# HBF Research Preview 1.0 — research and implementation record

**Freeze date:** 2026-08-21  
**Input specification:** HBF Network scientific-platform master implementation specification dated 2026-08-14  
**Audited source release:** v611 / platform 0.6.11  
**Release claim:** internal checks for declared cases; no independent scientific or experimental validation

## 1. Research method and limits

This update used a targeted, reproducible survey of public primary sources and authoritative records. The order of preference was:

1. intergovernmental and government data repositories;
2. standards bodies and official specification publishers;
3. peer-reviewed primary papers and publisher records;
4. university thesis repositories;
5. official software documentation and source repositories;
6. the supplied HBF source tree, fixtures, tests, schemas, registries, and reports.

The survey is broad but is **not described as an exhaustive search of the entire Internet**. Search indexes are incomplete, some publications are paywalled, records change, and no process can establish that every private or unpublished source was found. No credentials, private databases, partner-controlled data, confidential manuscripts, paid datasets, or non-public university records were accessed. No private result, reviewer, institution, partnership, approval, or endorsement has been invented.

Where a source was unavailable for independent transcription, the corresponding model remains disabled, abstains, or is labeled as an unresolved external gate. A citation is not evidence that HBF reproduced a result.

## 2. Main implementation conclusions

The supplied specification and code audit led to ten controlling decisions:

| ADR | Decision | Implementation |
|---|---|---|
| ADR-001 | Six scientific modules, two lessons, three retired identities | `/labs/`, `/learn/`, and Apache redirects |
| ADR-002 | One canonical source/model/evidence registry | `/packages/source-registry/`, `/packages/model-registry/`, `/packages/evidence-registry/` |
| ADR-003 | Versioned RunSpec/ResultSpec and canonical SHA-256 identity | existing `/schemas/` contracts retained and linked from `/evidence/` |
| ADR-004 | Browser workers use the released scientific kernels | canonical labs retain the audited Python kernels; workers run off the main thread |
| ADR-005 | Six separate evidence dimensions | `evidence-states.json`; public `VERIFIED_KERNEL` maturity badge retired |
| ADR-006 | DirectConvert owns conversion modeling | all canonical conversion links converge on `/labs/directconvert/` |
| ADR-007 | AlphaTrack remains synthetic until a plate-level real-data pilot | synthetic fixtures retained; real-data claim explicitly prohibited |
| ADR-008 | Finite-window and right-censoring semantics are mandatory | mirror and torus reduced viewers retain event/final-state reporting |
| ADR-009 | Upstream data is pinned and reviewed before promotion | source registry includes state, role, and unresolved transcription gates |
| ADR-010 | Scientific pages are separated from token/presale paths | canonical shell has no token, wallet, trading, or presale navigation |

## 3. Verification, validation, and uncertainty

### 3.1 Verification and validation

[NASA-STD-7009B](https://standards.nasa.gov/sites/default/files/standards/NASA/B/1/NASA-STD-7009B-Final-3-5-2024.pdf) and the [NASA-HDBK-7009B handbook published in 2026](https://standards.nasa.gov/system/files/tmp/NASA-HDBK-7009B_Final%2002-03-2026.pdf) distinguish verification against requirements from validation against a real-world referent for an intended use and domain. The release therefore does not use an internal test result as a validation label.

The public evidence vector has exactly six dimensions:

- software verification;
- numerical verification;
- literature reproduction;
- independent scientific review;
- experimental validation;
- data maturity.

The machine-readable allowed states and module-specific basis statements are in `/packages/evidence-registry/evidence-states.json`.

### 3.2 Uncertainty

The implementation does not collapse unrelated uncertainty into a decorative band. A complete future result must distinguish, when applicable:

- measurement and calibration uncertainty;
- source/evaluation covariance;
- parameter uncertainty;
- numerical and discretization error;
- model-form or structural uncertainty;
- sampling/Monte Carlo error;
- distribution shift and annotation disagreement;
- unresolved and not-quantified components.

The [BIPM Joint Committee for Guides in Metrology publication index](https://www.bipm.org/en/committees/jc/jcgm/publications) was checked on the freeze date. It lists the current GUM family, including 2026 publications; the release avoids falsely presenting a single interval as universal model validity.

