# HBF 7.5.0 implementation and research boundaries

The v750 work extends the latest uploaded Hostinger site while retaining the existing lab routes, apparatus graphics and original Torus Observatory. The requirement matrix covers all **96 tracked items**: nine reported defects, twenty ranked features, thirty contracts, ten release gates, eighteen data requirements and nine proposed milestone periods. It retains all **71 original heading segments** for traceability.

This document records integrated source review and executed verification on 11–12 September 2026. “Implemented” describes the scoped software present in the checkout; it is not a claim that a release, browser, independent solver, measurement or physical validation gate passed. Actual execution evidence is in `docs/v750-verification/`: 470 tests, 17 architecture/release gates, 14 Observatory cases, 24 scene/playback groups, source fixtures, replay and exact retained-PDF reproduction passed. Browser coverage is explicitly bounded there. The matrix separates partial work, missing data, external validation and deferred specialist/service work so a future session can continue from the same state.

## Changes in this increment

- **Release and calculation correctness:** Build-owned checker identity and strict asset/MIME/version checks; stable equal-temperature and inclusive-boundary handling; current creator identity distinct from historical kernel/schema/data versions; revision-owned queue/run/result state; quantity units; semantic, tolerance-aware A/B comparisons that retain raw differences.
- **Connected heating:** Reactor and Virtual use the existing heating-closure equations with typed wall-plug efficiency and already-included heating credit. Conditional and adjusted net are shown together. Deposited alpha energy and heater input are not credited twice, and no unmeasured recovery credit is added.
- **Joint particle handoff:** The existing particle-flux contract gains paired position, direction, energy, time, species, normalization and explicit fate accounting with legacy compatibility. Torus can hand off its declared scenario to DirectConvert. The handoff is a conditional source record, not an independently predicted escaping reactor population.
- **Collector and fields:** Raw collector CSV plus calibration, sign, background, source, geometry and electrical-boundary metadata; frozen prediction and session-holdout residual tools; point-field diagnostic CSV, named coordinate frames and covariance; current-run comparison; reference-job contracts; and a dimensional scaling worksheet. No actual apparatus measurement is bundled as a result of this work.
- **Scientific studies:** Typed settings and documented file formats retain an advanced JSON representation. Correlated affine uncertainty preserves shared draws, a transparent one-step Gaussian planner reports its assumptions, scoped evidence/claim manifests retain dependencies, and the standalone manufactured replay includes an independent analytic reference. These are not a calibrated end-to-end uncertainty chain or a validated active-learning system.
- **Detector infrastructure:** Immutable plate/calibration/image/annotation records, raw/pixel duplicate detection, strict plate/session/site split locks, frozen method identity, centroid benchmark/adjudication and bounded image batch/tile kernels. Dedicated UI/response finishing status is recorded in the matrix and finalization list; real calibrated images, trusted lock chronology and response measurements remain data gates.
- **Presentation:** Scene-first layouts with remembered panels and persistent primary results; synchronized selected-stage labels/table/keyboard controls in DirectConvert; scientific homepage numbers are displayed at their true values from first paint. The 3D appearance remains tied to the existing apparatus and stated model scope.

## Harness engineering

`AGENTS.md` provides a short repository map and points to durable plans and checkpoints. `tools/harness_check.mjs` mechanically checks generated release identity, numerical-module boundaries, prohibited hidden browser/network/unseeded-random dependencies in the reviewed core set, fixed first-paint scientific figures and a valid evidence path/status for every requirement. Numerical and event regressions live beside the implementation; `tools/validate.sh` remains the integrated release entry point.

These checks reduce avoidable integration errors and preserve a reproducible continuation point. They do not verify scientific truth merely because code passes. A failed check must remain visible and be corrected from its concrete cause; unresolved data or experimental dependencies stay explicit. Do not promote an agent’s proposed tests or historical counts to final integrated pass results.

