Wickie — An auditable beta screening framework for Iberian wildfire and drought risk: methodology note for release 0.4.0-beta.1

We describe Wickie, a screening framework that ranks Iberian regions by wildfire and drought stress to help catastrophe-risk analysts decide where to look first this week. Wickie is a screening signal, not a loss model and not — in this release — a statistically mature predictive model. It produces a 0–100 ordinal stress score per region as a documented weighted blend of three components: a hazard channel (fire-weather and drought), an exposure channel, and a bounded short-fuse alert lift from official meteorological warnings. The structural form, source provenance for every channel, the validation harness, and the promotion gate are public; release-specific weights, the alert cap, and SPI-to-channel knee values are versioned and exposed under an authenticated validation:read scope. The paper documents a structurally transparent screening framework with authenticated access to release-specific constants, not a fully open scoring rule. The drought channel is a real Standardised Precipitation Index (SPI; McKee et al., 1993) computed from a 30-year (1991–2020) NASA POWER climatology fitted at eight Iberian station points, with mixed-zero treatment after Stagge et al. (2015) and a Kolmogorov–Smirnov fit-quality diagnostic (a diagnostic, not a calibrated p-value, since gamma parameters are estimated from the same sample used to test fit). A documented SPEI-like (P − PET) fallback complements SPI; its use of Thornthwaite PET is acknowledged as a limitation, and the SPEI-like channel is explicitly ineligible for promotion to the headline rank until a true log-logistic + Penman–Monteith implementation ships or empirical equivalence is demonstrated. In release 0.4.0-beta.1, the SPI channel is shadow: it is shown live in the driver detail and used to flag source disagreement (which lowers data quality, not the score) but does not yet move the headline rank. The headline hazard component is a documented blend of a demo hazard fixture and an EFFIS/FWI fire-weather component whose WMS layer availability is checked live; numeric point sampling for fire-weather remains fixture-backed pending a documented validation step. Validation runs against six named historical Iberian cases (Sierra Bermeja 2021; Pedrógão Grande 2017; Andalusia drought 2022–23 and 2023–24; central Portugal wildfires September 2024; a Costa del Sol heat-and-wind watch case) with seasonally and geographically matched negatives by construction. Reported metrics are Mann–Whitney AUC with non-parametric bootstrap 95% CIs, calibration deviation, Spearman rank correlation, median lead time, and informational lift over single-source public baselines (random, alert-only, populous, hazard-only, drought-only). Promotion of any channel to the headline rank requires that channel to clear every metric tolerance and show positive median lift over every baseline with no material degradation, over two consecutive snapshots, plus leave-one-case-out and weight-perturbation stability. No channel has cleared the gate in this release; the harness reports promotion deferred for both the demo bench and SPI. The discipline is the product; the released ordering reflects that discipline. Source-available under LicenseRef-CatRiskSignalEngine-1.0; commercial use requires a separate licence. Decision-maker accountability is treated as a contractual matter under the law of England and Wales (Hedley Byrne & Co Ltd v Heller & Partners Ltd 1964 AC 465).