A joint plan required 183 coral values versus 253 in a fixed ranking for 90% range cuts in three early Pacific designs

We tested recovery plans for missing coral bleaching percentages in four Atlantic and Pacific samples from 2003 to 2019. We used three fixed heat-response designs with different seasonal controls or survey weights. Under equal cost and exact recovery, binary optimization found the smallest set of values that removes at least 50%, 75% or 90% of each nonzero missing-data range. All 12 minimum counts were confirmed by rounded fractional lower bounds. In the early Pacific sample, a joint 90% plan required 183 values. Following the primary-model ranking until every design met the target required 253. At 50%, primary-only plans failed to halve every range in two of four samples. Gains were not uniform across regions or periods. No value was recovered. The plans assume available, meaningful missing percentages, fixed covariates and independent outcome limits from 0 to 100. Unequal costs, lost source records, survey selection and measurement error limit use. The results concern finite archive ranges, not confidence intervals or causal heat effects. Code, candidate weights, plans, source hashes and numerical checks are included.

coral bleaching · heat stress · observation bias · missing outcomes · data recovery · integer optimization · seasonal controls