Make port scanner input, skipped results and workload controls honest #15

Confirmed October 10, 2026 review

The scanner already caps host expansion at 256, uses bounded concurrency, and has a first-run authorization modal. Keep those existing protections; adding them is not outstanding work.

Concrete defects found:

• Custom 1-65535 silently scanned only the first 100 ports per range.
• Invalid custom port input silently selected default ports.
• Invalid IPv4 octets were normalized into other addresses (999.0.0.1/30 became 231.0.0.1 and .2).
• Ports skipped after discovery failed were recorded as filtered with latencyMs: 0, despite never being tested.

The current implementation work addresses these correctness defects with explicit input errors, a visible port cap, strict address validation, and nullable timing/not-tested results.

Remaining follow-up

• Add cancellation/stop that terminates pending host discovery and service probes.
• Surface aggregate host × port workload before large scans and choose an appropriate aggregate cap.
• Preserve the timing-inference limitation in every result/export; browser probes cannot prove a TCP port is open or closed.
• Add browser regressions for large/invalid input, discovery skips, and interrupted scans.