Skip to content

Author example raw and prepared taxon support

The source-bound ledger compares the raw input headers for all 17 sites in author grids 512 and 562 with the author's 31-target taxonomy lookup and each site's 31-row prepared LRAready matrix. It records exact SHA-256 identities for the author lookup, PPE table, target-name crosswalk, author formatting script, grid site lists and all 34 raw/prepared site files. The report SHA-256 is a00d3fa98f1eeacf512ee3afd8243be45305aed2dcc7c420e9624a5f13b2a65e. The private author source remains in the external source store; this report retains only derived support facts and source identities.

All 17 prepared matrices contain 31 target rows, but the raw headers have literal author-lookup names for only 12–24 targets per site. Across sites, 168 prepared target rows have no exact raw source-name lead. Most of those rows are all zero. At WHH05, however, five such prepared rows have positive values: Calluna vulgaris (sum 1,364), Fagus sylvatica (5), Fraxinus (1), Juniperus (2), and Poaceae (138). The R parser witness uses the retained author's read.csv(..., check.names = FALSE) call and its taxonomy lookup. R strips trailing spaces from those five WHH05 headers; all five parsed names match the author lookup and their raw column sums equal the corresponding prepared row sums. The author formatting script calls read.csv before matching column names. These five positive rows therefore have source-parser support. The remaining 163 literal gaps have no such whitespace lead in the checked raw headers. Prepared row sums across windows are not a source count denominator or proof of the original counting scheme.

Four raw files also carry duplicate column labels: blank headers at HARESDOWN, WINDMILL and NTWITCHEN, plus repeated Centaurea at LOBBSBOG and Hypericum at WINDMILL. The ledger retains the exact labels and multiplicities. Source-header presence is therefore only a literal lead, and header absence cannot be converted to a measured zero. This does not invalidate the separately verified reproduction of the historical author R calculation on its own prepared inputs. It blocks admission of an all-measured taxon mask for the separately reviewed covariance and joint-draw profiles.

The row-specific decision needed from the source author or a qualified taxonomy reviewer is how duplicate or blank columns are handled and whether each absent target in each site was actually observed as zero or was not recorded under that site's counting scheme. A source-backed counting denominator and sample/window relation are also required before scientific LRA use. No answer or approval is inferred here.

Replay into artifacts/ and compare the report byte-for-byte:

PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m \
  bijux_pollenomics_dev.essd_reproduction.author_taxon_support \
  --source-root /path/to/landclim/author-reference \
  --output artifacts/author-example-taxon-support/replay
cmp docs/report/source-relations/landclim/author-example-taxon-support/report.json \
  artifacts/author-example-taxon-support/replay/report.json
Rscript docs/report/source-relations/landclim/author-example-taxon-support/whh05_header_parity.R \
  /path/to/landclim/author-reference \
  artifacts/author-example-taxon-support/replay/whh05-header-parity.csv
cmp docs/report/source-relations/landclim/author-example-taxon-support/whh05-header-parity.csv \
  artifacts/author-example-taxon-support/replay/whh05-header-parity.csv

The producer checks the 17 site identities, exact 31 target and prepared axes, source-file membership, raw header identities and positive prepared rows without exact raw-name leads before retaining the result. The separate R witness script checks the actual parsed names, author lookup targets and source/prepared sums for the five WHH05 rows.

Full author formatting replay

The site comparison replays the released R formatter's harmonize, aggregate, missing-PPE-row fill and time-window operations for the 17 listed author sites. It consumes 871 raw sample rows. All 17 prepared files match the retained author files byte-for-byte; all 15,028 cells match, including taxon labels, window values and the zero-filled rows. The comparison CSV has SHA-256 13b26b7828ade00ba4037ac334207c9b865a5032f33c40ffe17dc951397321ba. The replay script transcribes the calculation with absolute source paths and per-site scratch files under artifacts/; it does not claim execution of the original Windows-path wrapper unmodified.

The separate count preparation audit recounts all 17 raw and prepared source files: 501,713 raw counts, 299,660 prepared counts and a 202,053-count difference. It verifies the older independent challenge's site totals from private custody while keeping its historical exclusion-category checks distinct from the fresh count sums.

The additional exact source identities are: run.all.taxa.100219.csv 055a38b01755b5c47510199b44f7adebaf7fb33c6e75927544a76e2ceeb9df72, bins.csv 62cac73bafacb8be8d8b7007e9871494024e661ab1a2f28774dcf7cb7046a71f, TWlookup.csv 888a2d5d035c8a49bd957917ca0658bb08824240ac15c1da4b717c633ddae3dd, and site.metadata.csv 72949e9e3faf417be0b7425fc111a9638b37c4a38aab8f8d82e0c3d8739da2de. The formatter SHA-256 is f7463b6e625781dbfdc58186fad70011af9c9f7f03f724ea29abe2489040e118; the ledger above hashes that script and all 17 raw/prepared pairs. Replay used R 4.6.0. The exact numerical match explains how this released source produces its prepared example files; it does not identify the historical inputs to the published Nordic archive, establish a source count denominator for each window, or convert formatter-inserted zeros to observed absences.

OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 MKL_NUM_THREADS=1 \
  Rscript docs/report/source-relations/landclim/author-example-taxon-support/author_prepared_replay.R \
  /path/to/landclim/author-reference artifacts/jobs/author-prepared-replay
cmp docs/report/source-relations/landclim/author-example-taxon-support/sites.csv \
  artifacts/jobs/author-prepared-replay/sites.csv