States reconsider: California's advanced-nuclear assessment (AB 2647) + fusion framework (SB 925)

status: open, opened 2026-10-03, last moved 2026-10-03. Energy nuclear fusion us california

• The thread: the states whose moratorium-era politics defined US nuclear hesitancy are writing dated, mandated re-assessments. As enacted Sep 30, 2026: AB 2647 (Calderon) — CEC assessment of advanced nuclear's role due Jan 1, 2028; SB 925 — fusion strategic plan due Dec 31, 2029 + a state fusion regulatory framework + streamlining eligibility for fusion-component manufacturing. No technology set-aside or procurement mandate in either.
• 2026-10-03: opened from the dig. As introduced in February, AB 2647 proposed clarifying that the 1976 moratorium does not include advanced reactors — the enacted version is the assessment mandate, but the posture turn is now signed state policy. Both deliverables ledgered (AB Jan 2028; SB Jan 2030 review). Irony on file: the Sep-21 data-center-bills state is simultaneously hosting the AI buildout backlash and building the paperwork to add firm power. Watch: CEC assessment drafting (public-input phase), whether other moratorium-state legislations copy the AB-2647 shape, fusion pilot- plant siting competition under the 2040s resolution.
• 2026-10-03 (cross-ref): SB 925 is the first state fusion regulatory- framework process on the US record; sits beside the federal $10B fusion bill (fusion-race thread) as the policy layer's two levels.- 2026-10-04 (late evening): the independent runner exists — built it in one session. 160 stratified newsfeed items (Sep 28–Oct 4, all six topics, seed 42) + 128-row editorial gold subset (113 digest-covered + 15 hand-flagged tva FPs) through the same three SystemOne questions: Jev (hosted), pplx-decider-27b (hosted), Strands 2B (local CPU), Clef-flash (local CPU), plus production baselines (keyword gate as deployed; gemma4:e4b at the deployed prompt). Gold binary: pplx 85.2% (2/15 FPs), clef-flash 83.6% (5/15, free 9B on old CPU), strands 87.7% (n=81, 0/10 FPs), Jev 77.3% (6/15) vs the deployed gemma4 filter 71.9% (3/15) and the deployed gate 0/15 by construction. Every replacement that parses outruns production; nobody detects "on-topic but not news" (needs story memory — same reason stale-update ~25/25-unanswerable per-item). Calibration ~10 pts hot everywhere (ECE 0.09–0.13) incl. Strands' "95% at 0.9" claim. Qwen3.5:9b probe: non-viable under the deployed format:json call (answer lands in the thinking channel, parser fails 10/10 → production would default-relevant everything). Economics: 245 items/day × ~630 tok = $0.20/month hosted (pplx/Jev); free local. Ollama packaging remains the category's gap (community Strands GGUF broken — 320/324 tensors; Clef custom-code no GGUF).