Known limitations
- Unsigned builds. Windows builds are not code-signed until signing with the maintainer's Certum certificate is set up, and macOS builds are not notarised, so both systems warn on first start.
- Platforms. Windows x86_64, macOS 12+ (Apple silicon and Intel) and Linux x86_64. No Windows on ARM or Linux ARM builds yet.
- IP database placements are hints. The fallback database places a backbone address where its operator registered the prefix, which can be a head office far from the router. Route Map rejects placements the round trip rules out, but a wrong city that is still within reach is shown, labelled
ip-db. Some addresses are known only to a country (hollow marker). - The physics bound is an upper limit. It proves a hop cannot be farther than about 100 km per millisecond of round trip; it cannot prove a hop is where a source says. Slow replies (routers answer ICMP at low priority) only loosen the bound, never tighten it.
- What traceroute cannot see. MPLS tunnels hide hops, the return path can differ from the forward path, and hops that do not answer are listed but not placed.
- Origin from your public IP is a city-level guess and is wrong behind a VPN or proxy; set the origin in Settings.
- Behind a VPN or SASE client a trace shows the path through the tunnel. A driver-level client can ignore the probe interface; hop 1 is checked, but a consistent hop 1 is not proof of a direct path. The TCP interception check runs on Paths over TCP (Windows); it cannot see a proxy further than two hops away. An adapter the system does not clearly describe stays "unclassified".
- Repeated hops at one place. Hops at the same place that are not next to each other in the trace get separate markers; at world zoom the labels can spread far apart, with thin leader lines back to the real point. Zoom in to see them together.
- Hostname rules cover the operators Hoiho and the bundled site-code table know. A carrier with an unknown naming scheme falls back to the database; reports of such schemes are welcome.
- No submarine cables. TeleGeography's cable data is sold under licence (only its map images are CC BY-SA), so there is no cable overlay until a source the app may ship exists.
- No internet exchange labels. PeeringDB's terms do not allow bundling its prefix list; the code is there, switched off, pending their permission.
- ASN data is DB-IP's ranges, not BGP prefixes. The offline AS number is from DB-IP Lite ASN; the routed prefix and RPKI come from RIPEstat when Online lookups are on.
- Anycast is a likelihood. "Likely anycast" means the destination answered too fast for where it is registered, or belongs to a large content network; the served-from city is the largest one reachable in the round trip the destination added past the last placed hop before it.
- The typical latency is between countries, from the nearest RIPE Atlas anchors to the origin and the destination; small countries with no anchor have no baseline.
Continuous mode
- IPv4 or IPv6, as Settings > Trace > IP version chooses.
- The built-in ICMP prober only; the system
tracerouteandmtrare not used for it, and neither is RIPE Atlas. - One probe per hop per cycle, so ECMP siblings show up over several cycles, not in one.
- Sleep is told apart from a change of the system clock by a clock that keeps counting through sleep (CLOCK_BOOTTIME on Linux, CLOCK_MONOTONIC on macOS, the monotonic clock on Windows). If that clock cannot be read, the session says "Clock check off" with the reason, and a clock change can show as a gap. Either way nothing is counted as loss.
Paths
- A lower bound. Paths probes with ICMP on macOS and Linux and with TCP connection attempts to port 443 on Windows (Windows 10 version 2004 or later). Some load balancers spread one kind of probe and not the other, so a path can stay hidden either way.
- Networks that do not answer TCP probes. On Windows, routers past some hop may answer ICMP but not TCP probes. Paths cannot map past that hop; it says so in the heading ("Routers on this network do not answer TCP probes beyond hop N"), and Trace (ICMP) can map on. When the ICMP check is answered by the target itself, the heading says a firewall or load balancer in front of the target is the likely reason instead.
- Silent hops. Discovery stops after 5 silent hops, then asks the target once more at full TTL on each path found; a path reached only that way shows the silent stretch, and the target is numbered after the last hop sent.
- Rate-limited routers answer only some probes; Paths does not add flows there once more flows stop showing anything new, so a balancer behind a heavily rate-limited hop can be missed now and then.
- Per-packet balancing is named, not split into paths: every packet may take another router there.
- Discovery is not continuous: Live probes one flow per hop per cycle.
Reverse traces
- They need RIPE Atlas with your own key and credits (60 each), and Online lookups (to learn your public IP and the target's network).
- A probe in or near the target's network must exist; for some networks and small countries there is none.
- Your public IP address is published as the target (see the guide). Behind carrier NAT the reverse trace ends at your ISP's NAT address, not your machine; on IPv6 with temporary addresses it traces to the address you had at that moment.