Can I try again to get some opinions about the problem as I see it? Let's put aside for the moment the issue of parallax error. If I shoot a series of shots using a wa lens in an interior and stitch them, I get - unavoidably - the characteristic curvature of horizontals above and below the "equator".
No you don't get a wavy horizon, provided(!) that you have your panning axis perfectly aligned. You can still get rid of the 'wave' if you point your horizon (pitch parameter) at a higher or lower position than you are doing when the 'wave' strikes. The resulting stitch will have a somewhat curved top and bottom edge, so you'll lose some FOV to cropping, unless you can fill in the missing parts with CS5s content aware fill (unlikely to be successful for most interiors). However much you try to get your rotation platform leveled, there will almost unavoidably be a small amount of error, but the panostitcher should be able to figure this out automatically if you can set some vertical line controlpoint pairs.
So the combination of a 24 mm TS lens and the Agnos device looked to me like a perfect solution. No curvature, no parallax errors, just an effectively bigger sensor with a wide aspect ratio (assuming landscape orientation). I'm not so concerned about resolution so my main interest is in the effective FOV I could achieve overlapping 3 shots with this lens, by comparison with my existing 14-24 @ 14mm. The absence of the volume anamorphosis effect that 14mm exhibits would also be a big plus.
The Argos device looks like a useful addition, I just don't know how it does in practice (e.g. can you work quickly between lens+camera shifts in opposing directions). In my experience, and within the image quality limitations of a large image circle near the edges, I use the horizontal flat-stitch most. For that it was very easy to equip my RRS setup with a few simple additions to make a very compact solution that also works fast (e.g. when trying to avoid cloud-cover to change light conditions between tiles).
I set my camera with the left shifted lens in the clamp and shift it to the right stop, I then shift the lens to the right and the camera to the left stop. With the camera in landscape mode that's all it takes, 2 exposures and 4 shits, quite fast. The stops are pre-set at the correct width (corresponding with 2x the amount of lens shift + clamp width). The result is very easy to composite in Photoshop with mostly a relative horizontal shift (and perhas a bit vertical to compensate for a rotated sensor). This is impossible
to accomplish in Photoshop under all but the more favorable circumstances without the correction for entrance pupil shifts, especially in scenario's like interiors with furniture close to the camera. A good blender may get away with it under favorable circumstances, but I'm not going to second guess physics and hope Photoshop can manage (because when it can't, you're in big trouble). Prevention is still better than having to cure.
I would have thought that this requirement would be pretty common amongst those of us shooting in confined spaces using 35 mm gear. I confess to being at the low end of the food-chain in this respect; even a single TS lens is a big and possibly uneconomic outlay for me. Of course if it improves my results it may prove justified.
The requirement is there, and so is the solution.