inst/VALIDATION.md
In OECDdata: Fast and Friendly Access to OECD Data

OECDdata 0.2.5 validation record

Tested on 20 October 2026, Windows 11 x64, R 4.6.1 (ucrt). The discovered package defects are fixed. Complete CRAN release sign-off still requires the outstanding checks below; this record does not claim CRAN acceptance or completed GitHub Actions runs.

| Check | Result | |---|---| | Offline testthat suite | 347 passing assertions across 74 executed cases; zero failures, errors or warnings. One additional live case is skipped by default. | | New regressions | 11 cases covering 36 assertions; the original defects were reproduced before repair. | | R CMD build | Successful source archive with the rendered vignette. | | Windows R CMD check --as-cran --no-manual | Successful with remote incoming and clock verification disabled for the restricted environment; see the recorded log. | | checktor 0.2.0 | No policy, robustness or opinion findings; no suppressed findings. All offline checks, including spelling with Hunspell 1.7.4, executed. | | checktor URL liveness | Requested in the initial audit, but could not execute because R's outbound network access is blocked. Not counted as a pass. | | Installed package | The diagnostic runner unloads the development namespace and loads the newly installed archive before checking its public API. | | Real libcurl transport | Local HTTP integration passed gzip decoding, redirects, Retry-After handling, HTTP 204, HTML rejection, request headers and size enforcement. | | Real OECD responses | Fresh REST v1 and v2 requests via PowerShell returned identical 32-row datasets, including values and attributes. | | Cached integration | All seven assertions in the opt-in live test passed in strict offline mode against checksum-verified responses downloaded on 16 October 2026. | | API compatibility | The intended argument/default mapping to eurodata 1.8.0 passed. | | Offline coverage | 87.7% measured by covr 3.6.5. | | Parser benchmark | 120,000 synthetic rows; native and base parser results verified. Timings are retained in the diagnostic log. | | GitHub configuration | All three workflow YAML files parsed; jobs have not run on GitHub. | | Current CRAN name | No case-insensitive OECDdata match in the CRAN package index downloaded on 20 October 2026. This is not name reservation or approval. |

The fresh data requests used format=csvfile with PowerShell's HTTP client. The package's parser used the verified cached XML 2.1 metadata. This is fresh REST/data-format evidence, not an end-to-end fresh run of the package's R HTTPS transport. The recorded CSV 2 response was also exercised by cache replay.

Pin latest dataflow versions before asking OECD for referencepartial references. The former wildcard request was observed returning HTTP 400; dataflow-only latest resolution and the exact-version request both succeeded.
Return names for non-enumerated components such as TIME_PERIOD.
Honour qualified component references over default dataset options and reject conflicting explicit contexts.
Describe actual code availability even when no codelists are enumerated.
Validate cache status/time fields and apply size limits to cache hits.
Classify libcurl size failures correctly and avoid retrying them.
Treat XML false and 0 exclusions conservatively; do not advertise excluded values as observed codes.
Reject invalid or duplicate dimension positions and reserved component IDs.
Validate dataflow and DSD identities in CSV responses, including conflicting identity columns.
Stage development source before building so Windows logs under .qa cannot cause R CMD build to fail while copying the tree.
Verify the installed namespace and actual PDF output in release tooling; fail explicitly when requested URL checks cannot execute.

PDF manual: the initial R CMD check --as-cran summary incorrectly reported the manual as OK. Rdlatex.log recorded a failed MiKTeX texify command and no manual PDF was produced. MiKTeX cannot initialize its configuration under this session's permissions. A portable TeX Live attempt also failed on inaccessible ancestor-path information. Neither is a proven package documentation error. The updated scripts explicitly require a nonempty PDF; the full-manual GitHub job does the same.
Remote checks: run CRAN incoming checks and checktor with OECDTEST_CHECKTOR_URLS=1 using working R network access. The OECD service homepage was reachable through PowerShell.
Other platforms: execute the configured Windows/macOS release and Linux release, oldrel, devel and R 4.1.3 jobs. WSL was inaccessible locally.
Fresh R integration: run the manual live workflow or the opt-in test using the package's own transport.
Publication: this directory arrived without Git metadata or a specified GitHub remote. No repository was created or pushed, and nothing was submitted to CRAN. The maintainer still confirms submission and licensing declarations.

Rscript --vanilla tools/checktor.R .
Rscript --vanilla tools/check.R .
Rscript --vanilla tools/diagnostics.R .

The full check needs a working LaTeX installation; checktor spelling needs aspell or hunspell. Strict checktor execution is available with OECDTEST_CHECKTOR_STRICT=1. A partial manual-free check requires the explicit OECDDATA_NO_MANUAL=true switch and is labelled accordingly.

Local evidence is kept in .qa/, excluded from Git and the CRAN package: final-tests.csv, finish-audit-final.log, http-integration-final.log, checktor-spelling-final.log, replay-live.log, and release-diagnostics/ (stage results, build, installation, coverage, lint, benchmark and R check logs). The lint audit reported style findings only; there were no lint warnings or errors. Coverage and benchmark results are recorded without treating them as proof of correctness or network performance.

Checktor is an additional tool, separate from the CRAN policy and R's package checks. Its severity and skipped-check reporting are preserved in the local reports.