Nothing here asks for trust. The engines re-derive the frozen design points on load, the anchor regression pins every number, and the deposits let anyone reproduce the record from scratch.
The engine must reproduce every frozen anchor within its stated tolerance: 5 breeder anchors (config 22021), 5 burner anchors (the M-45 design point), the tritium surplus law, the analytic neutronics surrogate, and 5 high-fidelity surrogate checks. This table is the live output, re-run at build time — source in engine/kronos_toolkit/verify/, raw results in data/regression.json.
| Check | Expected | Got | atol | Verdict |
|---|---|---|---|---|
| breeder.Q | 3.076326 | 3.076326 | 0.0001 | PASS |
| breeder.P_fus_MW | 85.040577 | 85.040577 | 0.5000 | PASS |
| breeder.Ip_MA | 9.656483 | 9.656483 | 0.0100 | PASS |
| breeder.T_kg_yr | 3.836229 | 3.836229 | 0.0100 | PASS |
| breeder.f_n | 0.797120 | 0.797120 | 0.0001 | PASS |
| burner.Q_E | 1.318120 | 1.318121 | 0.0001 | PASS |
| burner.P_fus_MW | 4298.500000 | 4298.499304 | 0.5000 | PASS |
| burner.P_n_MW | 233.950000 | 233.950944 | 0.2000 | PASS |
| burner.f_n_pct | 5.442600 | 5.442619 | 0.0100 | PASS |
| burner.phi_i_keV | 248.750000 | 248.748154 | 0.0500 | PASS |
| surplus.1.34 | 1.690000 | 1.691289 | 0.0200 | PASS |
| surplus.1.80 | 3.980000 | 3.979503 | 0.0200 | PASS |
| neutronics.local_tbr | 1.151700 | 1.151700 | 0.0010 | PASS |
| neutronics.net_tbr | 0.742400 | 0.742400 | 0.0010 | PASS |
| hifi.gyrokinetic | 0.689655 | 0.689655 | 0.0000 | PASS |
| hifi.neoclassical | 0.632456 | 0.632456 | 0.0000 | PASS |
| hifi.struct_fea | 0.785000 | 0.785000 | 0.0000 | PASS |
| hifi.kinetic_mirror | 0.428388 | 0.428388 | 0.0000 | PASS |
| hifi.nucdata | 18.353000 | 18.353000 | 0.0010 | PASS |
The in-browser engine on these pages runs the same check on load (self_check()) before any control unlocks.
Every check in the reproducibility suite (V01–V23, 23 checks). PASS = reproduces/holds; REQUIREMENT = an open item stated honestly; INFORMATIVE = a scan finding. Click any check for its full claim, criterion and computed value.
| ID | Check | Category | Verdict |
|---|---|---|---|
| V01 | Hyperion frozen design point reproduces | design-point | PASS |
| V02 | Gain is SOLVED not scaled | gain | PASS |
| V03 | Fuel follows purpose (min Ip) | fuel-trade | PASS |
| V04 | Closing-config counts | fuel-trade | PASS |
| V05 | Toroidal field is not a second lever | field-lever | PASS |
| V06 | Neutron multiplication is D-T only | blanket | PASS |
| V07 | TBR is a design lever | blanket | PASS |
| V08 | Startup inventory inside civil precedent | fuel-cycle | PASS |
| V09 | Heating power density is physical | power-density | PASS |
| V10 | betaN respects the no-wall limit | stability | PASS |
| V11 | Peak field within demonstrated | magnet | PASS |
| V12 | Uckan / enclosed-current identity | current | PASS |
| V13 | Ash basis is a declared assumption | honesty | INFORMATIVE |
| V14 | Triangularity declaration (H27) | honesty | REQUIREMENT |
| V15 | External current drive is required | current-drive | REQUIREMENT |
| V16 | He-3 spectrum-destruction rule | fuel-cycle | INFORMATIVE |
| V17 | The current requirement is structural | fuel-trade | PASS |
| V18 | Design-space minimum-current finding | design-space | INFORMATIVE |
| V19 | Divertor exhaust is OPEN not adverse (H24) | exhaust | REQUIREMENT |
| V20 | Negative triangularity eliminates ELMs (H25) | stability | PASS |
| V21 | NT is not a detectable performance lever (H26) | confinement | INFORMATIVE |
| V22 | Steady-state and damage-life-limited (H29) | operational-honesty | REQUIREMENT |
| V23 | Per-MW tritium rates must name fuel and basis (H35) | fuel-cycle | INFORMATIVE |
The systems-engineering layer above the physics: every top-level requirement traced to its verification method, the multiphysics analysis that addresses it, and the specific deposit checks that are its evidence. Status is honest about maturity — nothing is hardware-tested because nothing is built yet; “verified” here means verified by analysis and reproducible validation against the frozen model.