## 4. Nuclear data and p–¹¹B reactivity

### 4.1 Authoritative data repositories

- The [Atomic Mass Data Center / IAEA AME2020 distribution](https://www-nds.iaea.org/amdc/) supplies AME2020 tables and covariance-related files. The Nuclear Data Explorer pins the relevant mass records instead of scraping an unversioned web display.
- [IAEA EXFOR](https://www-nds.iaea.org/exfor/) is the authoritative experimental nuclear-reaction compilation used for accession-level source discovery. “EXFOR contains a reaction” does not mean a transformation or combined curve has been independently reviewed.
- [NNDC NuDat 3](https://www.nndc.bnl.gov/nudat3/) is retained for evaluated structure and decay context.
- [NIST ASTAR/PSTAR/ESTAR](https://www.nist.gov/pml/stopping-power-range-tables-electrons-protons-and-helium-ions) is retained as a government reference for stopping-power/range context, not as a substitute for detector-specific calibration.

### 4.2 Reactivity sources and boundaries

The source registry preserves peer-review and transcription state rather than ranking sources by recency alone:

- Sikora and Weller (2016), [doi:10.1007/s10894-016-0069-y](https://doi.org/10.1007/s10894-016-0069-y), supplies an open peer-reviewed cross-section/rate evaluation. The included source record preserves the measured energy range, channel handling, statistical component, stated systematic component, transformations, and hashes. A known narrow-resonance omission bounds low-temperature use; the record does not hide it.
- Nevins and Swain (2000), [doi:10.1088/0029-5515/40/4/310](https://doi.org/10.1088/0029-5515/40/4/310), remains a named Maxwellian reactivity comparison with its piecewise domain. Coefficients must not be guessed if the source cannot be checked.
- Tentori and Belloni (2023), [doi:10.1088/1741-4326/acda4b](https://doi.org/10.1088/1741-4326/acda4b), is retained as a modern reactivity/radiation comparison source, not silently made the default.
- The 2026 p–¹¹B parameterization at [arXiv:2601.00241](https://arxiv.org/abs/2601.00241) is a preprint and stays non-default pending transcription and independent review.

FusionSim’s “partial literature reproduction” state means selected source checkpoints and declared refinement fixtures pass. It does not mean an independent laboratory reproduced the software or an experiment validated the plasma model.

## 5. CR-39 and AlphaTrack

The public evidence does not support a real-data accuracy claim. Relevant primary literature demonstrates why a controlled protocol is necessary:

- Lahmann et al. (2020), [doi:10.1063/5.0004129](https://www.osti.gov/biblio/1801151), reports CR-39 response behavior for ICF-relevant ions and shows that particle species, energy, filtering, etch, and calibration matter.
- Schollmeier et al. (2023), [doi:10.1038/s41598-023-45208-x](https://www.nature.com/articles/s41598-023-45208-x), addresses multi-MeV, multi-ion CR-39 spectrum differentiation and cites calibration and automated-image-analysis precedents.
- Fews (1992), [doi:10.1016/0168-583X(92)95367-Z](https://doi.org/10.1016/0168-583X(92)95367-Z), is retained as an early automated etched-track image-analysis precedent.

Consequences for WP2:

1. split by plate and batch before modeling;
2. freeze acquisition, etch, microscope, focus, image, and annotation protocols;
3. use licensed partner-controlled data and local-first processing;
4. blind the frozen model evaluation;
5. preregister primary metric, interval, subgroup, abstention, and shift analyses;
6. preserve human/model disagreement and deviations;
7. release favorable, unfavorable, or inconclusive results;
8. limit every conclusion to the tested detector and acquisition domain.

The bundled AlphaTrack images are synthetic. External real CR-39 data is not included.

## 6. Orbit, mirror, toroidal, and configuration modeling

The mirror and torus modules are deliberately reduced. Their visual quality does not raise their scientific fidelity.

- Qin et al. (2013), [doi:10.1063/1.4818428](https://doi.org/10.1063/1.4818428), supplies numerical context for the Boris particle pusher. HBF’s convergence statement applies only to bundled declared cases.
- V. Player’s 2024 UC Irvine dissertation, [Orbit Space Analysis of Resonant Beam-Driven Ion Acceleration](https://escholarship.org/uc/item/5r11w74x), and the related orbit-space literature were reviewed as context, not adopted as a benchmark without a frozen identical mathematical problem.
- Bradley Scott Nicks Jr.’s 2020 dissertation, [Novel Avenues of Wakefield Acceleration: Fusion Plasmas and Cancer Therapy](https://escholarship.org/uc/item/0v88226r), includes a bounded p–¹¹B plasma analogy and explicitly cautions that parameter changes matter. It does not validate HBF’s torus or mirror scene.
- Joseph Jerkins’ 2021 MIT thesis, [Fusion Fuels](https://dspace.mit.edu/handle/1721.1/143405), was retained for alternative-fuel radiation and activation-risk context; it is not implemented as an HBF kernel.

