# PF Simulation website enhancement validation

Date: 22 September 2026. Scope: the attached website packages, the corrected r34 scientific source, and the satellite/Qiskit additions. This is an implementation review plus scoped numerical execution, not a new full-platform, live-provider or experimental certification.

## Baseline and source integrity

The attached source and upload archives were used as the website baseline. Existing product source matched the working checkout; only the generated next-env.d.ts differed. All 377 selected top-level Python/Markdown and scientific-integration Python files checked against the corrected r34 release archive matched byte-for-byte. No PF scientific source was changed.

## Capabilities verified before implementation

| Capability | Verified description | Boundary retained on the website |
|---|---|---|
| Live satellite tracking | Current-UTC SGP4 propagation using periodically refreshed CelesTrak GP elements and supported TLE inputs; Astropy time/coordinate handling and validity policies | Predicted positions, not a telemetry stream; default element refresh 2 hours; solver runtime affects cadence |
| Quantum satellite workflow | Orbit predictions feed moving-endpoint photon interception, visibility, scheduling and configured state/channel calculations; optional QKD models use declared optics and detectors | Neither measured satellite performance nor actual secret-key exchange |
| Qiskit / Aer | Actual local density-matrix simulation, protocol/component models and seeded Pauli measurements | No IBM hardware execution; configured phase covariance is not inferred from geometry |
| Qiskit Experiments | Optional two-qubit state / one-qubit process tomography and declared count imports; saved-state companions | Code-audited this update; optional dependency absent, so not newly executed; no automatic SPAM calibration |
| Other scientific integrations | The complete 15-family capability registry, with bounded operations, applicability and prerequisites | No claim of full upstream API coverage; vendor and licensed execution remains separately qualified |
| Private AI and infrastructure | Five AI provider protocols, S3-compatible/NAS BYOS, and optional hosted services | BYOK/BYOS data boundaries and operator responsibilities remain explicit |

The detailed integration audit and exported capability matrix accompany the website. MATLAB licensed execution and genuine Swabian recording decoding remain unverified in retained release evidence; neither is presented as an executed Gallery result.

## Fresh numerical evidence

### Satellite tracking and historical replay

- 51 tracking/replay tests passed, with two expected timing-policy warnings.
- Fresh SGP4/WGS72 propagation of historical inputs produced 363 GCRS states (121 per object) over two hours from 16 September 2026, 18:32 UTC.
- A separate matched 120-second J2/DOP853 versus SGP4 comparison produced 39 orbit samples and 18 photon/quantum scenario epochs. Both branch validations passed and saved input/artifact integrity was checked.
- Largest model position differences: ISS 2.666875 m, TECHNOSAT 3.063653 m, HST 3.201510 m. These are model differences, not measured errors.
- Original public orbital data, timestamps, effective settings, source hashes and output arrays are retained. No current network acquisition or physical payload measurement was performed.

### Qiskit Aer phase-correlation study

- 42 actual density-matrix evaluations: two Bell encodings at 21 configured phase correlations.
- 10 tomography runs, nine Pauli settings each, 8,192 simulated shots per setting: 737,280 simulated shots.
- 210/210 exact-state reference/physicality checks passed. Maximum fidelity and purity reference discrepancies: 2.00e-15 and 3.11e-15.
- Exact-state checks are distinct from sampled estimates. Complete complex matrices, unprojected raw reconstruction, seeds and original simulated counts are retained.
- Qiskit 2.5.2 / Aer 0.17.2. No quantum hardware or paid AI service was used.

## Research and R&D value assessment

These benefits are reasoned from the implemented workflows and demonstrated outputs, not measured productivity or commercial claims.

| Research decision | What PF provides | Defensible value |
|---|---|---|
| Which orbit assumptions matter? | Matched initial states, separate propagation branches, photon calculations and retained differences | Supports controlled model comparisons and targeted follow-up studies |
| Which encoding fits the noise model? | Exact and sampled quantum-state outputs across configured phase covariance | Identifies the noise/correlation information needed before choosing an encoding |
| Can a result support a companion study? | Saved full states, clocks, rays and orbit fields with prerequisites and lineage | Reduces ambiguity in transferring model outputs between compatible studies |
| What should be measured next? | Bounded memory inference, sensitivity analysis and recorded-event replay routes | Supports hypothesis comparison and prioritization of characterization work |
| How can proprietary methods evolve? | Reviewable private layers, explicit activation, shared-core versioning and revalidation | Preserves ownership and change history around controlled research configurations |

Academic value is inspectable methods, numerical reproducibility and hypothesis comparison. Corporate R&D value is traceable design alternatives and experiment prioritization. No quantified cost/time saving, superiority, mission accuracy, hardware certification or PF advantage is claimed.

## Website changes and cohesion

The original visual style, AI research workflow, private-stack explanation, BYOK/BYOS boundaries and original Gallery studies are retained. A new Integrations page organizes scientific engines, recorded inputs, AI services and deployment connections. Science explains satellite and Qiskit research use; the homepage introduces both through concise examples. Resources adds original r34 guides. Two new Gallery studies include real data, methods and downloadable reproduction bundles.

The upload archive is a static site for Network Solutions. The application, live tracking and scientific backends still run in a separate PF environment. The public site does not request credentials, stream orbital data or execute private research.

## Upstream documentation cross-check

CelesTrak documents the GP/OMM/TLE data formats at [GP data formats](https://celestrak.org/NORAD/documentation/gp-data-formats.php). Qiskit documents the simulator at [AerSimulator](https://qiskit.github.io/qiskit-aer/stubs/qiskit_aer.AerSimulator.html). The scope statements above are based on PF r34 implementation, not inferred from the larger capabilities of upstream packages.

## Website verification

- Production static export and TypeScript validation passed.
- Nine generated HTML files (six content pages and fallback variants): 372 local links, assets and anchors resolved, with zero missing targets.
- Desktop integration layout inspected; Integrations, Science and Gallery inspected in 375-pixel-wide embedded mobile viewports (360-pixel content width after scrollbar); content width equaled scroll width on each.
- New satellite Gallery modal opened with its accessible title and download link, closed successfully, and Gallery anchor navigation was checked.
- Independent content audits of the satellite, Qiskit and other integration additions passed after three bounded wording corrections.
- Both scientific reproduction bundles passed CRC and byte-comparison checks against their generated evidence; satellite input/artifact hashes verified.

The existing release evidence and original Gallery studies were retained. Their complete validation suites were not rerun merely to update the website.