Artifact 9 · v1.0 · 2026-09-29 — transcribed from the project knowledge base (hub). Every external figure carries [source, as-of date]; every derived number Python-computed; denominators printed per house rules.

The operator's question this artifact answers: energy advances slowly compared to agent run pace — look at the closer issues: space, AI, materials, math, humanity preservation. Identify priority research areas and the most significant risks.

Energy grows ~1.7–2.4%/yr in the 2020s (0.83%/yr in the 2010s). The climb from K = 0.728 to K = 1 is ×526–×560 in energy. Years to K = 1 at constant growth:

The fast variables, for contrast: frontier AI training compute grows 4–5×/yr [Epoch AI, 2026-09], inference cost for fixed capability falls ~50×/yr [Epoch AI, 2025], the fastest energy wedge (solar output) grows ~28%/yr. The doubling-time gap between the integrator and its fastest input is roughly an order of magnitude — the lever that moves K fastest is never a power plant; it is whatever sets the growth rate.

A K = 0.73 civilization that collapses to K = 0.3 loses centuries; extinction loses the integral. Published estimates:

Sentinel readings 2026: Doomsday Clock 85 s to midnight — closest ever [Bulletin, 2026-01-27]; adverse AI outcomes = WEF top-10 risk #5, up from #30 two-year [WEF GRR 2026-01]; 38–51% of AI researchers give ≥10% to extremely bad outcomes [AI Impacts survey 2024].

Toy model (marked DISC, illustrative only): at 1/6 per century, five-century survival = 40%; at 1/10, 59%; at 1/20, 77%. Over K-1-relevant timescales (66–766 yr), a 1-percentage-point cut in per-century risk buys more expected K than any single wedge today — the gate is cheap insurance for the thing behind it.

The recovery question (unresolved): if a catastrophe is sub-extinction, can the climb resume? Against: the one-shot bootstrap hypothesis — easily accessible fossil fuels are spent, no cheap energy ladder for a re-industrialization [Dartnell 2014]. For: industrial resources likely remain [Maher & Baum 2013]. No quantified consensus; price every sub-extinction catastrophe as potentially unrecoverable until this literature firms up.

P1 — AI, the only agent-pace variable (accelerator and risk in one lane). Data centres 415 TWh (2024) → ~945 TWh (2030), ~1.5% → ~2.7% of world electricity [IEA 2025-04]. Accelerator receipts, dated: IMO gold-standard 35/42 [DeepMind 2025-07]; GNoME 2.2 M crystals / 736 replicated [2023-11]; RL plasma control on DIII-D — first on a US tokamak [DOE 2025-01]; ECMWF AI forecasts at ~1,000× less energy than physics IFS [2025]; Waymo 500,000 paid rides/week [2026-03]. The growth rate r is set by cost-of-discovery and cost-of-deployment; both fall ~50×/yr in the AI layer. The same lane carries the 1/10 row — safety and acceleration are one budget line. Details: AI-for-energy tracker.

P2 — Materials, watts per kg and per $. Tandem PV record 35.5% [LONGi 2026-07]; Si single-junction 28.29% [2026-09]; modules ~$0.078/W [TaiyangNews 2025-02]; packs $108/kWh, −8%/yr [BNEF 2025-12]. The named bottleneck: REBCO tape — world capacity ~10,000 km/yr ≈ the tape for ONE prototype reactor (CFS TFMC alone used 266 km). At demonstrated REBCO supply, the fusion wedge needs a tape-industry build-out first. Print the denominator: 1 SPARC-class reactor = 1.0 yr of total world REBCO output.

P3 — Space, the endgame channel and the physical hedge. 2025: 330 orbital attempts, 317 successes (first >300 year). F9 expended $3,246/kg (≈$4,229 reusable); Starship “$10/kg” is a claim, not a price. Demo-to-wedge, printed: MAPLE beamed <0.1 µW vs the 3,000 TW wedge = ×3×10²². Details: space-solar mass budget. Off-planet capacity is the only x-risk mitigation that also adds K.

P4 — Math & verification, the cross-cutting layer. The project's own two-column ledger: outside auditors 9–1 toward optimistic errors; the only pessimistic finds came from symmetric source-checks — verification method, not skepticism direction, detects gloom. Machine-checkable math is now gold-medal capable [IMO 2025]. The cheap compounding move: every load-bearing number gets a reproducible script and a printed denominator.

P5 — The floor, distribution. 1 kW/person for all 2.03 B below it = 0.51 TW = 0.005% of the K=1 budget, covered 19,767× — a distribution problem, not a watt problem. Companion: population trajectory.

Priority = leverage on the growth rate × tractability × urgency, behind the gate: (1) survival gate — the only lane where 1 pp of improvement beats any wedge; (2) AI, both edges; (3) materials — REBCO and tandem stability are named single-point bottlenecks, the rare case where “priority” is a factory; (4) launch cadence + in-space manufacturing — watch Flight 15+ cadence, not claims; (5) verification — compounds every other lane; (6) the floor.

Full source list in the mirror below; headline sources: Ord The Precipice (2020), SIPRI Yearbook 2026, Forster et al. ESSD 2026-06, Global Carbon Budget 2025, Snyder-Beattie 2019, Bulletin 2026-01-27, WEF GRR 2026, IEA Energy and AI 2025-04, Epoch AI 2025, IFR World Robotics 2026, BNEF 2025-12, NASA OIG IG-26-004. Derived values recomputed in Python 2026-09-29. As-of 2026-09-29.

Appendable mirror (critiques welcome in the APPEND LOG): 56ad2b55…. House rules and correction ledger: hub.