Official open-source projects were reviewed to clarify model boundaries:

- [FreeGS](https://github.com/freegs-plasma/freegs) solves Grad–Shafranov equilibrium problems and describes itself as work in progress.
- [DESC](https://github.com/PlasmaControl/DESC) provides equilibrium/optimization software.
- [STELLOPT](https://github.com/PrincetonUniversity/STELLOPT) provides stellarator equilibrium and optimization tooling.

These repositories are not bundled, endorsed, or claimed as independent comparisons. They demonstrate why an analytic orbit field cannot be relabeled as equilibrium, stability, transport, or whole-device modeling.

## 7. Direct conversion

- Barr and Moir’s government technical report, [OSTI 7341986](https://www.osti.gov/biblio/7341986), is retained for direct-conversion concepts and engineering limitations.
- Rax et al. (2025), [doi:10.1103/PRXEnergy.4.013007](https://doi.org/10.1103/PRXEnergy.4.013007), is retained for physical limitations of adiabatic direct energy conversion.

DirectConvert therefore exposes an ideal bound and assumption-driven reduced stages. Default loss factors are not measured component efficiencies. The release makes no hardware, plant, voltage-holding, reliability, thermal, or net-electric output claim.

## 8. Accessibility, security, and reproducibility standards

### 8.1 Accessibility

[WCAG 2.2](https://www.w3.org/TR/WCAG22/) is the normative accessibility target. The canonical shell provides:

- a keyboard skip link and one global primary navigation;
- visible focus with sufficient offset;
- 44 CSS-pixel minimum menu targets;
- reflow at 320 CSS pixels for portfolio and evidence pages;
- reduced-motion handling;
- semantic headings, landmarks, tables, status messages, and non-color state text;
- static content and explicit failure messages before optional scientific runtimes load.

Legacy lab-specific controls inherit the audited accessibility fixes and are checked by the packaged static tests. Browser/assistive-technology certification remains a human release gate.

### 8.2 Application and supply-chain security

- [OWASP ASVS 5.0](https://owasp.org/www-project-application-security-verification-standard/) is used as the current Level 1-oriented web verification reference.
- [SLSA build provenance 1.2](https://slsa.dev/spec/v1.2/build-provenance) informs the in-toto provenance statement.
- Canonical scientific directories enforce CSP, framing protection, MIME controls, referrer policy, and vendor-first dependencies. Pyodide is pinned at 0.26.4 and selected self-hosted-first, but its binary distribution is not bundled in this archive; the identical pinned CDN remains a functional fallback. Full self-only runtime operation is therefore an explicit gate, not a completed claim.

### 8.3 Research-object metadata

- The specification named RO-Crate 1.2. The current official release is [RO-Crate 1.3](https://www.researchobject.org/ro-crate/specification/1.3/index.html), published 2026-06-22, so the release metadata was upgraded to 1.3.
- [DataCite Metadata Schema 4.7](https://schema.datacite.org/meta/kernel-4.7/) is current. The included record is explicitly a draft because no DOI is registered. A fake placeholder DOI was not created.
- [Citation File Format 1.2](https://citation-file-format.github.io/) metadata is included as `CITATION.cff`.

## 9. Resilience and deployment findings

The supplied archive’s internal release checks passed before modification (56/56 unified gates and 5/5 quick external-validation package checks). A read-only check of the then-live site found six deployment mismatches: several v611 files and version/status responses were missing or stale. The upload package includes the missing files and a deployment verifier; only a post-upload run can establish that the host received them.

Scientific pages remain readable when JavaScript, WebGL, or Pyodide is unavailable. 3D modules retain a 2D/static fallback. Workers prefer local pinned assets and surface runtime failure rather than replacing a failed scientific result with a plausible number.

## 10. Remaining external gates

The following cannot be completed by software development or Internet research alone:

- independent scientific review by a named, conflict-disclosing expert;
- independent implementation and blind comparison for WP1 or WP3;
- licensed, blinded real CR-39 data, expert annotations, and preregistered analysis for WP2;
- institutional agreement on scope, data, publication, IP, safety, names/logos, and negative-result release;
- DOI registration and archive deposition;
- post-upload host verification;
- browser, assistive-technology, and human usability review;
- pinned Pyodide binary vendoring if a fully self-only runtime is required.

These remain visible in `release-manifest.json` and must not be converted to “passed” without exact evidence.

## 11. Release interpretation

This package is partnership-outreach ready in the narrow sense that it supplies runnable artifacts, bounded research questions, protocols, evidence states, and honest gates. It is **not** evidence that the scientific models are independently approved or experimentally valid. The preferred first request is one independent p–¹¹B reactivity reproduction. The AlphaTrack pilot is the differentiated alternative when a qualified detector laboratory and governed real dataset are available.