Conditional uncertainty on author example sites

The source-bound result uses the released LandClim II author example's grid 512, 225 BP midpoint window. The six sites with source pollen in that window are BROADOWN, HANGINGS, HARESDOWN, LOBBSBOG, WINDMILL and WINNEYS. Their source radius values span 10–200; the calculation passes those static values as metres to the source-compatible lake deposition function. The raw-to-prepared replay has already reproduced all six selected prepared files byte-for-byte from source raw rows. The report hashes the selected author files, 31-target PPE and covariance tables, crosswalk, site list, lineage receipt and calculation code. Its SHA-256 is 415185927514de10a71b6a7ba6c26d7bc1fd3ed8cc140760ec2e3e9f58f1c243.

The ten taxa with positive prepared counts at every selected site form an exploratory closed subcomposition. Their site totals are 289, 293, 2,223, 2,786, 4,251 and 562 counts, respectively: 10,404 selected counts out of 10,747 on the author's full 31-target axis. Every selected site/taxon cell is positive; no formatter-inserted zero is represented as an observed absence. This selection was made after reading the example counts. Its denominator is the sum of these ten taxa within each site/window, not the source's complete pollen-counting denominator or total vegetation cover.

The reviewed fixed-site profile treats each site's selected count total as a conditional multinomial draw and propagates the source PPE covariance. The joint draw identities retain 1,024 paired site bootstrap rows, one multinomial count draw per original site and one shared positive PPE vector per replicate. The PPE vectors use a lognormal distribution moment-matched to the source mean/covariance; the selected source covariance has zero off-diagonal entries. The closed cover draws and both reported means and standard errors are fractions on the ten-taxon axis. For example, Cyperaceae has fixed-site SE 0.014250 and joint SE 0.121566; Poaceae has 0.011599 and 0.101076. The difference reflects site selection under this declared bootstrap, not an estimate of vegetation accuracy. Six source sites are too few for a broad population interval.

The source site's historical basin outline, radius changes, independent vegetation truth and chronology uncertainty remain unverified. The author example's numeric radius is a fixed method input, not an admitted age-specific Nordic lake geometry. This result certifies a conditional calculation on a deliberately restricted source example. It does not replace the original 31-taxon author result, claim corrected parity with historical R standard errors, or admit Swedish or four-country LRA predictions.

Replay into artifacts/ and compare all governed outputs byte-for-byte:

PYTHONDONTWRITEBYTECODE=1 OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 \
  MKL_NUM_THREADS=1 .venv/bin/python \
  docs/report/method-certification/author-site-uncertainty/replay.py \
  --source-root /path/to/landclim/author-reference --repository . \
  --output artifacts/jobs/author-site-uncertainty/replay
for name in report.json draws.json cover-draws.csv; do
  cmp "docs/report/method-certification/author-site-uncertainty/$name" \
    "artifacts/jobs/author-site-uncertainty/replay/$name"
done