instrument named. Where a gate cannot be measured from outside, the page says so and claims nothing.
- Gate 1 Crawl Can a bot fetch the URL? robots.txt · server errors · CDN and WAF rules · budget spent elsewhere
- Gate 2 Index Did the engine keep it? thin · duplicate · canonicalised away · noindex
- Gate 3 Retrieve Was it pulled for this query? does not answer the sub-query the fan-out issued
- Gate 4 Cite Did it get named in the answer? present in the source, not lifted out
- Gate 5 Influence Did it shape what was said? cited navigationally, contributing nothing
Scope and instruments
| Item | This reading |
|---|---|
| Site | auspexseo.com, nine pages and a 404, static HTML on Cloudflare Pages |
| Date | 9 September 2026, with delivery readings from 7 and 8 September |
| Gate 1 | Requests to the live site from one location, each carrying a named crawler's user-agent string. Observed status codes and headers. |
| Gate 2 | The HTML as fetched, without rendering. Canonical, robots meta, sitemap membership, structured data, and whether the content is present before any script runs. |
| Gates 3 to 5 | Not measured. See below. |
| Delivery | Lighthouse 13.4.1 in Chrome 152, three sequential runs per page per form factor, 48 runs per reading, medians. Five readings between 7 and 8 September. |
| Not used | Search Console, server logs, a rendering check with a headless browser at scale, a prompt panel. A client audit uses all four. |
Findings, ranked
Two findings at medium severity, both in delivery, both found by the measurement and both fixed the same day. Nothing at high. Gates 1 and 2 pass on every check that can be made from outside. Gates 3 to 5 are not claimed.
| Finding | Severity | Gate | Basis | Status |
|---|---|---|---|---|
| Eight inline drawings added on 8 September cost 85 to 457 ms of mobile first paint on the pages that carry them, in proportion to how many each carries. Transfer weight rose 1.8 KB. Desktop was unchanged. The cost was the layout of SVG text before first paint on a throttled CPU. | Medium | Delivery | Provable. 48-run reading against the previous 48-run reading, medians outside the run-to-run spread on five of eight pages. | Fixed the same day. Layout of below-fold drawings is deferred until they approach the viewport. The next reading was within 4 ms of the pre-drawing state on every page. |
| The stylesheet was versioned by a query string. A request for a new version during the half-minute after a deploy was answered by the previous deployment and cached at the edge, under the new version, for a year. | Medium | Delivery | Provable. The cached response was 461 bytes shorter than the file deployed, with the edge reporting a cache hit and an age of 194 seconds. | Fixed the same day. The version is now in the file name, so an early request fails briefly instead of caching the wrong file. The deploy step refuses to upload unless every asset a page references is present. |
Both findings were introduced and found by the same people, on the same day. A client audit is done against a site whose authors have not spent a fortnight measuring it. It finds more.
Gate 1. Crawl
Every named agent receives a 200 on every page tested, from the homepage to the
deepest reference page. The CDN is Cloudflare with no bot rule in front of the
site. robots.txt allows every agent it names and distinguishes the
retrieval agents, which fetch to answer a live question, from the training
crawlers, and from the two tokens no crawler fetches under.
| User-agent presented | / | /geo/ | /pricing/ | Edge |
|---|---|---|---|---|
| Googlebot | 200 | 200 | 200 | Cloudflare, no challenge |
| Bingbot | 200 | 200 | 200 | Cloudflare, no challenge |
| OAI-SearchBot | 200 | 200 | 200 | Cloudflare, no challenge |
| ChatGPT-User | 200 | 200 | 200 | Cloudflare, no challenge |
| PerplexityBot | 200 | 200 | 200 | Cloudflare, no challenge |
| ClaudeBot | 200 | 200 | 200 | Cloudflare, no challenge |
| GPTBot | 200 | 200 | 200 | Cloudflare, no challenge |
| Applebot | 200 | 200 | 200 | Cloudflare, no challenge |
| Amazonbot | 200 | 200 | 200 | Cloudflare, no challenge |
| No user-agent | 200 | 200 | 200 | Cloudflare, no challenge |
What this shows is what the site returns to a request that presents a crawler's name from one address in Seattle. It does not show what the crawler itself receives, since an edge rule can be keyed to address ranges rather than names. A client audit reads the server logs and checks each crawler's address against the ranges its operator publishes. That is the difference between an observed response and a verified one, and this page has only the first.
