The world is drilling toward $30–45/MWh geothermal and cannot see what it is drilling into. We build the instruments, the yardstick, and the mapped path from $61.65 to $29.80 per MWh.
Electricity demand is repricing firm power: data centers alone grew 17% in 2025 and head toward ~945 TWh a year by 2030 (IEA base case verified). The scarce good is a megawatt that runs at 3 a.m. in February, anywhere. Geothermal is the only source both firm and everywhere. The barrier is price.
The U.S. Department of Energy set the public target at $45/MWh by 2035 (Enhanced Geothermal Shot). Superhot rock carries 2.2–2.9× the enthalpy per kilogram of 250 °C brine verified; one or two producers make a 45–50 MW block. Our mission is the step beyond the public target: $30/MWh, around the clock, around the world. We intend to get there first, and to own the ruler the finish line is measured with.
Capital reached the frontier: Quaise $180M, Mazama $135M, Hephae $17.8M, Fervo first power — one quarter, all verified press, 2026. The race to superhot rock is funded. The instruments are not.
Superhot wells die at transients: across the 20-well base rate, failures are dominated by thermal stress at quench and kick-off verified — the hours nobody has measured. The working tool stack tops out near 200 °C; the celebrated 350 °C tools are survival ratings for four-hour runs. Every rating on the market is a survival number. None is a duty record.
Every downhole instrument dies the same two deaths: heat drifts its calibration, and the well fouls its sensors. Cool-Brain removes the heat from the problem and corrects the fouling in place. A passive Inconel waveguide — acoustic ceiling near 1,000 °C — carries machined reflectors through the hot span, while the transducer and all electronics sit behind a thermal barrier below 100 °C. The architecture has one field precedent, IDDP-2, where tools stayed under 65 °C while logging 400–500 °C rock verified. We made it permanent, packaged for the drill string.
Echo time-of-flight yields a five-point temperature profile through the span. Two corrections nobody else performs in the well: reference-notch drift correction and dual-depth amplitude-ratio fouling metrology. In the simulated 500-hour run at 450 °C, the corrected readout holds worst-case 2.23 °C against a ±5 °C gate, where the uncorrected readout drifts to 9.63 °C simulated — the bench publishes the measured figure.
The claim, kept narrow on purpose: continuous multi-point temperature profiling through transients at 374–450 °C+ duty, with in-situ drift and fouling correction and drill-string packaging.
Point sensing at 500 °C exists (Sentek). Hours-scale 350 °C logging exists (Kaldera, the memory-gauge houses). Fiber profiles exist in the lab at 300–450 °C. We name the incumbents because the claim is a combination — profile × duration × transient × calibration × duty — and the combination exists nowhere else verified (market scan 2026-09-29).
ARPA-E funded three superhot categories: well construction, heat extraction, and testing/validation/quality-assurance infrastructure. The third has no occupant — no full-tool qualification service at 375–400 °C, no superhot test standard at all. The field's own gaps report calls the testing "a major gap"; our teardown refines that report. A developer qualifying its own tools convinces nobody. The referee position is empty, and referees are never dropped from the game.
So we ship three things at once: the instrument, the qualification basis (duty classes, test taxonomy, witnessed-report contract), and the first witnessed 500-hour record at 450 °C — entry number one in our own standard. Whichever developer wins the superhot race, their tools get graded on our yardstick. Adoption is a sale, not a loss.
What compounds behind it: claim surfaces filed before any vendor contact; measured maps no paper contains (hold/release/recovery surfaces, drift-and-fouling curves, cost breakdowns, cycle-life leak data); the results brand, kept by correcting the record in public; and the operating record — every installed machine linking revision to condition to measured result. Fleet data is the moat money cannot buy.
Deep tech earns belief with arithmetic that survives attack. Our cost model is recomputed and audited to the stored trajectories model: today's best coupled study lands at $61.65/MWh, and the ladder to $29.80/MWh names the invention that must deliver each rung. No step is wishful — each is a requirement with a bench gate.
Three lines of business stand on it: instruments (Cool-Brain, then the series); witnessed qualification — revenue from the day the record exists, because every tool maker needs it; and data and basis — the dataset and the standard itself. Product lines: Relay, Shift, Flex.
