Welcome

Read this room in one sitting or drop into any section. The reading path below mirrors the navigation. Two things carry it: the evidence, labeled at every number, and the stop rules, written before the spend.

Two series, one model. The spec rows are audited arithmetic; the real-cost rows are simulated at first-principles conversion. The room never merges them (CLAIM-TRACE S4).

The case

Superhot resource turns geothermal from a site business into a product business, only if somebody certifies the hardware that survives it. That certification does not exist above 350 °C. Building it is the company.

The ladder is one audited model, not a promise. Its rows are different project definitions and are never added together. The spec series says what would have to become true; the real-cost series says what the same project costs once the conversion physics and quoted-cost classes are honest. The reinforced run reproduces the audited $29.80 row exactly — to nine decimal places — before it re-prices anything.

p05/p50/p95 boxes from the declared stress box and the contract box. Boxes are declared ranges, not probabilities (CLAIM-TRACE S1).

The mission number is a design spec, not a forecast. R6D is the surface cost (about $970 per gross-kW) that would have to land for a 500 °C doublet to hold $30 across 80% of the declared stress box and 99.9% of the contract box. The contract box assumes fixed-price EPC and long-tenor de-risked debt. It is a commercial structure, not an engineering result.

The instrument — simulation-first qualification

This is how we work and what we sell at the bottom: a qualification layer. Pre-registered falsifiers, frozen predictions, gate checkers, hashed records. The bench confirms the prediction; it does not gate our ability to have results.

Evidence flow with the gate checkers at every hand-off. The same checker judges the virtual campaign and the bench record.

Verification chain: measure, correct, command, actuate, verify, record. E9 runs it end to end under injected faults.

The next wave (E6–E9 + Q)

Every item here is modeled to the bench: frozen specs, gate checkers written, fixtures designed, RFQs drafted. Section 07 prices the run order.

DC-4 duty, frozen before fabrication (E7-DC4-DUTY-SPEC-REV0). Checkpoints: leak and stroke at 1, 10, 25, 50, 75, 100 cycles; destructive strength at 50 and 100.

E7 specimen kit · assembled

Envelope geometry from the frozen spec: casing segment 219.1 × 300 mm class, cement sheath 273 × 300 mm, connection collar 245 × 120 mm, E5 stack envelope 80 × 30 × 8 mm, end plates 320 × 320 × 20 mm. Status: CONCEPT_SPECIMEN_GEOMETRY_ONLY — not a fabrication drawing.

E7 specimen kit · exploded

Parts P-1 to P-6, exploded for presentation. Kit clearance 1.0 mm.

E8 limiter ranking from the desk case. The verdict is INDETERMINATE and stays INDETERMINATE across every robustness variant (VP4) — so we stop claiming a lever until the matched-bit tests name one.

Test readiness — modeled to the metal

Nothing below waits on a decision of ours except the sends. Each run freezes its prediction before it starts and is judged by the same gate checker that judged the virtual campaign.

Run order 7 → 6 → 5 → 8 → 9, cheapest decisive measurement first (next-stage/HUMAN-QUEUE.md §2, assumed cost ranges).

What this session squeezed virtual (VP1–VP7)

Seven pushes ran here, each judged by the real gate checkers, each recorded in qa/room-virtual-*.json with its seeds, assumptions and hash-bound sources. Deliverability and completion hydraulics are now bounded with numbers. The E7 and E9 gates were walked through pass, leak, growth, lockup, strength, laundering and unlabeled paths — every cheat caught by a named contract, every silent failure refused. The E8 INDETERMINATE survived wider boxes and three leave-one-anchor-out variants. The stroke divergence is charted and resolves as a strain reading the campaign will verify. The duty-class corners nobody has run are enumerated with their (basis, duty state, hours) rows.

Risk framework

The risks below are the ones a technical reader will find anyway. Better they find ours first, with the kill criteria attached.

Blast radius × cost to resolve. Positions are the reading of the records (inference), not measurements.

VP6 stroke envelope (this session, SIMULATED): the model needs 25.8 mm per 10 m at the S1 quench class; the E7 brief quotes 2.5–3 mm per 10 m, which reads as relief strain 2.5–3e-3 and sits where the model says. The campaign decides which reading the hardware needs.

Stop rules are design features. The E7 gate stops after two consecutive breached checkpoints. The E8 case refuses to name a lever below the 0.90 stability threshold. The E5 seal filing halted when prior art blocked the claim, and the record says so.

Moat & partnerships

Qualification is a permissioned operating record. Every campaign we run, witnessed and hashed, widens what we can certify and what a regulator or insurer will accept.

Partner routes are named by class, not by logo: national lab and hot-rock facility class (Sandia/Welltec/NER), university deep-underground lab class (EPFL/ESRF, Bedretto/OU), operator replay class (EGI/FORGE), and independent test-house class for the witnessed records. Named partners appear only with consent and a signed role.

Team & firm

Missions, not headcount. Each one is a piece of the company someone will own end to end.

The people registry holds the research and outreach record. It publishes nothing here: the disclosure line for names is a signed role plus consent, and that gate is human-only.

Glossary