Complete
The bounded checks returned usable evidence. This does not verify a business, its products or fulfilment.
The evidence behind the result
Understand the check.
Know its limits.
A useful result tells you what was observed, what is missing and when it expires. ShopSafe has three different check suites. Their scores and scope are not interchangeable.
01 · For agents and applications
Quick URL verification.
Public DNS, HTTP and TLS, up to three HTTP redirects, static sensitive-form checks and Google Web Risk Lookup. This suite does not run a browser, JavaScript or a language model.
Rules quick-2026-10-19-v4. The same recorded observations and evaluation time produce the same result.
Read the machine-readable rules or use the MCP resource shopsafe://verification/rules. Reading the rules does not start a check or spend credit.
Risk points · 0–100
The strongest supported signal wins.
Use the highest triggered value below. Points are not added together, and a score is not a probability.
- Current URL reputation match95
- Google Web Risk matches the exact submitted URL or a responding HTTP redirect. Evidence must be current and scoped to that URL.
- Sensitive form exposed to HTTP70
- An associated password or card field is present on an HTTP page or has a supported submission path to HTTP in static HTML.
- Redirect from HTTPS to HTTP30
- A responding HTTPS page directs the check to an HTTP URL within the supported redirect chain.
Ordinary hosting, missing email records, certificate age, internationalised names, third-party payment providers and brand popularity add no quick-check points.
Partial
Some evidence is unavailable, stale or outside an allowance. A zero score with missing evidence stays unknown.
Incomplete
The page could not be observed and the submitted URL has no current usable reputation answer.
How to read the result
Check the exact URL, age and coverage before applying your own policy. A current match can still show high risk when page inspection is blocked.
Evidence expires after at most five minutes, or sooner if its source expires. Cached results retain their original observation time. An expired record remains history.
Usable observations and no reputation match.
0/100 · no_known_threats · complete coverage.
Page observed; reputation provider unavailable.
0/100 · unknown · partial coverage.
A responding intermediate redirect is listed; the final page is not.
95/100 · high_risk · complete coverage.
Submitted URL is listed but the page blocks inspection.
95/100 · high_risk · partial coverage.
A sensitive form uses HTTP and the chain downgrades from HTTPS. Points are not added together.
70/100 · elevated_risk · complete coverage.
Synthetic examples illustrate the rules. They are not real website assessments or a measured accuracy benchmark.
What static form inspection can and cannot establish
The parser associates supported password and card fields with their forms, including explicit form IDs and supported submit-button destinations. Text-only markup, inert template contents and non-submitting button actions are excluded. Supported card autofill fields include inputs, text areas and selectors. A third-party HTTPS payment form adds no risk by itself.
The limits are 100 forms, 2000 controls and 30000 HTML tags, within the bounded HTTP response. Exceeding these limits makes form coverage unavailable. Conditional forms, ambiguous table or foreign-namespace markup, certain script states and XML/XHTML also return incomplete evidence. The result identifies a parsing limit or unsupported structure while retaining usable HTTP and TLS observations. Embedded documents, JavaScript changes, custom widgets, runtime submission and the identity of a payment recipient are not verified.
Google Web Risk provides URL reputation evidence. Some unsafe sites may be missed and some safe sites may be flagged in error. Neither this evidence nor ShopSafe guarantees the safety of a site or transaction.
02 · For verified brand owners
Compare the brand.
Preserve the evidence.
Protect compares a candidate with the verified official homepage and up to two owner-selected official pages. Each page is assessed separately; points are not pooled across references.
Rules protect-2026-10-19-v9. Its priority score helps order review work. It is not the quick URL risk score or a finding that a business has committed fraud.
- Lookalikes are leads
- Bounded name variants, certificate history and owner reports supply candidates. A similar hostname or brand mention alone does not open a takedown case.
- Corroborating copied content
- Brand identity is compared with seven-word text fingerprints, exact official image addresses, retained image bytes and conservative pixel samples of static PNG, JPEG and WebP images. Pixel similarity is a weaker supporting signal than identical bytes; image evidence is counted once. Brand identity plus copied text needs a page action or brand-image match. An image match without copied text also needs a lookalike name and page action. The evidence goes to human review.
- Coverage stays visible
- Missing or stale official pages, blocked sites and uncollected images remain gaps. Pixel checks may identify resized or re-encoded rasters, but crops, recolours, animation and unsupported or undecodable images remain outside comparison. A missing match cannot establish originality. This is bounded discovery, not a search of every website or advert.
- Casework requires authority
- A current reviewed finding and customer authority precede a report. Staff approve and record actual provider submissions and follow-ups. Active cases are eligible for six-hour follow-ups within the capture allowance. Quieter watchlist pages have longer intervals, with first checks and revisits rotating across saved observations. A timeout is not proof of removal; closure needs separated corroborating observations and human review.
03 · The public website checker
A broader storefront review.
The public checker uses rendered page content, domain and infrastructure observations, available reputation sources and a bounded AI assessment. Its heuristic scoring differs from deterministic agent verification.
Result version 2026-10-18-url-scope-v2. The result lists the checks that ran, their findings and source coverage. A familiar brand does not bypass an exact URL threat match.
- Combined signals
- Deterministic infrastructure and content contributions are capped at 75 points. Half of the AI assessment weight is added, then observed trust indicators subtract at most 12 points. The result stays between 0 and 100.
- Reputation and missing evidence
- A current authoritative match for the checked URL sets a minimum of 85. If reputation sources are all unavailable, or the current rendered page cannot be observed, the result is at least 30 and coverage discloses the gap. That uncertainty floor is specific to this browser suite.
- Observed signals have limits
- A payment SDK, policy link, business identifier or familiar design does not by itself verify a merchant. Page claims and AI analysis can be wrong. A low score does not certify safety; a high score is a reason to inspect the evidence.
Review and correction
Evidence can change.
Operate a site with a result you think is wrong? Supply the URL and the finding you dispute so it can be reviewed.