Free read · outside-in
paulseperformance.com
Report/r/fa7a2c6ca6e5
13 of 85 checks need work
0 fail·13 weak·72 pass
- 98Performance
- 96Accessibility
- 100Best practices
- 100SEO
- 3/3Agentic
98 is Lighthouse's score for speed on this page, measured on a desktop with no throttling, which is close to what a visitor on a normal connection gets. It counts nothing above, and a page can score 100 there and still fail checks here. Lighthouse 13.5.0, desktop, run on a real browser.
The stress test, printed second because it is a test and not the norm: the same page on Lighthouse's mobile profile (1.47 Mbps down, 150 ms round trip, CPU slowed 4×, Moto G Power (2022)) scores 66.
Accessibility, best practices and SEO above are the phone run's: this reading was stored before the desktop run kept its own category scores. Performance and paint time are the desktop run's.
How this number was measured
Desktop, the number above
- Lighthouse
- 13.5.0
- Form factor
- desktop
- Throttling
- simulated in software, not observed on your connection
- Device
- a desktop window, 1350×940 at 1× pixel ratio
- CPU
- no CPU slowdown
- Network
- 10.24 Mbps down, 10.24 Mbps up, 40 ms round trip
- Repeat it
- npx lighthouse@13.5.0 <url> --preset=desktop --output=json
Stress test, printed second
- Lighthouse
- 13.5.0
- Form factor
- mobile
- Throttling
- simulated in software, not observed on your connection
- Device
- Moto G Power (2022), 412×823 at 1.75× pixel ratio
- CPU
- slowed 4×
- Network
- 1.47 Mbps down, 0.68 Mbps up, 150 ms round trip
- Repeat it
- npx lighthouse@13.5.0 <url> --preset=perf --output=json
The desktop run is what a visitor on a normal connection gets, so it is the figure the dials carry. The phone profile below it is Google's floor: a page that passes it passes anywhere, and a page that scores badly there can still feel quick on a real phone. This report was stored before the runner kept its full configuration, so the phone figures are the profile's own definition rather than a copy of the run.
What each layer scored
- Server runtime0/10%
- Email and DNS2/540%
- Discovery and AI readiness3/650%
- Security headers6/1155%
- On-page basics29/4171%
- Doors a scanner knocks on7/978%
- AI search4/580%
- Crawlability and indexation5/683%
- Content1/1100%
The short version
Nothing is broken. 13 of the 85 checks are worth tidying.
Some pages have nothing for a search engine to show under the link. The description is the sentence under the link. Without one, the engine writes its own.
Also worth a look: one H1 per page, structured data, compressed transfer, descriptions too short, pages with no H1, and 7 more, which fall into 2 in on-page basics, 2 in email and DNS, 1 in crawlability, 1 in AI search, 1 in security headers.
Section 02 lists every check, what it found, and what closes it.
Read from the outside. Nothing on the site was changed, and we did not log in.
Of 85 checks, 0 fail and 13 are weak. Nothing fails outright. This reads the instructions your site sends to browsers and mail servers, not the quality of what you make. Every check behind the three check figures is in section 02.
By what a failure costs when it lands: 3 fault, 10 polish. The order is about the kind of failure, not about your revenue.
A visitor on a desktop waits 0.7 s before the main content appears, and the page asks them to download 378 KB to see it.
01Change since the last audit
What changed since your last audit
Compared with the audit of 17 November 2026.
- The response headers this audit grades changed since 17 November 2026.
02Everything we checked
What you can fix yourself, without us
85 checks across 10 layers. A FAIL is something a scanner can act on. A WARN is a door that is common but does not need to be open.
- Email authentication needs work: the DKIM key is 1024 bits, and 2048 is the minimum; SPF does not end in -all.
- Give each page one H1 that says what the page is about.
- Write a meta description of 120 to 160 characters; it is the sentence search results show.
- Add the missing security headers (HSTS, CSP, Referrer-Policy, framing protection). They are free and they close the cheap attacks.
- Turn on compression for the HTML.
These are free, and you are welcome to do them with a developer you already trust. If that is enough, this audit has done its job.