| ID | Requirement | Subsystem | Method | Analysis | Validation evidence | Status |
|---|---|---|---|---|---|---|
| REQ-B01 | Fusion gain Q ≥ 3.0 at the design point | Plasma physics | Analysis | Physics Engine ↗ | V01V02 | ANALYSIS-VERIFIED |
| REQ-B02 | Tritium breeding — on-model, thin margin (net TBR ≥ 1.0; lever 1.03–1.20) | Blanket & fuel cycle | Analysis | Neutronics ↗ | V06V07V08V16V23 | MODEL-ONLY |
| REQ-B03 | Peak on-conductor field ≤ 20.1 T (REBCO limit) | Magnet systems | Analysis | Structural ↗ | V05V11 | ANALYSIS-VERIFIED |
| REQ-B04 | MHD stability — β_N below the no-wall limit | Plasma physics | Analysis | Structural ↗ | V10V20 | ANALYSIS-VERIFIED |
| REQ-B05 | Plasma current sustained (I_p, current drive) | Current drive | Analysis | Electromagnetic ↗ | V12V15 | ANALYSIS-VERIFIED |
| REQ-B06 | First-wall lifetime ≥ 2 fpy at 1.97 MW/m² | First wall / PFC | Analysis | Neutronics ↗ | V09 | MODEL-ONLY |
| REQ-B07 | Divertor exhaust within target material limits | Divertor / PFC | Analysis | Thermal ↗ | V19 | MODEL-ONLY |
| REQ-B08 | Structure survives disruption/EM + seismic loads | Structure / vessel | Test | Structural ↗ | — none yet | OPEN |
| REQ-B09 | Magnet quench protection (hot-spot ≤ 200 K) | Magnet systems | Analysis | Electromagnetic ↗ | — none yet | MODEL-ONLY |
| REQ-B10 | End-of-life waste ≤ Class C low-level waste | Materials / safety | Analysis | Neutronics ↗ | V13V14V22 | MODEL-ONLY |
Where two subsystems meet, the controlling parameter and its frozen value are the interface. These are the interfaces an as-built machine must hold to the design.
| ID | Interface | Controlling parameter | Frozen value | Owner |
|---|---|---|---|---|
| ICD-B1 | TF magnet ↔ vessel structure | Peak field / reacted hoop load | 16.84 T · hoop reacted by coil case | Structural ↗ |
| ICD-B2 | Plasma ↔ first wall | Combined heat + neutron flux | 1.97 MW/m² | Neutronics ↗ |
| ICD-B3 | Blanket ↔ tritium plant | Tritium throughput | ~4.0 kg/yr · TBR ≥ 1.05 | Neutronics ↗ |
| ICD-B4 | Divertor ↔ coolant loop | Peak surface heat flux | ≤ 10 MW/m² at 90% radiated | Thermal ↗ |
| ICD-B5 | Cryostat ↔ magnet cold mass | Insulating vacuum / temperature | 1.2×10⁻⁸ Pa · 4.5 K | Thermal ↗ |
2,000 Monte-Carlo samples run through the real evaluator, varying the genuinely uncertain inputs (H₉₈, T_i0, He-4 ash, Greenwald fraction, TBR) around the frozen point. Regenerated at the freeze from dt_evaluator.py — not the pre-freeze scan.
The layer that closes the digital thread: measured geometry checked against the frozen design, with each deviation propagated back into the multiphysics analysis and the requirement it affects.
⚠ Representative — no as-built article exists (nothing is built). The as-built values below are synthetic deviations at typical fabrication tolerances, shown to exercise the workflow. This tab is the receiving structure for real CMM / laser-tracker / photogrammetry / NDE data at manufacture.
