Model card
Virtual Reactor Lab
A browser virtual laboratory of a reduced linear-mirror model — a simulation platform, not a
reactor. The lab kernel vrlab_core COMPOSES the tested reactor3d_core solver (one
calculation truth) and adds synthetic sensors, DirectConvert stage records, an electrical energy chain and a
component manifest. The renderer only draws kernel output.
No net-energy, reactor-performance, engineering-feasibility or high-voltage construction
guidance. All sensors are SYNTHETIC; loss factors and conditioning efficiency are ASSUMED and labelled.
Fidelity ladder (declared, not faked)
- VIS — hall/tour visuals only; no scientific output claim.
- L0 — analytical ideal upper bound (loss-free collector, no stopping).
- L1 — reduced loss-aware model (default): static plasma profiles, weighted alphas, detector efficiency, parameterised DC losses.
- L2 — field-map trajectories: planned; the kernel REFUSES to fake it.
- L3 — WarpX / Geant4 reference runs: planned (offline research path).
- DT — true digital twin synchronised to hardware: future. Per NIST usage, the current system is a model, not a digital twin.
What is calculated (browser instant path)
- Child reactor run via
reactor3d_core: reactivity, Poisson events, Boris trajectories, detectors, fixed-stage DirectConvert, exact alpha ledger. - Per-alpha DC stage events (entering energy, assigned stage, recovered, loss, rejection reason) summing EXACTLY to the ledger channel.
- 10+ synthetic sensors derived from the run state (or declared assumed), deterministic noise seeded by run identity.
- Energy chain: gross DC → one assumed conditioning step → usable model power into a simulated load.
Traceability
- Two-level identity for the LAB run (input_fingerprint + run_hash) plus the child reactor identity in outputs.
- The default lab run's child IS the canonical Reactor 3D run — provable lineage, pinned by tests.
- Custom runs get distinct identities and are never linked to the canonical Hub record.
- Component manifest with stable IDs (VESSEL-001, COIL-MIRROR-L, DC-GRID-01…) generated by the kernel.
Honest display rules
- Particle markers enlarged; each is a weighted macro-particle (banner shown).
- Physical time vs slow-motion multiplier displayed side by side.
- Sensor badges: synthetic / assumed / ok / saturated — no reading exists that is not derived from the run state or declared as an assumed input.
- Bremsstrahlung stays a separate plasma diagnostic outside the alpha budget.
Verification (vrlab tests)
- Pinned canonical lab identity + child lineage.
- Sensor determinism; calorimeter == ledger channel; DC current charge == stage events (exact).
- Energy chain closes exactly; L0 gross ≥ L1; non-negative channels.
- Planned fidelities rejected with a clear message.
Research path (specified, not deployed)
- FastAPI + PostgreSQL + object storage + job workers for sweeps and L2/L3 reference runs (this static site runs the instant path only).
- Blender/glTF asset pipeline, WebGPU renderer and WebXR: planned upgrades documented in the architecture note.