Policy Based Routing

I'm considering making a couple of changes to my networks that I think will require some form of PBR and I'm trying to understand (a) if that's right, and if so, (b) what is the appropriate way to do that. It seems that there are two approaches to PBR, using either mwan3 or pbr, and I'm struggling to understand what are the scenarios that each are designed for.

Some background: I have a two site setup, connected by a wireguard site-to-site VPN. Each site has its own internet connection; site1 has FTTC/VDSL with a static IP address, and site2 has 4G cellular, dynamic IP address and CGNAT.

Firstly, I'm adding a VoIP phone to each location. The phone in site1 works perfectly, but the in-bound audio stream to the phone at site2 is being blocked (I believe) by the CGNAT. To try to resolve this I would like to route that phones packets across the site-to-site VPN, and out of site1's internet connection with its static IP address, I believe this will require PBR, but only on the site2 router.

Secondly, I wish to change ISP for site1 as I can now move to a symmetrical 1Gbps FTTP connection. However, that ISP will not supply a cost-effective static IP address. To resolve this I'm hoping to also use an "overlay" ISP via an L2TP connection for static IPv4 and IPv6 addresses, but this will be limited in bandwidth and capped with a data limit, so I will want to restrict its use to only those devices that require a static IP (so, my servers, the site-to-site VPN and the VoIP phones etc). So I think I will also need PBR on the router for site1.

I don't think I need or want features like wan failover or traffic balancing. This will just be about routing specific devices via specific links.

Any advice (or education!) much appreciated.

If you are in charge of site to site VPN it might be a good idea to explore mptcp, new in upcoming v24 ?

Thanks for the suggestion, but from a quick look at the RFC, I'm not sure how that will help with what I'm trying to achieve.

If you need special rules for individual devices (IP) I think the manual approach fits better.