Not requested: the PII audit. Nothing in this report says anything about it.
| Check | State | What it is, and what closes it |
|---|---|---|
| Email and DNS2 of 5 need work | ||
| SPF | WARN | one record, but it does not end in -all"v=spf1 include:_spf.mx.cloudflare.net ~all"Publish one SPF record ending in -all, under ten lookups. |
| DKIM | WARN | key on resend, 1024-bit (2048 is the current minimum)Publish the DKIM key your mail provider issues. |
| 2 passed Receives email at the domain, DMARC. 1 not graded Branded email. | ||
| Security headers1 of 11 need work | ||
| HSTS | WARN | max-age=86400Send Strict-Transport-Security for a year with includeSubDomains. |
| 6 passed Legacy XSS header, Content Security Policy, MIME sniffing blocked, Clickjacking protection, Referrer policy, Server banner. 4 not graded HSTS preload, security.txt, Permissions policy, Cross-origin opener policy. | ||
| AI search1 of 5 need work | ||
| An answer block where a searcher lands | WARN | 2 of 4 pages carry an answer block; none on /, /forge/./, /forge/, /case/momsbalm/, /case/motohorek/Add a 40 to 90 word answer under the H1 that begins with the question. |
| 4 passed llms.txt maps real pages, Who says it, in JSON-LD, Crawler rules for answer engines, Every claim of freshness is checkable. | ||
| Crawlability and indexation1 of 6 need work | ||
| www and bare domain agree | WARN | www ends at www.paulseperformance.com, bare at paulseperformance.comPick www or the bare domain and 301 the other to it. |
| 5 passed Search engines allowed, Indexable, Canonical tags, HTTP redirects to HTTPS, Missing pages return 404. | ||
| On-page basics8 of 41 need work | ||
| Meta descriptions | WARN | 1 of 7 pages are missing a description or sit outside 80 to 200 characters/forge/ 79Write a 120 to 160 character description naming the page's subject. |
| One H1 per page | WARN | 1 of 7 pages do not: 0 on /forge/Use one H1 per page, near the top, matching the page's subject. |
| Structured data | WARN | 1 of 7 pages publish no JSON-LDAdd JSON-LD for the page's own type: Article, Product or LocalBusiness. |
| Compressed transfer | WARN | the home page arrived with no content-encoding, so every visitor downloads it wholeno content-encoding header on the home pageEnable Brotli or gzip for HTML, CSS and JavaScript at the edge. |
| Descriptions too short | WARN | 1 of 7 descriptions are under 80 characters and say little/forge/Expand these descriptions past 70 characters with the page's subject. |
| Pages with no H1 | WARN | 1 of 7 pages have no H1, so nothing states what the page is about/forge/Add one H1 to every page, near the top. |
| Text to markup | WARN | 6% of the page is words, so a reader waits for markup8292 characters of text in 129009 bytesReplace image-only or script-heavy markup with the text it stands for. |
| Cache validators | WARN | no ETag or Last-Modified, so a repeat visit re-downloads the pageSend ETag or Last-Modified so caches can revalidate instead of re-download. |
| 29 passed Page titles, Social preview tags, Image alt text, Page language declared, Mobile viewport, Cache policy on the page itself, Mixed content, Duplicate page titles, Pages with no title, Titles too long to show, Pages with no description, Descriptions too long, Pages with more than one H1, Pages with no canonical, Canonicals pointing elsewhere, Pages with no social image, Images with no alt text, Standards mode, Character encoding declared, Twitter card tags, Internal links on the home page, Inline scripts on the home page, URLs are short and readable, Time to first byte, Page weight, Third-party hosts, Images served at the width they render, Smallest image format the site supports, Above the fold eager, below it lazy. 4 not graded Encoding in the response header, Language alternatives declared, Resource hints, Vary on encoding. | ||
| Contentall 1 pass | ||
| 1 passed No path from the build machine appears in served output. | ||
| Discovery and AI readiness3 pass, 3 not graded | ||
| 3 passed XML sitemap, Sitemap named in robots.txt, Favicon. 3 not graded Pages read, Web app manifest, RSS feed. | ||
| Doors a scanner knocks on7 pass, 2 not graded | ||
| 7 passed readme.html, Author names via API, XML-RPC, Login page, Uploads listing, Debug log, .env file. 2 not graded What the site runs, Version disclosed in markup. | ||
| Server runtime0 pass, 1 not graded | ||
| 1 not graded PHP runtime. | ||
no container found in the HTML. A container is optional: this says nothing about the site tracking you, only that this page loads no tag manager.
