HBF Development Platform

One connected research chain — not a demo pile.

From transparent physics, to detector evidence, to energy-conversion modelling, to independently reviewable research objects. Every calculation returns one universal run envelope with identity, warnings and lineage; every claim links to its inputs, model version and reviewer state.

Run DirectConvert in your browser → See the evidence chain

0
automated tests
0
DC-VAL checks
0
connected products
0
Python kernels
Physics simulation Detector evidence Energy-conversion model Research object Independent review
Honest boundary. This is a software, simulation and research-data platform — not a reactor, radiation-source, laser, vacuum or megavolt hardware manual. Nothing here is a validated engineering or physical result. See the scientific limitations.

The products

🔬

FusionSim

Transparent p-¹¹B physics engine (cross-sections, reactivity, bremsstrahlung, L0–L2).

Maturity: Research preview Live demo: In browser Evidence: equations + sources
Open FusionSim →
🧫

AlphaTrack AI

Detector-image evidence: calibrated counts, morphology, human review, uncertainty.

Maturity: Synthetic demo Live demo: In browser Scientific validation: Pending (real CR-39) Tests: 27 automated
Try synthetic demo Model card

DirectConvert

Spectrum-aware charged-particle energy conversion: ideal (L0) and loss-aware (L1) stages, exact energy ledger, optimizer, uncertainty. Model, not machine.

Maturity: Interactive model Live demo: In browser Verification: DC-VAL 1–9 (internal)
Run in browser Model card
📚

Research Hub

System of record: projects, sources, equations, datasets, runs, releases and provenance.

Maturity: MVP pages Live demo: Browse Export: RO-Crate / DataCite
Browse the hub →

Verification is not validation

✓ Technical verification (internal, done)

  • Canonical hashing is byte-identical across Python and JavaScript.
  • DirectConvert energy conservation to machine precision (DC-VAL-01).
  • Optimizer cross-checked against brute force; η = n/(n+1) literature case.
  • AlphaTrack plate-level splits with no tile leakage; deterministic runs.
  • Reproducibility: regenerate-from-export matches; tampering is detected.

These confirm the software computes what it claims — not that the physics matches reality.

◷ Scientific validation (required, pending)

  • Real CR-39 benchmark with blinded experts and inter-rater study.
  • Detector-specific calibration and partner-lab protocols.
  • Independent trajectory-code comparison for DC-L2.
  • Reviewed, unit-checked field maps for trajectory studies.
  • External specialist review before any released performance claim.

These are required before any real-world performance or physical claim.

Get the source

The full monorepo — five Python kernels, FastAPI service layer, PostgreSQL migrations, workers, validation, docs and the Next.js app.

Releases, checksum & source → Test report Download .zip