Infinity Engine — Locked Pre-Registered Forecast for Solar Cycle 25 Declining Phase and Cycle 25 → 26 Minimum

This is a falsifiable, pre-registered, SHA-sealed forward prediction generated by the Infinity Engine's harmonic-regression forecaster, applied to the SILSO daily total sunspot number V2.0 record (1976-01 through 2026-03, monthly mean smoothing, 603 samples). Method (locked): Least-squares harmonic regression with 3 spectral components (no polynomial trend) over the period range 730–20000 days, with bootstrap (n=300) prediction intervals and physical clipping at 0. The locked configuration was selected via honest hindcasts on solar cycles 23, 24, and 25 — both adding more components and adding a polynomial trend produced WORSE out-of-sample errors despite higher in-sample R², a classic overfitting signature. The Occam choice (3 components, no trend) was retained. Components recovered: Schwabe at 10.80 y (canonical 11.0 y), 8.62 y sub-harmonic, and Hale magnetic at 21.87 y (canonical 22.0 y). Fit R² = 0.7151. Hindcast performance characterization: ~30% mean amplitude error and ~2 y mean timing error on the past three solar cycles. The cycle 25 hindcast (training data ending December 2018) reported 192 against truth 159 (+20.6%). Forward predictions for the SILSO smoothed monthly total sunspot number: 2026-09: SN ≈ 82.6 [95% CI 75.4 – 90.8] 2027-03: SN ≈ 51.0 [95% CI 43.9 – 60.0] 2027-09: SN ≈ 22.5 [95% CI 15.8 – 31.8] 2028-03: SN ≈ 0 (predicted cycle 25 → 26 minimum, ~Q1 2028) Falsifiability protocol: Each forecast point is independently checkable against future SILSO releases. The prediction succeeds at a given lead time if the actual smoothed monthly value falls within the combined 95% bootstrap interval plus the ~30% method-error envelope characterized from hindcasts. The cycle minimum prediction succeeds if a deep minimum (SN < 10) occurs within ±6 months of Q1 2028. No selective reporting: every checkpoint will be reported as it arrives, success or failure. Honest limitations: The forecaster uses no solar physics — no dynamo model, no precursor (geomagnetic activity at minimum), no terminator method. It is regularized harmonic regression on the past time series alone. Methods in the physics-informed class (Dikpati, Choudhuri, McIntosh terminator) use information that pure harmonic regression does not. The locked configuration was chosen after seeing hindcast results, which introduces selection bias; future predictions on other phenomena should lock the method before seeing test data. SHA-256 lock: 661365300298254985a47f776fd12a5243de7221664677fb697eaab7df2d328d Locked at: 2026-04-11. Data provenance: SILSO daily total sunspot number V2.0, Royal Observatory of Belgium / SIDC, fetched live via the LISIRD-style endpoint at https://www.sidc.be/SILSO/DATA/SN_d_tot_V2.0.txt (CC-BY-NC 4.0; Clette & Lefèvre 2016). Source code: infinity/forecaster.py within the Infinity engine repository. Classification: [METHODOLOGY — predictive harmonic forecaster, deterministic, fully traceable. First pre-registered falsifiable prediction. Hindcasted, locked, SHA-sealed, uncertainty-bounded. Awaiting falsification by future data.]