Water first. Oil next. Cells and shafts only if they pass.

1. 1. Close the water rig

   XP-C01. Two square meters, four facets, water at 80–120 °C. Gate G0: the energy balance closes within 5%, the instruments can actually see a 5% error, and the residual sits inside that uncertainty. No coating claim before this. No cell. No shaft.

2. 2. Prove the oil tile, then the tanks

   RC-T, then two-tank storage. Cavity against a flat control. A selective surface that survives real flux. A concentration that still pays for itself. Storage that meets a stated loss and cost. Process heat is already a product at the end of this step.

3. 3. Earn the cold tile on a spare rail

   E0 through E5, and they start after the water rig, not after the oil tanks. A flux map the chosen cell is allowed to see. Junction under its limit. DC against optical power, in the 24–32% band until a purchased cell replaces the model. Coolant on its own meter. A written reason the tile beats the simpler alternative. A dead cell that does not take the rail with it. Fail, and the field stays all oil.

4. 4. Then, maybe, a shaft

   Only against hot oil, on a brake. At least twelve steady points. Balance within 12%. At least 300 watts on one kept point. At least 30% isentropic, uncertainty inside 3 points. Leak fraction on every point. Thirty minutes without a rub. No generator on that test. A generator is the next gate, and it reports net watts after the pumps. Then the scroll, the disc, and a bought micro-ORC sit on one header. Score net electrical energy per stored thermal energy, dollars, leak, and hours before service. Unbolt the loser.

G0

Measurement. Nothing else counts until the water rig can close.

G1–G4

Oil tile, in order: cavity, absorber, concentration, storage.

E0–E5

Cold tile, its own ladder. E0 waits on G0, not on the tanks.

GY, GS

A thermal year, and dollars per megawatt-hour of heat versus gas. Not a split-field year.

Shaft, then generator, then bake-off

The disc does not get a generator until the shaft is honest.

The thermal model is every acre on oil. For a first island of 120 pods, about 4,800 m² of glass, packing so the mirrors do not shade each other, the lot is about 2.6 acres. A typical West Texas year on that model is about 3.7 GWh of heat. Noon is not that number. August is fat and January is not, and the model is a hypothesis, not a bid.

Dollars, when we publish them, are dollars per megawatt-hour of that heat, against the gas boiler, with the cost assumptions written down. Not against silicon.

A split field does not get an annual kilowatt-hour until the year is rerun with two tiles, hourly sun, cell temperature, diffuse loss, demand, and a storage state. Until then the 3.7 GWh figure stays labeled thermal path.

• Drawn: the pod, the secondary and the dump, the oil tile, the tanks, the water rig, the uncertainty budget.
• Not drawn: the cold tile. No fake dimensioned cell.
• The disc sheets are pictures. Gap stays an experiment inside 0.4–1.0 mm. It is not frozen at 0.1 mm, and it is not the number printed on the image.
• The scroll is a catalogue machine. It does not need a heroic drawing to be the baseline.

Open the sheets