What the availability score is
The availability score is the percentage of our checks that succeeded over the last 30 days. We check the panel's website and, where we hold read-only API access, its API.
It does not tell you whether orders are delivered, whether followers are real or stay, whether refill works, or whether refunds are paid. A panel can answer every check and still deliver poorly, and the reverse. Price, number of services, verified ownership, advertising and how often a catalog changes never raise it.
The four things on a profile are separate: availability, catalog activity, reviews from this site's members, and Trustpilot's own figures.
The checks
- Website: the home page is fetched about every 60 minutes.
- API: the read-only service list is requested about every 6 hours, only for panels whose owner gave us a key. The key is never shown or logged.
- Each scheduled time slot is one check. If a slot fails we retry up to 3 times; the retries are attempts inside the same slot, never extra votes.
A check is a success only if
The page is a real page on the panel's own site (not an error page, parked or for-sale page, default server page, empty document or a redirect to an unrelated domain), or the API returns a usable service list. HTTP 200 alone is not success.
A check is a confirmed failure only if
The failure is the panel's (timeout, refused connection, unresolvable domain, broken certificate, server error, unusable response), it repeated on every attempt in the slot, and unrelated sites were reachable from our server at the same time.
A check is inconclusive when
The cause could be ours or unclear: a rejected, expired or missing API key, rate limiting, bot protection, a robots.txt rule, our own network or DNS being down, an error we cannot attribute, or a wave of failures across many panels at once (treated as our problem). Inconclusive and not-run checks are never counted as success or failure, but they lower coverage.
The formula
Success rate = 100 × successful checks ÷ (successful checks + confirmed failures)
Coverage = checks with a definitive result ÷ checks expected in the period
Availability score = 0.7 × API success rate + 0.3 × website success rate
The weights (70% API, 30% website) are a product decision, not a scientific standard. They are stored with a version number. If the API cannot be checked, its weight is not given to the website: no combined score is shown, and the website result is displayed on its own.
When a score is shown
- Under 14 days of monitoring: no score. The page says data is being collected. It never shows 0.
- 14 to 30 days: a provisional score, only if both channels have at least 90% coverage of the time we have actually watched the panel.
- 30 days or more: the full 30-day score, again only with at least 90% coverage in both channels.
- Not enough coverage: “Insufficient data”.
Next to every score the page shows the period, number of checks, coverage per channel, the last successful check, and whether it is provisional.
Current status is separate
The badge at the top (“Latest checks passed”, “Disruption detected”, “Status unknown”) comes from the newest definitive checks, independent of the 30-day score. A high historical score cannot hide a current outage. Status turns unknown when the newest definitive check is older than 3 check intervals.
Catalog activity
Activity describes what changed between catalog fetches. It is not a quality score and earns no ranking.
- The first catalog we read for a panel is a baseline. Its services are never counted as “added”.
- A service is “disappeared from catalog” only after it is missing from 2 complete fetches in a row. We say disappeared, not deactivated, because a catalog tells us nothing about status.
- A catalog that looks cut short (below 70% of the known size) is held back and changes nothing until the same content is seen again.
- When almost every price moves by the same factor, that is a currency-rate change and is kept out of the price counts.
- Emoji, spacing and punctuation edits to a name are not changes.
- Times are when we observed a change, not when the panel made it.
- “No change observed” does not mean inactive. Days without a complete fetch are shown as “no data”.
Refill, cancel and drip-feed shown on a profile are capabilities declared in the catalog. They do not prove the feature works.
User reviews and the adjusted rating
Only reviews from this site's members count. One user is one voice per panel: several reviews by the same person are stored but averaged into a single rating before anything is counted. Optional ratings marked “did not use” are left out, never counted as zero.
The star display is the plain average with the number of users, the star distribution and the date of the latest review. Fewer than 5 independent users is a limited sample and the panel is left out of “top rated” lists.
For ordering only, an adjusted value is used: (n × R + m × C) ÷ (n + m), with n independent users, R their average, m = 10 and C a baseline: 3 until the site has 50 eligible reviews from 20 users, then the mean over users of each user's average. The adjusted value is never shown as the panel's rating. This is not an anti-fraud measure: duplicates, bursts and links go to a manual queue separately.
Negative reviews are not removed for being negative. A verified owner may reply; a reply does not mark a complaint resolved. “Order proof checked” appears only after a person has looked at private order details, which are never public and are deleted after review.
Trustpilot
Trustpilot figures appear only through Trustpilot's official data API, cached and refreshed at least every 24 hours, and only when the site's licence allows showing other businesses' data. Without that, only a verified profile link is shown. Out-of-date numbers are hidden and the link stays. These numbers never enter the site's own ratings.
Limits
- Periodic checks are samples, not continuous monitoring. The score is “percent of checks that succeeded”, not exact uptime.
- A panel behind strong bot protection may be inconclusive often and show insufficient data.
- Our server being unable to reach the internet cannot lower a panel's score, but it does reduce coverage.
- Reviews can be gamed; manual checks reduce this and do not remove it.
Settings and versions
Formula availability-v1, settings version #1, effective 2026-11-03. Every stored score names the version it used. Changing a weight or threshold creates a new numbered version; earlier versions are kept.