Methodology
Evidence first. Limits stated.
SiteSentry reports what it can observe in the responses it fetches. When it can't observe something, it says so instead of guessing, and it never calls an issue fixed without re-checking it.
Coverage
49 deterministic checks in 9 categories
The same input always produces the same findings. You choose which categories an audit runs, and the audit records which pages and checks it actually covered.
Crawlability & HTTP10 checks
Status codes, redirects, robots.txt and sitemaps
Indexability9 checks
Robots directives, canonical tags and intended indexing
Metadata & structure14 checks
Titles, meta descriptions and headings
Internal links3 checks
Broken and redirected internal links, orphan candidates
Structured data2 checks
JSON-LD syntax and basic required properties
Images4 checks
Alt attributes, dimensions and broken internal images
Content (heuristic)3 checks
Word counts and duplicate titles/descriptions
Language & hreflang3 checks
html lang and hreflang annotations
Social metadata1 checks
Open Graph and Twitter card tags
Confidence
Every finding says how sure it is
- Verified
- Measured directly from the response: a missing tag, a 404, a noindex directive. If it's reported, it's there.
- Heuristic
- Based on a threshold or convention, such as a title over 60 characters or fewer than 250 words. Useful, but a judgement call; thresholds are configurable.
- Needs review
- Measured correctly but possibly intentional, such as an image with empty alt text. Only a person can decide.
Priority
A score you can read, not a black box
Each finding shows the reasons behind its priority, P1 (most urgent) to P4. The score adds up as follows.
- Base severity
- High +3, medium +2, low +1. Workspace settings can override the severity of any check.
- Important URLs
- +1 if the issue affects a URL you marked as important, +0.5 if it affects the homepage.
- Scope
- +0.5 when it spans 2–4 URLs, +1 for 5 or more.
- Certainty
- −1 for heuristic and needs-review findings, so a guess never outranks a verified defect of the same severity.
- Kind
- −0.5 for improvement opportunities and observations, which aren't defects.
Example: images without width and height are a low-severity layout-stability improvement (they can cause layout shift), so they usually rank P4, not as a major SEO defect.
Health score
One number for triage, with the arithmetic shown
A SiteSentry triage indicator, not a search-engine metric. Issues are counted per check, not per page, so one template bug is one deduction.
100 − 10 × high − 3 × medium − 1 × low
Counted per open issue type (one check failing on one site) at its worst severity, never below 0. 80 and above reads as good, 60–79 needs work, below 60 is poor.
Example: one high, five medium and seven low issue types give 100 − 10 − 15 − 7 = 68. Fixing the high-severity issue adds 10 points; fixing one image template that affects 25 pages adds 1, because it is one fix.
The score says how much outstanding work an audit found. It is not a ranking factor, and it can't predict traffic.
History
When is an issue resolved?
Each issue has a stable fingerprint: the check, the normalised page URL and any expectation that shaped it. Wording and observed values are excluded, so an issue keeps its identity when its details change. Audit results are stored as snapshots that later audits never rewrite.
An issue is resolved only when a later audit inspected the same page with the same check and didn't find it. If that audit skipped the page, timed out, or didn't run that check, the issue is not re-checked. A regression is an issue found where the previous audit ran the same check and found nothing; issues outside the previous audit's scope are reported as first detections instead.
Limits
What it can and can't see
What it observes
- The HTML and headers each page returns
- robots.txt, XML sitemaps and redirect chains
- The status of internal link and image targets
What it marks as unknown
- Pages that timed out, failed or were blocked
- Pages over the size limit (truncated)
- Issues the latest audit didn't re-examine
What it doesn't claim
- Content added by JavaScript after load
- Rankings, traffic or Core Web Vitals field data
- That a change caused a ranking movement
| Profile | Limits | Rendering | Use |
|---|---|---|---|
| Quick | 25 URLs · 3 clicks deep | Off | Pre-sales check, single section, public hero form |
| Standard | 1,000 URLs · 10 clicks deep | Sampled (1 page per template) | Monthly client audit |
| Full-site | 50,000 URLs · unlimited depth by default; budget set per audit | Sampled (1 page per template) or Full (every page) | Migrations, large e-commerce |
| Custom | 1,000 URLs · 10 clicks deep by default; budget and depth set per audit | Off or Sampled (1 page per template) or Full (every page) | Include/exclude regex, subfolder-only, staging with auth |
- Every crawl needs an account and a domain verified as yours. Free accounts run Quick audits.
- Up to 60 link targets are status-checked.
- 10s timeout per request, 2 MB of HTML per page, at most 5 redirects, each one re-validated.
- Only public http/https addresses. Private, loopback, link-local and cloud-metadata IPs are refused, including at connect time.
- robots.txt is respected. Character encoding comes from the Content-Type header only.
- Rendering loads pages in headless Chromium after the crawl (412 px wide, no CPU or network throttling, ads and analytics blocked, 20 s per page). Its LCP and CLS are lab numbers from one load, never field data. Links that only appear after JavaScript are reported, not crawled.
- No search-performance data yet. Page HTML alone can't show that a change moved rankings; that would need Search Console data.
Try it on a page you know well.
The fastest way to judge an audit tool is to point it at a site whose problems you already know.