The Kronaxis Polling Method: A Pre-Registered, Hash-Anchored, Ward-Level Methodology for English Local Council Election Forecasting

We present the Kronaxis Polling Method (KPM), a falsifiable, hash-anchored, statistically grounded methodology for predicting English local council election outcomes. The current production version, KPM-v17.10, combines per-ward Uniform National Swing projection with Reform-target detection from ONS Census 2021 demographics and Hanretty 2016 Brexit Leave estimates, falling back to the hand-crafted KPM-2.2 v15.1 fragmentation rule in mid-confidence cases. On a held-out test set of 40 May 2026 councils, v17.10 achieves 72.5% accuracy — statistically indistinguishable from the v15.1 baseline at the same 72.5% (bootstrap 95% confidence interval on the difference: [-17.5, +17.5] percentage points). The headline overall-accuracy null is, however, structurally informative: v17.10 has approximately 2× the recall for Reform UK council wins compared to v15.1's broader n=130 sample, at comparable precision. Both methodologies share a 0% recall blind spot for Conservative and Green council wins on the 2026 cycle. Three large-language-model approaches were tested as alternatives and all under-performed both the hand-crafted rules and a trivial always-NOC baseline; we report the failure record openly. The methodology is pre-registered for the 2027 county council cycle and the 2028 metropolitan and London borough cycles with frozen SHA-256 hashes, public reproducibility tests, and a published falsification criterion. All code, data, predictions, and audit history are open at github.com/Kronaxis/kpm. Reproducibility tests: run python3 -m tests.test_backtest (verifies KPM-2.2 v15.1 = 59.2% on n=130) and python3 -m tests.test_v17_10_backtest (verifies KPM-v17.10 = 72.5% on n=40). Both run nightly in public CI. Production methodology hash: 2ea86b8d1e25ee68ebf66c6f59496e2480a43dba2a81ff17012ce0094531f018.