Data and code: The lost years are unrecoverable by ocean-drift back-calculation. A measured predictability horizon for sea-turtle natal-origin inversion

Analysis code and derived data accompanying a single-author study of sea-turtle "lost years" dispersal in the North Atlantic, establishing a forward/inverse asymmetry and measuring where the inverse problem fails. Forward reconstruction works. Virtual hatchlings seeded at real NW-Atlantic nesting beaches and advected through reanalysis currents reproduce the Florida to Sargasso to mid-Atlantic loop, and predict real tracked green-turtle positions at AUC 0.91 when compared like-for-like on matched drift durations. The gyre-retention loop reproduces across three independent current products, so the corridor is not a single-model artifact. The one fully independent spatial test, against basin-wide occurrence density, is near chance (AUC 0.59), and that gap is reported. Inverse reconstruction is provably ill-posed, and the horizon is measured. At the empirically measured gyre diffusivity (about 4000 m^2/s, from 615 drifters), origin uncertainty exceeds the 150 km inter-rookery spacing within about two weeks. The harsher model-free bound comes from relative dispersion: trajectory pairs beginning within 15 km separate past the rookery spacing in about 13 days, holding on 171,823 drifter pairs across 1989-2024. Real turtle active swimming raises the diffusivity further, so inversion is harder for animals than for passive particles. Turbulent transport destroys positional origin information almost immediately, which is why the field assigns lost-years provenance with genetics rather than back-tracking. The asymmetry is organized by a single relation: inversion is possible only while the age at observation stays below the diffusive mixing time across the origin spacing. This places the European eel (invertible, caught days old from a diffuse spawning region) and the sea turtle (erased, observed months old from closely spaced rookeries) on one phase diagram. A heritable rookery-genetic marker traces origin through a separate, complementary channel that the ocean cannot mix away, and that channel is quantified rather than waved at. No climate claim is made. Whether the delivery corridor drifts as the ocean warms is unresolvable on current data. The corridor's decadal drift is product-dependent and underpowered, and a turtle drifts for years before its endpoint is observable, so the post-2016 window is intrinsically a handful of cohorts in any dataset. That bound is stated rather than a finding. An earlier version of this work led on a "warming in place" result that did not survive adversarial recompute and was cut. Contents. The forward corridor model, the predictability-horizon twin experiment, the model-free relative-dispersion measurement, the empirical diffusivity estimator, the genetics channel quantification, robustness and specification-curve checks, all derived outputs behind the headline numbers, the manuscript figures and their generators, and the companion notes each result traces back to. Raw third-party data (Dryad turtle tracks, NOAA drifters, CMEMS and HYCOM currents, OBIS and GBIF occurrences) is public and is cited at its canonical archive rather than redistributed. License: code MIT, data and text CC-BY-4.0.