Missing follow-ups permit gains and losses in four high-heat coral cohorts in 1998-2018

We tested whether the sign of next-year mean hard-coral cover change can be determined after high survey-date heat. HeatCRD version 1 gave 856 Atlantic and Pacific site-source-years with TSA_DHW of at least four in 1998-2018. Only 243 had a next-year cover record from the same source. Observed pairs had mean changes from -2.85 to -0.58 percentage points. Under a 0-100% range for missing follow-up cover, all four full-cohort bounds permit gains and losses. Bounds span -11.90 to +50.17 and -8.81 to +81.28 points in the Atlantic, and -18.62 to +45.74 and -21.56 to +50.70 in the Pacific. Same-month matching leaves fewer records and wider bounds. These are finite-cohort possibility bounds, not confidence intervals or causal heat effects. Missing baseline heat, archive gaps, mixed survey methods and survey-date exposure limit inference. The package gives code, source hashes, aggregate evidence, a figure and an independent reconstruction. External record data remain outside the bundle.

coral cover · marine heat · missing follow-up · partial identification · observation bias · HeatCRD