We design to $27/MWh and sell at $30–45 — the margin math demands it: the $30 row survives only 0.90% of electricity shortfall or 0.91% of capital overrun at nominal. We know exactly which cost lines decide it (installed power block ≤$1,000/gross-kW; subsurface $15–25m per pair), and we are attacking them by name.
Every invention runs a full virtual qualification campaign before any hardware: physics simulation, Monte Carlo, pre-registered falsifiers, and gates identical to the bench gates — judged by the same gate-checker code the bench campaign will use. Five mechanisms have run their campaigns: 56 contracts green, every gate and falsifier executed simulated. The E1 release-latch campaign passes its real decision rule in 95.8% of virtual campaigns (force window 7.92–14.88 N, release channel 54–137.6 µm).
The discipline is qualification engineering's own: analysis alone cannot qualify hardware (IEEE-323). Sims are design authority; gates close on measurements. Failure is informative and published — attenuation maps and leak modes become named redesign levers. Every number keeps its label, in every document, forever. In a field of mythology, the corrected record is the brand.
The next ninety days produce three numbers, each of which converts rooms into capital: R1, the chain proof — the waveguide hears its notches at room temperature, days away; R2, first hot data with deliberate drift and fouling attacks, plus the commanded Shift-latch release and the power-block cost-gap memo; R3, the 500-hour qualification record at 450 °C and Shift repeatability on three well pairs. Behind them, the money ladder: sprint capital in hand now, the $2–3M pre-seed open against this schedule, Series A on repeatable service, project finance at the fleet rungs.
Then the physical ladder: a three-well design-partner program proving repeatable service, the first 45–50 MW block landing the band, and a fleet at $30/MWh whose 25-year operating records compound the moat. By the fleet rung we manufacture the power-block modules, the strings, and the qualification services. Think in gigawatts. Build in metal. Design the first machine for the next thousand.
One founder today: a self-taught engineer who co-founded and shipped a data platform in mining, then built this field's first adversarially-verified evidence base, corrected its folklore in public, reduced the instrument to practice, and wrote the qualification basis the segment never had. Four founding seats are open — ultrasonics, HT packaging, qualification, drilling operations — each with founding equity and named artifacts: their name on the paper and on the record. The team is a use of proceeds, not an afterthought.
Capital: the $2–3M pre-seed (lead $1–1.5M) against the R1–R3 schedule, milestone use of proceeds. Operators: a duty profile and one failed-instrument story — design-partner conversations open now. Tool makers and labs: review of the qualification basis; furnace and test-house time. Engineers: a founding seat and a defining piece of work to own.
Thirty minutes for a technical review. That is the whole ask. The numbers, the gates and the falsifiers are ready to be attacked.
The full evidence trail is claim-traced in our evidence repository. The honest index for the numbers above:
| Claim | Label | Source |
|---|---|---|
| $61.65 → $29.80/MWh spec ladder; $34.06 and $47.46 rows | model | audited arithmetic, stored trajectories — recomputation in the QA record |
| 20-well failure base rate — thermal stress at quench/kick-off | verified | supercritical well-base study, claim W1B-019 |
| 2.2–2.9× enthalpy per kg vs 250 °C brine; 1–2 producers ≈ 45–50 MW block | verified / inference | claims W1C-019/020 |
| IDDP-2 tools <65 °C logging 400–500 °C rock | verified | field record, claim W1A-026 |
| Corrected 2.23 °C vs uncorrected 9.63 °C over simulated 500 h @ 450 °C | simulated | digital furnace, nominal parameters; bench run publishes the measured figure |
| 5/5 mechanisms, 56 contracts green; latch window 7.92–14.88 N | simulated | virtual-qualification campaigns, same gate checkers as the benches |
| Capital wave (Quaise $180M · Mazama $135M · Hephae $17.8M · Fervo first power); no qualification service at 375–400 °C | verified | 2026 press and filings; market scan 2026-09-29; SGW-2025 gaps report |
Nothing in this document is measured plant performance. Every simulated figure is labeled simulated. That discipline is the product: it is what a qualification company sells.