Trust the trail,
not the badge.

What was checked

Run timestamp (UTC): 2026-09-22T02:47:27.537472+00:00. R reads unchanged inputs with haven; Python reads them with pandas. Each language independently calculates every case before comparing structured values. Shared input and analysis specifications are intentional; shared computed results are not used.

Download analysis evidence (JSON)

Package quality / risk

The existing R package QC tool has not been represented as successfully integrated. Its exact permitted local entry point has not been confirmed. Package-risk checks remain not run. The local import adapter requires an explicit tool identity, version, timestamp, package/version and individual checks; unsupported status values are rejected. A passing numeric comparison does not establish package quality or regulatory acceptance.

Presentation testing

Browser evidence belongs to a separate layer: responsive layouts, filtering, clipboard, actual downloads, keyboard use and network behavior. Local screenshots and logs live outside public assets. The analysis-run record above precedes these separate browser executions.

Data provenance

One public source, pinned and attributed

Source: CDISC SDTM/ADaM Pilot Project. Commit 667511d4b183871d74392ba691c935c38d431d39. Retrieved in this local run on 2026-09-21 UTC. CDISC supplies these as clinical test data. “Copilot” is not a separate verified dataset identity in this release.

The upstream Terms of Use govern the inputs; there is no blanket permissive software license for the data. They require attribution, prohibit misleading origin claims, prohibit distribution for a fee and contain restrictions on altering data. The raw XPT files are kept unchanged outside the website. Download links here contain newly calculated aggregate results, not repackaged participant datasets. This release shares attributed educational aggregate analyses; the upstream terms continue to govern the unchanged source data.

No employer datasets, private study results, or invented clinical endpoints are used. No additional synthetic dataset is mixed into these cases. The small known-answer numerical fixtures are separate artificial test vectors.

Download provenance manifest · Source data dictionary (define.xml)

Coverage before chart choice

Oncology response, waterfall and swimmer plots are not part of the CDISC Pilot cases: no suitable response endpoint was established. The separate Figure Library uses fictional teaching inputs for these templates. The supplied time-to-event endpoint is dermatologic safety, not overall survival. Participant profiles are deferred to keep this version focused on aggregate interpretation.

Reproducibility

The environment that actually ran

{
  "python": {
    "python": "3.13.13",
    "packages": {
      "numpy": "2.4.6",
      "pandas": "3.0.3",
      "scipy": "1.17.1"
    },
    "fixtures": "pass",
    "seed": "not applicable: deterministic"
  },
  "r": {
    "R": "4.6.0",
    "packages": {
      "haven": "2.5.5",
      "jsonlite": "2.0.0",
      "survival": "3.8.6",
      "ggplot2": "4.0.3"
    },
    "fixtures": "pass",
    "seed": "not applicable: deterministic"
  }
}

No new R, Python, npm or CDN dependency was installed. The local worktree reuses the existing frontend dependencies. R: haven (MIT), jsonlite (MIT), survival (LGPL ≥2), ggplot2 (MIT). Python: pandas / NumPy / SciPy (BSD family); plotting exports use the installed matplotlib. Versions are recorded, not claimed to be the newest available.

Official references: haven, ggplot2, survfit, SciPy Welch test. pharmaverse candidates admiral, rtables, tern, gtsummary and ggsurvfit were considered; this descriptive ADaM-input workflow does not need a new derivation or reporting dependency. They are not all installed here and no regulatory validation is implied.

Source ↔ result association

Executed source hashes and exact aggregate result hashes are recorded together. The page embeds source from the same files distributed for download.

Architecture decision

Static delivery is enough for these questions.

R/Python execution takes place before publication. Core scripts, fonts and data are same-origin. Lab pages have a restrictive content security policy, no analytics integration and no outbound data submission. A static server is required to fetch JSON; opening HTML with file:// is not supported.