Incident reading of the dress...
I believe you mean "reflected reading" and not "incident reading". Its not possible to take an incident reading
of a subject or portion of a subject.
Back to the OP's question:
There is no single answer, hence your not finding one. The problem is that the dynamic range of different films and different digital sensors vary. In the case of digital, a single sensor will deliver a difference dynamic range at differing ISO settings. This means that the difference, in stops, between that middle or average tone in the picture and the point where highlights begin to clip varies.
There are two approaches:
1. trial-and-terror: shoot a test shot and review the histogram to judge whether significant highlights are clipping. If so, take another test at a somewhat reduced exposure and check the histogram again. Eventually you will find the right exposure. With experience, you'll find a nice average amount of exposure shift that your camera, with either your chosen film or your preferred digital ISO, requires. You'll be able to use that without any further testing most of the time.
2. Calibrate your camera's sensor and meter: This derives from the old B&W film "religion" known as the Zone System. You first need to disable all the metering brains in your camera, generally by selecting a spot metering mode. Meter from the dress only. Now take a series of pictures starting with this metered exposure and reducing the exposure in each subsequent shot. Now find the best picture, in terms of reproducing the dress, and determine how many stops different the exposure was compared to the spot reading. You can now use this difference in future shots, so long as you use the same film or the same digital camera at the same ISO, by spot metering a dress in manual mode and shifting the exposure by the pre-determined amount.
Both of these methods are approximately the right way to go. They will yield a nice looking dress, but other things in the picture (e.g. skin tones) may not be optimally exposed. Also, the Trial-and-Terror method is a little bit wonky unless you do it for each and every shot. If you use the correction from one shot on another and are using a meter with its own intelligence you have two brains trying to correct the same problem. This brings up images of the old hook and ladder fire truck sequences in the old Keystone Kops movies, and similar, where there was a second driver for the rear wheels who would no always steer around the same side of an obstacle as the primary driver.
