Event 5156 is written by the Filtering Platform Connection subcategory whenever the Windows Filtering Platform allows a connection. Its mirror, event 5157, is written when the platform blocks one. The subcategory also covers port bindings and listen actions, but these two are the ones analysts read. Microsoft's own table for the subcategory shows General Success auditing as No for domain controllers, member servers and workstations, and Failure as Yes. The reasoning is in the comment: Success typically generates one event for every connection made to the system, so it is far more useful to audit blocked connections.
That explains the forum threads. A Splunk Community thread reports about 25,000 of these events per minute from servers, and a Spiceworks poster describes about fifty a second. The usual cause is a policy that turned Success auditing on for this subcategory; the urtech.ca write-up traced it to Audit Object Access in the Default Domain Controller Policy. The fix is the one Microsoft implies: set the subcategory back to Failure only, for example with auditpol /set /subcategory:"Filtering Platform Connection" /success:disable, and turn Success on only for the high-value hosts where you need to see connections to and from untrusted addresses. If you must keep it everywhere, filter at the forwarder rather than discarding whole ranges blindly.
Reading the event takes patience because the fields behave differently from other Security events. Application is a kernel-style path such as \device\harddiskvolume3\windows\system32\svchost.exe, not C:\Windows\System32\svchost.exe. Microsoft points to diskpart's list volume command to map the number to a drive. Any rule written for 4688 or 4663 paths will miss it, so your regex needs a volume wildcard. ProcessID is documented as hexadecimal, but the sample event shows a plain decimal value, and that matches what a SIEM extracts. Compare it with Task Manager directly, and convert it before matching the hexadecimal new process ID in a 4688.
Protocol is the IANA number: 6 for TCP, 17 for UDP and 1 for ICMP, with 47 (GRE), 50 (ESP) and 51 (AH) among those Microsoft lists. FilterRTID is the run-time ID of the filter that made the decision, and netsh wfp show filters writes a filters.xml where you can look it up. LayerName and LayerRTID identify the Application Layer Enforcement layer, resolvable with netsh wfp show state. In the sample, %%14610 is the message code Event Viewer renders as Receive/Accept, and the layer ID is 44. RemoteUserID and RemoteMachineID both show S-1-0-0 in Microsoft's sample.
Test this before you build detections. Microsoft describes SourceAddress in 5156 as the address that initiated the connection and DestAddress as where it was received. The page for 5157 flips that, describing SourceAddress as the local address, and the sample events put the listening process on the source side. The documentation contradicts itself on which end is which. Do not argue it out on paper. Start a simple listener on a lab host with auditing on, connect to it from a second machine, and see which field holds which address. The sample on this page follows the initiator convention: 10.20.4.88 reaching port 3389 on 10.20.4.37, an inbound RDP session between two workstations, which is a lateral movement path worth a rule.
Once the event is flowing, put it to work on three jobs. The first is an egress inventory by process: which executables talk to which external addresses, and from which folders. Microsoft suggests flagging an Application that is not the expected one, that sits outside standard folders, or whose name contains strings such as mimikatz. The second is exposure of administrative ports. Inbound connections to 445, 3389, 5985, 5986 and 135 from outside your management subnets map to remote services (T1021) and deserve review. The third is boundary hygiene: protocol numbers other than 1, 6 and 17 on an endpoint are unusual, and Microsoft calls them out.
Event 5157 is the blocked half. The fields are identical, FilterRTID names the filter that dropped the traffic, and it is logged as an audit failure, which is why you probably collect it already. It is the cleanest place to see scanning. One remote address hitting many different ports on a host and being refused is the shape of network service discovery (T1046), and blocked inbound attempts on 445 or 3389 from a workstation subnet show lateral movement that the firewall stopped. A blocked outbound connection from a server that should never reach the internet tells you something tried to call home.