03What we measured
The pages we read
| Page | Title | H1 | Description | Canonical | Alt text |
|---|---|---|---|---|---|
| / | 18 ch | 1 | 119 ch | self | 100% |
| /case/momsbalm/ | 52 ch | 1 | 108 ch | self | 100% |
| /case/motohorek/ | 60 ch | 1 | 111 ch | self | 100% |
| /case/portfolio/ | 38 ch | 1 | 111 ch | self | no images |
| /case/stack/ | 62 ch | 1 | 163 ch | self | 100% |
| /forge/ | 33 ch | 0 | 79 ch | self | no images |
| /verify/ | 35 ch | 1 | 155 ch | self | no images |
Every check below ran on every page in this table, and each finding names the worst page it found it on.
Speed, measured by Lighthouse
Lighthouse 13.5.0, desktop, run on a real browser
The stress test, printed second because it is a test and not the norm: the same page on Lighthouse's mobile profile (1.47 Mbps down, 150 ms round trip, CPU slowed 4×, Moto G Power (2022)) scores 66, with a 3.3 s largest paint.
Measured from a server in Helsinki, Finland. The headline figures are the desktop run, with no CPU slowdown; the phone profile, printed as the stress test, is throttled to a mid-range phone on a slow mobile connection. Where that server sits is part of the score: visitors closer to it see faster times than visitors further away, so a site with customers on the other side of the world can score lower here than it feels in practice. Every audit is measured from the same place, which is what lets you compare one report against another. One run also moves by a few points on its own, so treat a change as real only when it repeats.
Since the last reading
First reading of this address, so there is nothing to compare against yet. The next reading of it will list what moved.
Lighthouse
- Performance98
- Accessibility96
- Best practices100
- SEO100
- Agentic3/3
- Largest paint0.7 s
- Page weight378 KB
- Phone, stress test66
Accessibility, best practices, SEO and the agent fraction in this panel are the phone run's: this reading was stored before the desktop run kept its own. Performance and paint time are the desktop run's.
The rows that did not pass on the phone profile, worst first, with the number each one measured:
- Minimize main-thread work6.2 s, fails
- Reduce JavaScript execution time1.4 s, fails
- Touch targets do not have sufficient size or spacing.fails
- Forced reflowfails
- Network dependency treefails
- Render-blocking requestsEst savings of 580 ms, fails
- Max Potential First Input Delay410 ms, 16%
- Total Blocking Time1,110 ms, 23%
Lighthouse 13.5.0, desktop, run on a real browser.
The stress test, printed second because it is a test and not the norm: the same page on Lighthouse's mobile profile scores 66, with a 3.3 s largest paint.
Agentic is a fraction, not a score out of 100: this page passed 3 of the 3 agent checks that could run. 2 could not run at all (tool schemas, ai-catalog.json), because the site does not publish them. A missing file is skipped rather than failed, so a clean fraction here does not mean an agent can read or act on this page.
The complete Lighthouse report, all 5 categories and every audit it ran, exactly as the browser produced it.
AI search
Whether a machine that reads pages to answer a question can find the answer here, tell who says it, and quote it without distortion. It measures this page and the files around it; it does not promise a ranking, a citation or traffic in any assistant.
- OK llms.txt maps real pagesThe /llms.txt names pAulse Performance and maps 7 of the site's own pages; 7 checked and answering./ /case/momsbalm/ /case/motohorek/
- WARN An answer block where a searcher lands2 of 4 pages carry an answer block; none on /, /forge/./, /forge/, /case/momsbalm/, /case/motohorek/ Fix Add a 40 to 90 word answer under a question heading on the home page.
- OK Who says it, in JSON-LDpAulse Performance is stated once as organization, with 3 sameAs profiles; 2 resolve.https://github.com/pAulseperformance https://zpe.vc/
- OK Crawler rules for answer enginesThe star group disallows nothing. no answer-engine reader is named; none blocked by name; no Content-Signal line.no Content-Signal line
- OK Every claim of freshness is checkableThe page states 2026-11-19, and the sitemap agrees (2026-11-19).2026-11-19 / lastmod 2026-11-19
4 of 5 checks pass.
