Evidence collection
Radar accepts only HTTP and HTTPS destinations, including explicit ports from 1 to 65535. It rejects credentials, hidden control characters, ambiguous backslashes, malformed authorities, private-network destinations and unsafe DNS answers before allocating scan capacity. Every redirect is revalidated, repeated hops stop immediately, DNS and response work have deadlines, and the full static retrieval path has one bounded wall-clock budget. URL text is parsed as structured data and is never interpolated into a shell command. Requests are also bounded by bytes, redirect count, queue size and concurrency. Static analysis does not execute page scripts.
Analysis layers
Independent checks cover URL and brand context, reputation, domain and hosting context, DNS, TLS, redirects, response content, forms, scripts, clipboard behaviour, documents and file-delivery indicators. Selected risks may also use a separate credential-free browser worker.
Scoring and confidence
Every scoring finding is labelled as candidate, corroborated or confirmed and assigned to an evidence category. A single weak evidence category cannot produce a High verdict; independent categories or directly observed high-confidence behaviour are required. Findings from selected same-site pages remain page-scoped, and their points are not added together across unrelated pages. Established Microsoft, Google and Apple authentication endpoints are recorded as legitimate delegation context rather than unknown credential collectors. Scripted credential collection requires a credential form plus a locally correlated collector endpoint, outbound send, submission hook and credential-field read; endpoint mentions alone do not score. Direct ClickFix instructions, command-like clipboard delivery, unsafe credential destinations and reputation-confirmed threats are never suppressed by these controls.
Defensive indicator extraction
Radar extracts bounded URLs, hostnames, IP addresses, SHA-256 values, active filenames and blockchain addresses from inert content for analyst review. Raw commands and payload content are not stored; URL fragments, query values, bot tokens and webhook identifiers are redacted before retention.
Browser-analysis boundary
The isolated worker uses a pinned public-network request broker, blocks private destinations and service workers, limits requests and bytes, and never runs operating-system commands or downloaded files. Eligible signals may receive rendered comparison, bounded observation at multiple checkpoints and one constrained verification transition. Radar may submit one same-origin verification form only when it contains no editable, credential or file fields; it never enters values or submits the credential form revealed afterwards. It records only bounded counts and boolean signals when a self-hosted gate silently measures browser or interaction characteristics; field names, values and raw scoring code are discarded. A verification transition can be correlated with a branded password prompt on the same host or another host, while unbranded same-site logins remain negative controls. Only after ClickFix instructions appear may Radar perform one copy-only interaction. Radar can recognise multilingual security-intent wording, intercept clipboard writes for in-memory classification, decompose supported command workflows into safe defensive properties, and record privacy-bounded resource fingerprints. When evidence images qualify for retention, Radar stores only the strongest event image rather than every near-duplicate candidate. It never pastes or executes commands, connects to clipboard-supplied destinations, opens downloaded files or submits forms containing visitor-entered data.
Privacy and retention
New free scans create public reports available for 30 days and listed on the dashboard for 24 hours. Public report views reduce HTTP addresses to their website origin so paths, query values, fragments and embedded credentials are withheld. The submitting browser can display its own exact entered address from session-only storage. Reports remain unindexed. After public expiry, URL Radar removes the public lookup index and keeps the report and policy-approved evidence in a private owner-only archive for defensive research, regression testing and training. Credentials, authentication tokens, executed commands and downloaded payloads are not retained.
Known limitations
- Targeted, geo-specific or authenticated content may not be visible.
- Access challenges can prevent inspection of the destination.
- New threats may have no reputation history.
- A point-in-time result can become stale as a website changes.
- Low risk is not proof of safety; high heuristic risk is not proof of malicious ownership.