Gate 2. Index
All nine pages carry a self-referencing canonical and appear in the sitemap. The
only noindex on the site is on the 404 page, where it keeps the
real 404 that Pages serves out of the index. Every page's headline and every
paragraph are present in the fetched HTML before any script runs. Each page has
exactly one script element and it is the structured data, so nothing on the site
waits for rendering.
| Check | Observed | Basis |
|---|---|---|
| Canonical | Self-referencing on 9 of 9 pages | Provable |
| Sitemap | Lists the 9 pages and this one. No priority, no changefreq, no lastmod, since none would be accurate | Provable |
| Robots meta | noindex on the 404 only | Provable |
| Rendering dependence | None. Content complete in fetched HTML. One script per page, all of it JSON-LD | Provable |
| Structured data | Organization, Person and WebSite on the homepage. Article with breadcrumbs on the reference pages. OfferCatalog with two Offers on pricing | Provable |
| Eligibility | Each page is fetchable, canonical, listed and not excluded, so each is eligible for indexing | Inferred |
| Indexation | How many of the pages the engines hold is not part of this reading | Not claimed |
Eligible is not indexed. The count of indexed pages, and the reasons any page is excluded, come from Search Console reconciled against a crawl. That reconciliation is the first deliverable of a client audit, and it is left out here because the reading is limited to what anyone can observe from outside.
Gates 3, 4 and 5. Retrieve, cite, influence
Not measured. No panel has been run for this site's own name, so nothing is claimed about whether an engine retrieves these pages, names the practice, or is shaped by what the pages say.
A reading of these gates takes a frozen panel of at least 120 prompts, approved in writing, five repetitions per prompt per engine, reported per engine with a confidence interval, with the instrument declared on every number and a noise floor measured first on the category and an approved competitor set. A single prompt asked once carries roughly 44 points of error. A figure produced any other way would be a guess with a decimal point, and one of the things this site declines to sell is a figure like that. When a panel is run for this site it will be reported here, per engine, with its interval.
The measurement design, with the numbers each part comes from →
Delivery
The engineering layer is where the two findings were. Every page is between 3 and 10 KB of HTML over the wire, one 9 KB stylesheet and one 15 KB font file, five requests in all and none of them script. The largest paint on every page is a text node. The response headers carry a content security policy with no inline allowance, strict transport security, and a frame denial.
| Page | HTML over the wire | Requests | Largest paint |
|---|---|---|---|
| / | 5.1 KB | 5 | Lead paragraph |
| /geo/ | 9.4 KB | 5 | Answer paragraph |
| /methodology/ | 5.6 KB | 5 | Answer paragraph |
| /services/ | 6.7 KB | 5 | Answer paragraph |
| /pricing/ | 4.1 KB | 5 | Answer paragraph |
| /glossary/ | 6.0 KB | 5 | Answer paragraph |
| /about/ | 3.3 KB | 5 | Answer paragraph |
| /contact/ | 4.2 KB | 5 | Answer paragraph |
- 89–90
- 7 September · before the redesign
- Largest paint 2,723 to 2,797 ms. The whole gap was a cross-origin stylesheet on the critical path, rendering-blocked for about 820 ms.
- 89–91
- 8 September · after the redesign
- Within the previous run-to-run spread on every timing metric. Reported as flat, because it was.
- 97–99
- 8 September · typeface self-hosted
- Largest paint 1,575 to 1,640 ms, a median improvement of 1,138 ms. One request removed. Layout shift to zero on every page.
- 96–98
- 8 September · eight drawings added
- Largest paint up 85 to 457 ms on the pages with drawings. Finding 1 above.
- 97–98
- 8 September · drawings deferred
- Within 4 ms of the self-hosted reading on every page. The drawings now cost 2.3 KB and nothing else.
What this reading cannot see
- It is taken from outside. Search Console and the server logs are what turn an observed response into a verified one, and neither is used here.
- It presents crawler names from one address. A rule at the edge keyed to address ranges would not show.
- The delivery figures are lab readings on a simulated connection, three runs per cell. They are a baseline for comparison, not a claim about any real visitor's experience. There is no field data for the domain yet.
- The two findings were caused by the site's authors and found by them the same day. A site nobody has been measuring for a fortnight has more to find.
- Gates 3 to 5 are not measured. A page that presented a figure for them without a panel behind it would be doing what this site argues against.
What the client version adds
The same five gates, with the instruments this page lacks. Indexation coverage from Search Console reconciled against a full crawl, so every excluded URL has a stated reason. Crawler identity verified from your logs against the published address ranges. A rendering check on any page that depends on JavaScript. The frozen prompt panel, the competitor set and the noise floor for gates 3 to 5, reported per engine with intervals. And a fix list ranked by the clicks and impressions at risk against the effort to remedy each item, which your developers can act on with or without us.
Start with the audit
Ten business days, and a ranked fix list you can act on with or without us. Send your domain and a sentence on what you sell. You will get an initial read on whether an audit is worth your money before anything is quoted.