The numbers above, drawn. Each one shows how it is on the left and what it should be on the right.
A visitor waits 0.7 s before the main content appears
The page asks for 378 KB and paints its main content after 0.7 s, measured on a desktop.
0 s0.7 s
That is what a new visitor sees: an empty screen, then the page. The score above is this, once.
The email record is missing its last word
The record reads ~all: receivers are told to doubt a forged sender, not refuse it.
as it is today
"v=spf1 include:_spf.mx.cloudflare.net ~all"as it should be
"v=spf1 include:_spf.mx.cloudflare.net ~all"One word decides whether a forged sender is doubted or refused. On a site that mails customers, soft fail is mail that arrives in the junk folder.
One headline, and the crawler knows what the page is
Measured on /: 1 of 7 pages do not: 0 on /forge/ /
what a reader sees
All 1,000 designs
The field
I build and ship products, end to end.
Selected work
Mom's Balm
Motohorek
what the crawler reads
<h2> All 1,000 designs <h2> The field <h1> I build and ship products, end to end. <h2> Selected work <h3> Mom's Balm <h3> Motohorek (1 h1 on this page)
A reader sees a page with a title. A crawler sees this list, and picks one of them as the page's subject.
This report read 7 pages
Everything below is measured on what could be opened from outside.
A sample is a sample. The claim is what was read, never the whole site.
04The next step
This pass read 7 pages. The whole site crawl is free: every page, every internal link, and an inventory you can hand to a developer. It asks for an address, and the report is a link that keeps working after you close the tab.
Put this audit on your own site. Claim the report first: the badge then links back to it, with the date and the counts on it, so a visitor can check the claim instead of trusting an image.
Want the deeper report, the half that needs read-only access to your server? Ask for it.
05Limits and help
Where we could help
- Email: a mailbox on your own domain, aliases for hello, sales and billing, and SPF, DKIM and DMARC set up and tested.
- Security hardening: the headers this audit found missing, set and tested.
- On-page basics: titles, descriptions, headings and alt text, written per page rather than templated.
- A migration: the site as static files on Cloudflare's edge, with the editing workflow kept or replaced entirely.
What this audit cannot see
It ran without server access, so it cannot see plugin versions, the database, backups, logs, or your invoices. Three measurements a Worker cannot make are also absent rather than guessed: TLS certificate details, the HTTP protocol version, and whether the origin compresses. So is the PHP version under your site, unless the server volunteers it in a response header, which none of the twenty-two sites we sampled did. That layer is worth naming because two remote code execution flaws in PHP itself are being exploited in the wild as of 21 October 2026, and no scanner outside your server can see whether you are patched. They are in the deeper report.
The launch gate, read against this site
Our own pre-launch gate: fourteen gates and 110 lines, the list every site of ours is walked through before it serves a stranger. This is not that list repeated as advice. It is the part one reading from outside can settle, with the rest named and the reason it cannot. 16 settled, 8 not met, 24 of the 110 lines checkable this way.
1. it answers, everywhere, correctly 3 of 5 settled
- settled Unknown paths answer a deliberate 404, never a stack trace real 404
- not met One canonical host: apex and www agree www and bare domain agree: www ends at www.paulseperformance.com, bare at paulseperformance.com
- settled HTTPS is the way in yes
- not met One h1, headings in order One H1 per page: 1 of 7 pages do not: 0 on /forge/
- settled Alt text on every image every page with images describes at least 80% of them
- cannot be seen from outside mixed content, what a browser fetches at load, the page with JavaScript off, the layout at 390 and 1280, the redirect map of old addresses, and whether a preview worker still shows a mock these need a browser running the page, or the tree the site was built from
2. the copy is defensible none of it can be settled here
- cannot be seen from outside every line of it: whether a claim traces to a source, whether a promise is regulated, whether the proof is real, and whether the price is shown a reading measures what a page says, not whether it is true; this gate is the owner's own, and the answers are paperwork
3. money and legal none of it can be settled here
- cannot be seen from outside the payment rail in live mode, refunds, terms, privacy, consent, labels, filings and retention each is a decision with a document behind it, and none of it is visible from one request
4. measurement, before the first visitor none of it can be settled here
- cannot be seen from outside whether analytics was proven by a real visit, the conversion event, the UTM convention, Search Console, the alert path, and the number read weekly the audit can find a tracking container; it cannot read the store it writes to
5. discovery, so the site can be found and described 7 of 9 settled
- settled robots.txt answers and points at the sitemap 1 Sitemap line(s)
- settled Sitemap answers, and the pages in it are the site's own routes found at /sitemap.xml
- settled robots.txt does not disallow the pages that matter no site-wide block
- settled A canonical tag on the page self-referencing and absolute on all 7 pages read
- not met Structured data parses and validates Structured data: 1 of 7 pages publish no JSON-LD
- settled A unique title, within 60 characters all 7 pages title themselves in 65 characters or fewer
- not met A unique description, within 155 characters Meta descriptions: 1 of 7 pages are missing a description or sit outside 80 to 200 characters
- settled A share card: og:title, og:description, og:image, twitter:card all 7 pages preview properly
- settled llms.txt at the root The /llms.txt names pAulse Performance and maps 7 of the site's own pages; 7 checked and answering.
