A reef-building coral is two organisms. The animal builds the skeleton; single-celled algae live inside its tissue, photosynthesize, and supply most of its energy and nearly all of its color. Under stress, usually heat, the coral expels them. What is left is nearly transparent tissue over white skeleton. The colony looks white because you are seeing its skeleton through it.
So a bleached colony is alive. Starving, possibly dying, and possibly about to recover completely. And bleaching is a state rather than an event: it can appear and resolve between two visits, so a record showing none does not mean none occurred — only that none was there on that date.
White, and white. Bleached tissue is white. Recently dead skeleton is also white. They mean opposite things, and in a photograph they can look identical. The cue is texture, not color: living tissue, even bleached, forms a smooth film that softens the fine corallite structure, while bare skeleton shows it sharply. There is a clock on it too — dead skeleton does not stay white, because turf colonizes it within weeks and turns it green-brown. That is why recent mortality is a short-lived observation.
This is hard at ordinary resolution in ordinary water. It is where you will be least certain and where the machine is least reliable, for the same reason: the difference lives in fine detail the water has softened.
Run the governing test on it. A blurry, badly lit frame makes the two harder to separate. If they were a single class, a worse frame would produce a larger damage figure and an easier claim. Keeping them apart, and allowing indeterminate, means a worse frame produces a smaller confident count and a bigger honest gap. That is the right direction.
Coarse classes, on purpose. Bleaching is recorded in a few ordinal steps — unaffected, pale, partly bleached, fully bleached — with recent mortality separate. Not as a percentage of whiteness. A number carried to two figures would claim a precision the method does not have: the color you see depends on depth, distance and light, and the honest resolution is a handful of steps. A percentage would be more precise and less true.
There is no formula for the uncertainty in a class assignment the way there is for a proportion. It is measured instead: two scorers classify the same frame independently, and the rate at which they agree is the number you report.
This is what the health reference is for. A class assigned from impression — that looks pale to me — is not reproducible by a second scorer, or by you next year on a different monitor. A reference lying in the frame beside the coral fixes that, because the comparison happens inside the photograph. The reference and the coral were shot under the same light, at the same distance, through the same water; whatever the water did to one it did to the other. So you are never asked what color the tissue was — a question the image cannot answer — only which square it matches.