Bounded report errors permit both signs of a coral bleaching slope contrast in four ocean-period samples from 2003 to 2019

We tested how report errors can change a published heat slope contrast for coral bleaching. We compared the slope above 10% bleaching with the slope for the first 10 points. Four Atlantic and Pacific repeat-site models used 5,336 monthly units in 2003-2019. All observed contrasts were positive. Exact fixed-record bounds permit both signs in all four models when each report can change by up to five percentage points. The later Pacific contrast can reach zero at a uniform error limit of 2.10 points, or 3.57 points if reported zeros stay zero. These are worst-case limits, not estimates of actual error or confidence intervals. Errors can move all reports together and can depend on heat. Missing reports, exposure error and sampling uncertainty remain outside the bounds. The result concerns a distribution contrast, not a causal heat effect or species loss. Code, report evidence, endpoint assignments, source hashes and reproduction checks are included.

coral bleaching · heat stress · observation bias · measurement error · sharp bounds · sensitivity analysis