Mist shows the site as up without issue, and our pings are showing zero packet loss for this site on either the ISP gateway or the usable IP. However, our monitoring software Zabbix shows consistent 100% packet loss spikes on the main circuit.

The behavior is expected. 

Zabbix monitoring is just an API call (https://api.mistsys.com/api/v1/orgs/org_id/stats/ports/search?type=gateway&port_usage=wan) to the Mist portal.

The latency/jitter/loss returned by the the org level ports/search API are from the underlay WAN probes (i.e. ICMP pings to 8.8.8.8/1.1.1.1 via the local breakout path on each WAN interface), while the latency/jitter/loss shown on the Peer Path Stats graphs on Insights page and under Topology Details on Device page are measured via BFD/FPM between peer SSR routers over the SVR. The discrepancies are as expected as the latency/jitter/loss from pinging 8.8.8.8 via local breakout could be different from the latency/jitter/loss measured between two SSR routers over the SVR. We return the underlay WAN probe results in the ports/search API because SVR is not supported on all ports and we want to provide apple-to-apple comparison across all WAN links within the org.

Currently the only place to see the underlay WAN probes results on Mist UI is to hover the cursor over a path on Application Path Insights graph

There is already an outstanding feature request to add a time-series graph for underlay WAN probes results, which will be addressed in future updates.