Pre-registered Gaia DR4 predictions for nearby white dwarfs with excess astrometric noise

We select white dwarfs within 50 pc whose Gaia DR3 astrometry is poorly fitted by a single-star model (RUWE > 2), with no DR3 orbital solution, no sign of a resolved double and no bright neighbour within 5 arcsec. Nine objects pass. Six of them are warmer than 7000 K and have published binarity or photometric masses of 0.31-0.45 Msun, too low for single white dwarfs, so the selection mainly recovers unresolved double degenerates. LHS 3236, which has a published astrometric orbit (P = 4.03 yr, a0 = 4.37 mas), is used as a control: our DR3-based amplitude estimate (4.1 mas) reproduces it. Two objects, UCAC4 238-179131 (WD J2126-4224, 25.6 pc, RUWE 9.1) and LSPM J1318+7353 (36 pc, RUWE 3.5), have no published companion that we could find. Before the Gaia DR4 release (2026-12-02) we fix, for all nine white dwarfs: the expected photocentre amplitude, the minimum mass of a dark companion as a function of orbital period, falsifiable predictions (persistence of the excess noise, amplitude within a factor 2, P < 107 d for G 177-31 from its known RV variability, recovery of the LHS 3236 orbit, and a test of the predicted gap near 1 au in the double white dwarf separation distribution), and explicit rejection rules. A secondary list of 55 high-RUWE stars with constant Gaia RVS velocity is included. The verification script (b_verifica_dr4.py) is fixed in this deposit and will be run unchanged on DR4. The analysis was carried out with the assistance of an AI model (Claude, Anthropic); see the section "Author contributions and use of AI" in the note.