LandClim II author-example correspondence

The row-level ledger, generated report, and numerical summary compare the installed LRA publication profile with an independent exact author-R replay on the 17 author sites in grids 512 and 562. All 14,074 supported fractional cover and standard-error values pass the declared source tolerances. Maximum absolute residuals are 1.83881e-13 for means and 7.15331e-14 for standard errors. The 25-bin source profile has unavailable site-bins, which the report counts separately.

The source and output manifest records the producer and the original retained run. The original CSV was newline-normalized for the repository; the generated Markdown and JSON retain their original bytes. The current producer writes LF line endings directly. Replaying from the retained original candidate and author-R reference reproduces all three governed files byte for byte. An independent R/native execution produced the same row-level CSV and generated Markdown. Its JSON differs only in the reference receipt digest because that R execution has a new receipt.

Replay the governed correspondence from the original candidate and author-R reference retained in the private source store:

OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 MKL_NUM_THREADS=1 \
  NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 \
  PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=packages/bijux-pollenomics-dev/src \
  .venv/bin/python -m bijux_pollenomics_dev.essd_reproduction.author_example_correspondence \
  --run ../bijux-pollenomics-data/landclim/two-grid-process/run/run001 \
  --reference ../bijux-pollenomics-data/landclim/author-reference/reference \
  --output artifacts/jobs/author-example-correspondence/replay

This is two author grids, not the complete historical paper input set or a vegetation validation.