Skip to content

Paired local stochastic LOVE draws

The source-bound report compares the installed DISQOVER 0.9.14 LOVEr and lov functions with the current Python analytical LOVE engine. Both consume the exact same pollen, parameter and ring rows, two copies of one regional cover vector and explicit local PPE/count draws. The regional vector is the source-compatible deterministic Python REVEALS output; its corresponding R behavior is certified separately by the original oracle's deterministic REVEALS cases. Its source three-decimal percentage rounding leaves the fractional vector summing to 0.99999, which both LOVE implementations preserve.

The R replay only substitutes supplied stochastic samplers and the supplied regional-cover return at the function boundaries. It executes the installed source's ring deposition and local equation. Six paired local PPE/count realizations cover two repeats and the three eligible rings; the other seven rings remain source-invalid. Across Mean, Q10, Q90 and SD, all 120 finite R/Python cells agree within the declared 3e-10 absolute tolerance (largest difference 6.93e-14); all 280 missing cells occupy the same positions. The full R values, Python values and R session are retained. Counts and PPE draws vary by repeat and ring, so this checks more than a deterministic wrapper call.

This is exact-draw numerical parity for one bounded ten-taxon source fixture. It does not imply that R and NumPy produce the same random stream from equal seeds, that the source's historical uncertainty is calibrated, or that the unbounded local estimates are valid vegetation compositions. The separate optimizer comparison evaluates solution quality and achieved objectives without claiming equal search paths.

First build the source-locked R library with the current R reference producer. Replay into a new absent output below artifacts/ with that library and Rscript:

PYTHONDONTWRITEBYTECODE=1 OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 MKL_NUM_THREADS=1 \
  NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 \
  .venv/bin/python docs/report/method-certification/paired-love-draws/replay.py \
  --repository . \
  --library artifacts/jobs/native-r-reproduction/library \
  --rscript /usr/local/bin/Rscript \
  --output artifacts/jobs/paired-love-draws/reproduction

The generated library path is a local run product, not a portable dependency. The producer verifies the exact source archive and installed marker, hashes its input closure and outputs, retains failures under the artifact run, and only writes a passed report after every numeric and missing-position comparison.