## Defect and ranked-feature outcomes

| ID | Status | Implemented scope / remaining work |
| --- | --- | --- |
| B01 | implemented | The source correction generates the installation checker from build-owned release identity, validates requested assets and rejects version/MIME/HTML mismatch. This software status does not waive final regenerated-manifest, root/subfolder and mixed-release execution gates. |
| B02 | implemented | Integrated stable mass weighting and exact equal-temperature identity use one documented inclusive numeric range; boundary and unequal/extreme-mass regression definitions exist. This status records source implementation, not a newly executed test result. |
| B03 | implemented | Run/research/export creators use current app release and generated build/source identity separately from historical schema/kernel/data identity. Archived records keep their original identity. Final generated hashes/manifest and archive reproduction remain release gates. |
| B04 | implemented | Revision-owned queue/manual completion and guarded async commits are integrated; helper labels read actual run completion. Regression definitions cover interrupted debounce, stale jobs and invalid paths; final rendered rapid-edit workflows still require root execution. |
| B05 | implemented | Typed units/definitions/null meaning and quantity-aware formatting are integrated for the reviewed scientific quantities, including all changed FusionSim A/B paths. Unknown ambiguous definitions fail rather than invent units; final UI/export smoke remains required. |
| B06 | implemented | Scientific comparison retains raw differences, explicit material/tolerance/identical categories and scoped semantic alias deduplication. Density-invariant ratio regression definitions exist. Numeric display tolerances are distinct from physical uncertainty. |
| B07 | implemented | DirectConvert shows only the selected stage annotation and synchronizes stage dropdown, table focus/click and Up/Down/Home/End selection with the scene. Selection is display state, not a scientific reassignment. Final rendered narrow-width/keyboard checks remain required. |
| B08 | implemented | Scene-first initial presentation, remembered pane choice, drawer-aware viewport resizing and persistent primary summary are integrated without replacing original apparatus geometry. Final browser camera-fit, keyboard-focus and narrow-layout evidence remain a separate gate. |
| B09 | partial | Typed shared studies, main heating, dedicated collector/field metadata, detector registry and Torus supplementary tools are integrated and exercised in the browser. Advanced JSON remains for complex job/response/annotation objects. Arbitrary instrument column/unit conversion remains unsupported. |
| F01 | implemented | Build-owned release/checker/source identity, revision state, runtime assets and complete release-manifest checks pass in the integrated software. Root/nested/encoded-path and mixed-release/MIME/HTML rejection are exercised in the unit harness. Actual live-host cache/header behavior still requires deployment acceptance. |
| F02 | implemented | Main Reactor and Virtual scenarios now invoke existing heatingClosure, include typed heating inputs in scientific identity, and show conditional and adjusted net together with no extra heat recovery credit. The output remains conditional, not measured net electricity. |
| F03 | partial | Immutable plate/calibration/image/annotation registry, acquisition metadata UI, strict plate/session/site/raw/pixel split locks and frozen methods are integrated. Real licensed calibrated plates, independently identified acquisition sites and trusted evaluation chronology remain absent. |
| F04 | partial | Baseline morphometry evaluation and calibration/domain/overlap abstention exist. True watershed/learned comparison, instance-mask split/merge metrics, calibrated intervals and real-data robustness are not implemented or demonstrated. |
| F05 | partial | Joint @2 contract preserves paired phase-space samples/fates and supports @1; actual Torus scenario handoff reaches the dedicated DirectConvert workbench. Isotope/mirror escape-source production, CSV/openPMD serializers and general geometry transport remain absent. |
