Excellent questions, I'll field them as best I can, if there's clarification needed just ask.
- Voice data passes through the POTS lines to our VoIP cluster, eventually getting routed to another call channel or to a recording. Freeswitch is the node of choice here, though we're doing rather simple things with it so eventually we'll probably replace this with something more homegrown. Every vendor along those legs can listen in, just like normal POTS lines. Until people get off the POTS system, this is just going to be reality. Plans and plumbing are in place when viable to make this a strictly end-to-end pass-through fully encrypted service if you're not going over POTS. VoIP is pretty awesome stuff.
- Any vendor servicing the lines can listen in. Unless the NSA has stopped, they're definitely processing some of it. If this is a concern, you really just don't want a phone number at all. I'd love to offer better privacy there, but it isn't in anyone's hands but users' really.
- No SLA for the consumer product yet. We have a track record of no downtime outside our maintenance, but we've only been doing this for a few months so that's easy. That said, everything's modular and I can replace any given piece with a new virtual server in under 10 minutes from start to finish.
- Redundancy in the web portion is there, redundancy in the database is there, redundancy in the VoIP level is sort of there, but if any node along your call chain goes down you're going to lose that call and have to start it again. This is yet another failing of voice communications in general, and not something we can do anything about.
- POTS quality is the best we (or anyone) can do. No compression, just straight uLaw without processing so it's as clear as can be expected from a phone. You can actually see this phenomenon displayed from any cell phone currenty. Call another cell on your same network, you get pretty good quality comparatively. Call another cell on another network and depending on peering you could degrade quite a bit. Call a line connected to any copper telco and you're back at the lowest common denominator. Someday if we can get people off telephone numbers (a haughty goal, unfortunately) we can do quality limited only by your hardware on each end.
- Does the voice data pass through Phone Janitor, or is this strictly at the SS7 control level?
- Who's listening in? How do we know?
- What's the service level guarantee?
- Do you have telco levels of redundancy, or is this some 1U server running Asterisk?
- Will this reduce voice quality to low-end VoIP levels? How much latency do you add? How much compression?