# HBF v750 executed verification

Integrated checks ran on 11–12 September 2026. The release is a static research workbench, not experimentally validated apparatus.

## Software and numerical gates

- `unit-tests.tap`: **470 passed, 0 failed, 0 skipped**.
- `validation.log`: integrated syntax, generated runtime, **17 harness gates**, unit suite, numerical/reference and manifest checks passed. The optional raw Playwright script was not used; supported supervised browser interactions are documented below.
- `numerical-benchmarks.json`: **14/14** current Observatory cases, including four independent Python/SciPy flux arrays. Calculated physical values and acceptance outcomes are unchanged. Application/source identity and the reproduction package byte count reflect the added creator/exporter metadata.
- **24/24** scene, playback-regression and restored-feature groups.
- **41** Sikora–Weller transcription points and **40** Xie radiation fixtures verified.
- Frozen uncertainty/reference example: **22** artifacts verified by standalone replay.
- `pdf-reproduction.txt`: retained semantic HTML and tagged PDF reproduce exactly after installing the pinned build dependencies. Historical document identity is preserved.

Earlier failure logs are retained where useful. A changed science source hash initially made the historical Observatory provenance record stale; new v750 evidence was generated without altering the v748 historical record. Missing PDF dependencies were resolved and the exact-byte check passed. Intermediate unit failures exposed missing new registration symbols in the test harness, a cosmetic random-ID fallback and not-yet-created documentation paths; all were corrected before the final passing suite. The final responsive helper also received explicit label associations after the accessibility contract check caught its implicit-only labels.

## Actual browser observations

Reference environment: supervised Chromium, desktop viewport 1363×936, CPU projected 3D. WebGL is disabled by the browser environment. The original apparatus and fallback renderers load; GPU material/bloom quality is unverified.

| Workflow | Observed outcome |
| --- | --- |
| Installation delivery | Browser checker passed all 40 file checks; downloaded JSON retained as `installation-browser.json`. |
| Seven main lab routes | FusionSim, Isotope, Virtual, Torus, AlphaTrack, DirectConvert and Reactor all loaded and produced their default scientific results/3D state. No lab 404 occurred in the preview. |
| Virtual and Reactor | Scene-first view shows original apparatus, timeline and primary conditional/heating-adjusted result. Opening the drawer reserves scene width. Playback advances stored physical frames; Experiment fields remain available. |
| Reactor heating | Default conditional net is about +254.79 MW; heating-adjusted result about −1719.9 MW. Both stay visible with exploratory scope. These are model outputs, not observed electricity. |
| Torus | Full-window original assembly, camera rail, computed 24 field lines, pause/seek playback and live interpolated coordinates. Field-line progress is arclength, not particle time. |
| Torus field CSV | Metadata populated native controls; manufactured diagnostic import yielded +0.0001 T x residual at both points. Downloaded record retains original bytes, frame, uncertainty and model input hash. |
| Torus typed uncertainty | Default affine example produced standard uncertainty 7.348469 s⁻¹. Editing the source mean discarded its output and disabled stale export. Native parent run/input identity remained explicit. |
| FusionSim | Equal 200 keV inputs were inside the inclusive 200–500 keV benchmark. Manual Run completed. Doubling density produced correct reaction/radiation scaling; comparison rows displayed units and material/identical categories. |
| Isotope | Both original CM-event and product-transport animations are reachable. Default per-event energy/momentum closure and explicit frame/units remained visible. |
| DirectConvert | Keyboard End from Stage 3 selected Stage 8 in table and selector; stage power/current summary updated. Only selected scene annotation is enabled by the model contract. |
| Collector I–V | Synthetic raw/sign/gain/background import produced 4 A signal at all three rows and 40/80/40 W signal power; total boundary powers were 45/90/45 W. Held-out residuals were +0.1 A and +1/+2/+1 W. |
| Detector evaluation | Loaded generated 3-image/192-pixel fixture, locked split, froze actual numerical source, ran batch and built report: one eligible evaluation image, TP=1, FP=FN=0, precision/recall/F1=1. Physical validation remained false. |
| Detector round trip | Downloaded complete workbench, re-imported it, rebuilt the same frozen benchmark. |
| Native export identity | FusionSim native JSON identifies creator release 7.5.0 and the final source/build hashes; the retained 7.4.5 numerical-contract identity remains separate. Historical creator/import preservation is checked by five regression tests. |
| Detector response | Independent manufactured binned-response study calculated successfully. Rank/convergence refusal and calibration rules are exercised in the 17 numerical tests; no current-image species claim is unlocked. |

Actual exported fixtures are included beside this record: `hbf-field-diagnostics.json`, `hbf-collector-holdout-comparison.json`, `alphatrack-benchmark-report.json`, and `alphatrack-workbench.json`. These are generated verification artifacts, not acquired experimental data. File-chooser and download-event API timing occasionally failed during automation; selection was retried after inspecting state. The field download event and FusionSim download click acknowledgement timed out, while the actual files arrived and their bytes/results were verified. `fusionsim-native-v750.json` retains the final native creator metadata. Earlier workflow exports remain dated execution evidence; their source hashes are not retroactively rewritten after the provenance update.

## Narrow layout check

At 390 px iframe width, Torus document width and scroll width were both 390 px. Its stage was 390×457.25 px, with camera and playback controls visible. A narrow Virtual header exposed shrinking labels; its buttons now retain their natural width and scroll horizontally. Virtual reactor also loaded within the same width. Its final mode bar measured 362 px visible width and 507 px scroll width; all eight buttons retained `flex-shrink: 0` and readable natural widths. The corrected screenshot is retained. An earlier document-width measurement call timed out. These are iframe CSS checks, not physical mobile qualification. See `torus-390px.jpg` and `virtual-390px.jpg`.

## Limits

Production Apache/LiteSpeed headers, cache replacement, CSP enforcement and live-host installation were not exercised. Root, nested and encoded-space routes plus mismatch/MIME/HTML-as-JavaScript rejection are checked mechanically; run Installation check after uploading the complete package.

No real mobile device, GPU/WebGL materials, complete accessibility audit or realistic large microscopy throughput was qualified. A responsive iframe fixture under `tools/responsive-preview.html` checks CSS viewport behavior only. Synthetic morphology tests do not measure actual detector error, calibration coverage, fusion performance, sustained burn or net electricity.

Final payload integrity is in `release-manifest.json` and `SHA256SUMS`. The archive builder verifies all extracted bytes and creates the ZIP twice to check deterministic archive bytes. Integrity verifies content identity, not scientific validity or origin authentication.