| F06 | partial | Calibrated raw I–V CSV import preserves raw bytes/hashes, signs, paired background, declared electrical boundary and shared calibration covariance. Frozen-prediction/session-holdout current and power residuals are integrated. No real independently calibrated apparatus sweep or trusted preregistration is supplied. |
| F07 | partial | Joint-normal covariance propagation, shared draws, domain refusal and separate numerical/model/omission categories are integrated as affine study tools. No calibrated physical source–detector–collector uncertainty chain or complete discrepancy model is established. |
| F08 | partial | Standalone replay with dependency hashes and separately authored analytic Python uniform-field reference is bundled. A full external electrode/equilibrium adapter campaign, independent operator and review remain required; analytic matching is not higher-fidelity apparatus validation. |
| F09 | partial | Measured Cartesian B-vector CSV import, named frame and calibration metadata, shared covariance, outside-domain refusal and current-model residual views are integrated for legacy fields and Torus Observatory. Reference-job contracts exist. Actual calibrated map/position/geometry records and withheld measurements remain absent. |
| F10 | partial | A binned forward response μ = T R φ + b, detection efficiency/losses, rank/proportional-column diagnostics, bounded nonnegative Poisson EM, quadratic-penalty sensitivity and residual/deviance study is integrated. Current-image inverse output and physical species claims remain unavailable. Actual independent response/exposure/background calibration, uncertainty coverage and held-out physical performance are absent. |
| F11 | partial | Transparent one-step linear-Gaussian information/cost ranking with random and space-filling baselines is integrated. There is no validated surrogate, sequential campaign or realized improvement claim. |
| F12 | partial | Existing AME/LiveChart/p–B sources, imported cross-section/distribution checks, covariance exercises and radiation comparisons remain; shared normalization is added. Full evaluated covariance, differential/angular records, parser coverage and permission review are incomplete. |
| F13 | partial | Dedicated UI freezes actual source/settings, runs/cancels bounded imported pixel batches and builds locked adjudicated centroid reports. Browser manufactured three-image batch and exported-bundle re-import passed. Realistic large microscopy memory/time, TIFF/OME streaming, mosaics and mask topology require further work. |
| F14 | partial | Observable/scope evidence manifests, V0–V5 semantic gates and component evidence studies are integrated. Full convenient component-to-claim authoring and actual independently accepted measurement evidence remain incomplete. |
| F15 | deferred | No time-dependent species/ash inventory or maintained nonthermal evolution is implemented. Requires a supervised reduced-physics work package; retain current fixed-source limitations. |
| F16 | external-validation | Uniform-field reference request/replay exists, but no measured low-current collector baseline or geometry-specific self-consistent field/particle solver is integrated. Select and review one external solver only after regime/residual evidence justifies it. |
| F17 | implemented | Typed dimensional inputs calculate E/\|q\|, qV/E, angular spread, gyroradius/geometry, L/mean-free-path, conventional and generalized perveance and retain the surface record. Results state that matched ratios do not establish target-regime behavior; material/current extrapolation still needs specialist and measurement evidence. |
| F18 | deferred | Local evidence/permission metadata and retained research records are available. There is no multi-institution authenticated storage, enforced embargo or shared review service; build only after actual institutions agree requirements. |
| F19 | deferred | Existing fixed-boundary equilibrium remains explicitly scoped. No new free-boundary, stability, guiding-centre or collisional solver is claimed; specialist extension must answer a selected experiment question. |
| F20 | partial | Existing 3D apparatus is retained; scene-first layout, synchronized selected-stage labels, compact primary results and typed studies improve access. Scientific homepage numbers are final from first paint. Complete browser/accessibility qualification and new focused teaching modules remain incomplete. |

