SOVEREIGNTY CASE · WORKING NOTE · T2 — INTERNAL & TRUSTED DEMOS
With the ratified levers live, the posture choice is real: commit 78 of 108 TWh (78 productive, 0 exported) — or hold at 77.6 and buy sovereignty rungs
CM-19 (our anchor-selection model) runs with the full ratified stack live — the D-052 execution-probability haircut, the hard concentration cap (limits how much of the surplus any single buyer can take), and the D-056 levers (0.70 utilisation dial, export_residual candidate) — falsification-tested (put through a test designed to disprove it, which it must survive) by exact enumeration. Three findings; the policy gate is closed.
1 · The levers worked: e-fuels returns at its cap-feasible size, export competes honestly, three real postures emerge
Under D-052 alone, e-fuels (synthetic fuels made from surplus renewable electricity) could never be cap-feasible and 72 TWh (terawatt-hours) sat unaddressed. The D-056 levers change the geometry: with export competing and the dial at 0.70, e-fuels re-enters at 28 TWh — its cap-feasible size, not the old 60 — and max-commit postures (the named strategic options this note compares) place 78 TWh (78 productive + 0 exported). Resilience trades a data-centre slot for cold-chain flexibility. Sovereignty-first stays at 77.6 TWh by choice, spending slots on ammonia-fertiliser and grid-storage — the basics ladder over volume. All portfolios pass falsification inside the cap.
Exhibit 1 — Surplus allocation by posture (P10 whiskers dashed)
2 · What remains is the honest trade: 0 TWh separates maximum commitment from the sovereignty posture
The 0 TWh export slot is a POLICY statement made visible: that energy is sold, not converted into domestic capability. The sovereignty posture holds 30 TWh in hand and converts fewer TWh into more basics rungs (steps up the industrial value-add ladder, from raw commodity toward finished product) — and with the AU basics data staged (D-058), which rung each contract climbs becomes computable. Neither posture is wrong; the tool's job is to make the trade legible. Every anchor's falsifiable claim — an anchor is a large, patient customer that takes surplus power on a long-term contract — stays conditional on queue reform (the D-038 floor).
Exhibit 2 — Signing calendar, balanced posture (cumulative anchored TWh)
Sign-by years: earliest credible contract close per class (D-051 library)
3 · The selector earns its keep — it beats the field, and shows its test
Portfolio selection is falsification-gated: the chosen portfolio must beat both the naive count-baseline and the p90 of the random-feasible null, or CM-19 fails its own test. On the current pool it passes all four postures (balanced: 0.60 selected vs 0.00 count-baseline vs 0.48 random p90, across 18,564 feasible subsets). This is the honesty apparatus a ministry can interrogate — every anchor also carries a falsifiable claim conditional on queue reform (the D-038 complementarity floor).
In plain terms: there are 18,564 different ways to pick five anchors from the 17 candidates. Before we trust our pick, it must win two races. First, beat the no-model pick — the five you would sign without any of this maths; if the optimiser cannot beat that, it adds nothing. Second, beat the lucky guesser — we score hundreds of random five-picks and take the 90th percentile, the score a genuinely lucky guess achieves; if luck can match the model, the model is finding noise, not structure. Win both and the recommendation carries real information. Lose either and CM-19 declares itself invalid — the tool must survive the test built to kill it before it is allowed to advise.
Exhibit 3 — Falsification strip, balanced posture
Selected must exceed count baseline AND random-null p90
Decisions this note asks for
Policy gate CLOSED: D-056 executed — vectors, 0.70 dial, export_residual live; this note reflects the ratified configuration.