| Feature | System | Nominal | Tol | As-built | Δ | Analysis | Status |
|---|---|---|---|---|---|---|---|
| Major radius R₀ | Laser tracker | 1.200 m | ±2 mm | 1.201 m REP | +1.30 mm | Physics Engine ↗ | IN-TOL |
| Minor radius a | Laser tracker | 480.00 mm | ±1.5 mm | 480.60 mm REP | +0.60 mm | Physics Engine ↗ | IN-TOL |
| TF coil radial position | Laser tracker | 500.00 mm | ±0.5 mm | 500.70 mm REP | +0.70 mm | Electromagnetic ↗ | OUT-OF-TOL · MRB |
| TF-to-TF spacing (×16) | Laser tracker | 405.00 mm | ±0.5 mm | 405.30 mm REP | +0.30 mm | Electromagnetic ↗ | IN-TOL |
| First-wall gap | CMM | 20.00 mm | ±1 mm | 20.40 mm REP | +0.40 mm | Thermal ↗ | IN-TOL |
| Blanket module fit-up | CMM | 5.00 mm | ±1 mm | 6.20 mm REP | +1.20 mm | Neutronics ↗ | OUT-OF-TOL · MRB |
| Divertor target alignment | Laser tracker | 100.00 mm | ±0.3 mm | 100.42 mm REP | +0.42 mm | Thermal ↗ | OUT-OF-TOL · MRB |
| Vessel roundness (datum) | Photogrammetry | 0.00 mm | ±3 mm | 2.10 mm REP | +2.10 mm | Structural ↗ | IN-TOL |
| CS concentricity (datum) | Laser tracker | 0.00 mm | ±0.4 mm | 0.31 mm REP | +0.31 mm | Electromagnetic ↗ | IN-TOL |
| Vessel weld penetration | NDE | — | — | — | — | Structural ↗ | PENDING |
Every out-of-tolerance feature is propagated to the analysis field it perturbs and the requirement it puts at risk — the as-built geometry re-enters the model rather than sitting in a report.
Where the Hyperion breeder sits against reference tokamaks. This is honest positioning, not a superiority claim: Hyperion is a compact spherical tokamak optimised to breed tritium, neutrons and helium-3 — not to maximise Q. The burner is a D–³He tandem mirror, a different confinement class, compared separately by net electric output.
Every headline claim rests on a pre-registered test that regenerates from the open deposit. These are the five that matter — shown with their verified verdicts. You don't have to trust us; you can run them.
Verdicts below reproduce from the deposit (Tier-1 byte-identical where marked). The heavy re-runs execute on the record machine via the published runbook and the kronos-py / kronos-ml / kronos-quantum environments — not in this browser. Integrity (Test 1) was last re-verified here: 182 / 182 OK.
| # | Test | What it checks | Verified result | Tier | Regenerate from |
|---|---|---|---|---|---|
| 1 | Artifact integrity | 182 recorded artifacts vs their md5 | 182 / 182 OK | Tier-1 · byte | repro_driver.py --verify (kronos-py) |
| 2 | RL safety clamp | Envelope escapes over 150,000 control steps | 0 escapes / 150k | Tier-2 · pre-registered | clamp_activation_stats.csv (kronos-ml) |
| 3 | Calibration spine | Model-to-data improvement factors | 1691× / 36.8× / 7475× | Tier-1 · byte | calibration_loop.csv (kronos-ml + MC engine) |
| 4 | Quantum advantage | Fault-tolerant resource crossover this decade | no crossover (classical wins) | Tier-1 · byte | crossover_verdict.csv (kronos-quantum) |
| 5 | CBF safety certificate | Control-barrier faces + calibration-through-proof | 3 / 4 faces · 10 / 10 | Tier-2 · pre-registered | track_R3_1 formal-safety (kronos-ml) |
unzip. 2. Build the three environments — kronos-py, kronos-ml, kronos-quantum (conda-forge; versions pinned in the runbook). 3. python track12_repro/repro_driver.py --verify → expect 182 OK. 4. Regenerate track-by-track (repro_driver.py prints the plan) and re-verify. Tier-1 must return byte-identical; Tier-2 (trained-model metrics) reproduces within the pre-registered 5-seed tolerance — a few Tier-2 files reporting CHANGED after retraining is expected, not a failure.Every headline reproduces from the open deposit. Validation is public; the science is falsifiable. IP posture is open results with protected implementations.
| ID | Check | Result | Verdict |
|---|---|---|---|
| V01 | Frozen design point reproduces | worst rel 0.0e+00 (bitwise, NumPy 2.x) | PASS |
| V02 | Gain is SOLVED not scaled | Q 3.0763 from power balance | PASS |
| V03 | Fuel follows purpose (min I_p) | D–T 4.93 MA vs D–D 15.52 MA | PASS |
| V04 | Closing-config counts | D–T no-wall 213/10800 | PASS |
| … | V05–V23 (19 more) | field/blanket/current-drive checks | PASS |