## Source and data acquisition status

The repository already contains the Sikora–Weller transcription, a twelve-nuclide AME2020 mass slice, a nine-nuclide LiveChart ground-state snapshot dated 4 September 2026, and the Xie radiation comparison/fixtures. They retain their historical source identity; this update does not describe them as newly measured or as complete evaluated datasets.

Selected EXFOR/IBANDL accessions, evaluated mass covariance/index mapping, known detector response exposures, partner images and labels, electrical/field calibration certificates, surface characterization, and actual held-out collector/field campaigns remain missing. Source-register entries record what was actually inspected and whether source contents were retrieved. A landing page, publication or API description is not evidence that its underlying private, licensed or accession-level dataset was acquired.

The accompanying deep-research report and source register cover relevant public primary literature, government/evaluated data, public industry-affiliated work and reproducibility guidance. NIST Standard Reference Data must not be assumed unrestricted merely because NIST is governmental. OPEN-ADAS incorporation/redistribution requires permission for the relevant use. A paper does not automatically license raw images, weights, code or private measurements. No credentials, private records or licence permissions were invented.

## Model and evidence limits

The connected plant figures remain conditional outputs of included channels, empirical source models and assumed efficiencies. They do not demonstrate sustained p–¹¹B fusion, practical direct conversion, ignition, confinement or measured net electricity. A negative adjusted heating-closure result must remain visible rather than being hidden by the conditional headline.

The existing fourteen-case Observatory record and four-array Python reproduction are historical numerical verification. A manufactured independent analytic accelerating-gap comparison has a narrow software/reference scope. Neither proves a physical collector, calibrated detector, plasma performance, radiation safety or a working power plant. This release creates no V3–V5 physical evidence by itself.

Detector hashes detect changed bytes and declared duplication; they cannot authenticate ownership, acquisition conditions, annotators or timestamp chronology. The strict cross-site split needs genuinely distinct sites and sessions. A one-site pilot needs a separately named and narrower protocol. Pixel morphology does not establish species or energy without a calibrated and identifiable response model.

Collector current and power are correlated observables. The first-order CSV adapter assumes the documented calibration/noise/background relationships; other correlations need a different adapter or full covariance input. Field comparisons require the same declared frame and an appropriate domain. Similarity ratios do not justify MeV, current-density, surface or reactor-scale extrapolation.

Deferred work includes autonomous reactor control, a chatbot presented as a scientific validator, a private-data marketplace, a new general-purpose PIC/MHD solver, full reactor cost forecasting and unsupported high-fidelity expansion. Maintained nonthermal/ash kinetics, free-boundary/stability, self-consistent collector field/space-charge coupling and multi-institution services require bounded specialist and experimental projects.

## Executed release checks and remaining qualification

The integrated validation command completed successfully: 470 tests, 17 harness gates, 14 current Observatory benchmarks, 24 scene/playback/restoration groups, 41 cross-section transcription points, 40 radiation fixtures, 22 replay artifacts and exact retained HTML/PDF reproduction. The initial benchmark mismatch was source identity only. The final native-export provenance correction changed two source hashes and reproduction ZIP length; calculated physical values and benchmark pass criteria did not change. Earlier missing PDF dependencies were restored at their pinned versions.

Actual supervised Chromium checks cover the seven lab entrypoints, CPU 3D, original reactor/Torus playback, equal-200-keV classification, density comparison units, selected-stage keyboard behavior, typed uncertainty invalidation, manufactured field and collector imports, and detector lock/freeze/batch/benchmark/export/re-import. See `docs/v750-verification/README.md` for exact observations and limitations. Source test fixtures are not experimental evidence.

Production Apache/LiteSpeed headers/cache behavior, actual GPU materials, physical mobile devices, comprehensive accessibility, realistic microscopy throughput and independently reviewed measurement campaigns remain qualification work. The installation checker and portable-path/mismatch regression harness support deployment acceptance but do not establish that a live host has been updated.

The full requirement matrix has 24 implemented software outcomes, 40 partial outcomes, 18 data requirements, 7 external-validation requirements and 7 deferred requirements. An implemented software outcome does not complete an experimental programme period. Partial outcomes remain listed so future work can resume without substituting invented data.

Review-time evidence hashes are dated snapshots. Use the final release manifest and SHA256SUMS for delivered bytes; generated integrity records cannot hash themselves. No external message, application, funding request or deployment was performed.

## Proposed programme and resumption

`V750-PROGRAMME-PROPOSAL.md` preserves the existing collaboration page’s four phases and adds the supplied nine-period 24-month milestone proposal, first 30 days, stop/narrow rules and illustrative budget assumptions. None is a record of secured funding, access, partnership, institutional approval or completed research months. No external message or application was sent.

Resume from `RESUME-CHECKPOINT.md`, the original user development plan, this matrix and recorded verification failures; do not silently restart a new product or replace retained graphics. The JSON evidence-file hashes are a dated static-review snapshot. Regenerate them after source integration or retain their date explicitly. Historical fixtures and records must retain their original scientific identities.