A frame is not a measurement until somebody has scored it. That happens topside, on a screen, and it is a different skill from taking the picture.
Points, not outlines. The obvious way to measure how much of a frame is coral would be to trace every colony and add up the areas. Nobody does it, because tracing an outline is a judgment about where a boundary lies, made hundreds of times per frame, and boundaries on a reef are genuinely ambiguous. Where does live tissue end and turf begin? Those judgments do not cancel. A scorer who draws generously produces systematically higher cover, every frame, forever.
Point intercept replaces the judgment with a sample. Points are scattered across the rectified frame and you answer one question at each: what is directly underneath this one? The answer is a category, not a boundary. The proportion of points on hard coral is your estimate of hard coral cover.
Not fewer than 25 points per frame, and you record how many you used. Twenty-five is not a number this course invented — it is what the established monitoring programs use, and matching it is what lets your Entry sit alongside their data instead of alone.
More points buy less than you would think. Halving the sampling uncertainty costs four times the scoring. And there is a floor underneath it: once that error drops below your own disagreement with a second scorer, more points refine the arithmetic and not the answer. More points make the number tighter, not truer.
The class list is not yours. You score into the benthic classes in general use — hard coral, soft coral, dead coral, macroalgae, coralline algae, turf, other invertebrates, and non-living substrate split into sand, rubble and rock. You could invent a better list for your reef. It would describe your site more finely and be worth less, because nothing could be compared with it. Most of the value of a monitoring record is the value of the corpus it joins.
A machine may propose; you decide. Software can label every point in a frame and turn an hour into minutes. The confirmed classification is the score of record, the raw model output is logged and never reported, and the model version is written down. Models change silently between versions — if the raw output were the record, a software update between two visits would appear in your data as change at the reef. That is a faded reference chart, in software.
Two things come free. Because you review every proposal, your agreement rate with the model costs nothing and is worth reporting. And the distinction this course cares most about — bleached tissue against recently dead skeleton — is the one a model gets wrong most often, which is exactly where you should not stop looking.