- cannot be seen from outside hreflang pairs, whether the share card renders on X, LinkedIn, WhatsApp, Telegram and Slack, the AI crawler policy chosen in robots.txt, whether the money page is two clicks from the root, the backlink list, and a link check over the whole built tree cards and policy choices are decisions, and the render check needs five other companies' surfaces
6. security audit 4 of 6 settled
- not met HSTS, with a year and includeSubDomains HSTS: max-age=86400
- settled A Content Security Policy on the response present, script-src locked, style-src still allows inline
- settled MIME sniffing blocked nosniff
- settled A referrer policy strict-origin-when-cross-origin
- settled Clickjacking protection SAMEORIGIN
- not met A permissions policy Permissions policy: absent
- cannot be seen from outside whether the CSP still allows unsafe-inline somewhere, the dependency audit, the secret scan over the tree and the bundle, the rate limits, server-side validation, what an error response says, PII in logs, the backup and its restore, and who holds the deploy credential each needs the tree, the logs, or the account, and a response header is the only part of this gate that faces outward
7. page and accessibility audit 1 of 1 settled
- settled Speed measured on a throttled phone, not estimated measured, and the score is in this report
- cannot be seen from outside an axe scan with zero serious findings, the keyboard pass, contrast on every pair, the performance budget as a number to fail, images at the size they render, self-hosted fonts, HTML validity, and the page in print one of these needs a headless browser with a network profile, and the rest need a person with a keyboard
8. email 1 of 3 settled
- not met SPF published for the sending domain SPF: one record, but it does not end in -all
- not met DKIM published DKIM: key on resend, 1024-bit (2048 is the current minimum)
- settled DMARC published policy p=reject
- cannot be seen from outside a real message read in Gmail, Outlook and Yahoo, the authentication results from the received headers, every transactional path, the reply path, unsubscribe, bounces, and the list export the DNS records face outward and are read here; the message itself needs an inbox
9. form and signup testing none of it can be settled here
- cannot be seen from outside the happy path with a read-back from the store, bad input, duplicates, spam, the rate limit, the confirmation, the mobile keyboard, and the removal path a form that writes into a hole looks identical to a working one from outside
10. load and stress none of it can be settled here
- cannot be seen from outside the capacity number, the load test at 1x, 5x and 10x, the breaking point, how it fails, and the cost at 10x the audit makes one request at a time, which is the opposite of a load test
11. operations, so a bad day is survivable none of it can be settled here
- cannot be seen from outside uptime and certificate-expiry alerting, the incident runbook, the rollback and its drill, scheduled backups with a tested restore, DNS TTL, log retention, and where the credentials live these are the owner's arrangements, not the site's behaviour
12. launch day none of it can be settled here
- cannot be seen from outside the fifty places worked in order, one message in three channels, who answers replies for 24 hours, and who may pull the launch this one is the launch, and it happens after the gate
Past launch day the gate changes shape: the post-launch runbook runs D+1 to D+90 and every stage ends in one of four decisions, keep, change the copy, change the channel or stop. Its D+1 job is that every row of the product register answers at its own address. That is the same job as the broken pages in the full crawl, one address at a time, and the crawl is the faster way to do it.