Instrumental catalogue of convective surface outflows in the AVAMET network: episodes, observability masks and analysis code

Research artefact for an instrumental catalogue of convective surface outflows detected blindly in the citizen-operated AVAMET mesonet of the Valencian Community (eastern Spain), together with the measured selection function that says at which spatial scales the network was able to observe them at all. The organising result is negative and is the reason the artefact exists: a failure to detect an outflow is not evidence that no outflow occurred, and how far it falls short of being such evidence is a measurable quantity. Injected outflows of 4, 8 and 12 km core radius were recovered by the unmodified detector with probabilities of 0.039, 0.247 and 0.636 respectively, and that pattern reproduced on a calendar year held out until the method and its predictions had been frozen and deposited. Contents. The 107 consolidated episodes with their externally assigned radar levels and documentary annotation; the observability, exposure and detectability tables; the synthetic-injection results for the development archive (5,040 injections) and the 2024 holdout (840); the radar control denominators; the ERA5 environmental characterisation; the leave-one-station-out, robust-regression and quality-control analyses; and the code that produced all of it. What is not here. The raw AVAMET surface observations are owned by AVAMET and are not redistributed: access was granted for research use on conditions that cover access rather than redistribution. What is released is derived and event-level, including station identifiers and coordinates. A reader who wants the minute-level series must obtain their own authorisation from AVAMET. OPERA/EUMETNET reflectivity products and Copernicus ERA5 fields remain subject to their own data policies. Companion deposit. The freeze record (10.5281/zenodo.21862876) holds the specification, the eight predictions with their registered criteria and inconclusion rules, and the SHA-256 hash of every frozen file. It contains no results, because none are needed to check the claim it exists to support: that the method and its predictions predate the holdout. This deposit is the other half, and holds the results themselves.