All other factors being equal, does TTL ignore the ambient exposure and add a fixed amount of light, or does TTL take the ambient exposure into account so that the total exposure is the constant? This question is relevant when thinking about balancing ambient and flash when they overlap. I took some test shots and found that for my equipment combination the answer appears to be somewhere in between, though closer to the latter (i.e. constant overall exposure). YMMV.
Test protocol: the test target was a plain white cardboard. I took control shots using ambient only light (which was from yellowish fluorescent light) in manual exposure (in matrix metering), with the light meter at 0, -1, -2, -3, plus one shot at sync speed with practically zero ambient. I took similar shots with TTL flash and TTL-BL flash. Equipment used: D300 and SB-800.
First column: ambient-only
Second column: ambient + TTL flash (0 FEC)
Third column: ambient + TTL-BL flash (0 FEC)
The top row is with the light meter at 0, and the next rows are with the light meter at -1, -2, -3 respectively. The bottom row is with ambient "deleted."
For the ambient-only column, the changes in exposure are what I expected. The second (TTL) and third column (TTL-BL) seem to show a more-or-less constant exposure. If the amount of flash that is added to the image is constant, I would have expected the exposure to vary about as much as the ambient-only exposure. The fact that there is seemingly no variation implies that TTL does take ambient exposure into account to keep the final overall exposure (the sum of the ambient exposure and flash exposure) close to constant. (Note: in the actual shots, the histograms were a bit different, and crept upwards with increasing ambient light.)
In the top and second rows, the colors are a bit different because a greater percentage of the overall exposure comes from the ambient exposure (which is yellowish in color). I tried using a simple black and white conversion in Corel PSP X3 to see how close the luminance values are, and here's what I got:
The numbers in the boxes represent the 8-bit RGB value in the middle of the box (I don't know how to display the luminance value on a pixel level but I take it that the value shown here is a decent proxy for luminance). Note the noticeably higher value in the top row for the second and third columns. This reinforces the concept that the ambient must be at least slightly underexposed relative to our desired exposure (the flash can't take away light after all - it can only add light).
Application: if indeed TTL makes overall ambient exposure constant, then balancing flash and ambient when they overlap becomes much easier than otherwise, because we would be free to adjust the ambient to the desired exposure, leaving TTL to do the heavy lifting (in calculation terms) to reach a 'correct' overall (ambient + flash) exposure. The only limit appears to be that the ambient must be at least slightly underexposed if we don't want to overexpose it when we add flash. My results are limited however to my equipment. You can test if your equipment combination also behaves similarly by taking test shots with the protocol above.