Skip to content

Neotoma source relationship validation

The acquisition scope receipt binds retained inventory JSON rows and requested dataset IDs to the nine-part manifest. It records six differences in current country selection and the absent original HTTP pagination and response-byte authority.

The observation attribute trace follows one pinned raw datum through its normalized convenience fields and native parent IDs. It records source nulls and the absent denominator explicitly. The native variable surface reconciliation checks all 1,262 relational variable identities, names, source taxon groups, other source semantics and units against 370,936 native datum rows. The compact sample summary derivation checks a displayed site count against pinned relational samples and retains a reverse trace to one observation.

The Nordic lake-parameter review retains the exact five public source matches, the 54-site candidate denominator and a replay against the externally retained page capture. It admits zero historical basins or LRA samples.

The Swedish LandClim site comparison replays all 77 current Swedish prerequisite sites against 84 published LandClim sequences. It retains 34 same-name point leads, including 16 within 10 m, while refusing source identity and radius transfer for every row. The older 37-site comparison is contained row for row in this wider cohort.

The Swedish pollen cohort availability result checks a current Neotoma metadata query and a bounded PANGAEA/EPD profile screen before target counts are opened. Its 99 queried Neotoma site IDs are all retained. Two PANGAEA leads collide with existing core handles; three more apparent leads are retained UNASSIGNED cores, while three named handles need core, age and rights review. It does not close independent confirmation.

The Abborrtjärnen core comparison retains a same-name, 107-sample and 0.334 m point lead but refuses to transfer the published radius or age labels to the Neotoma core.

The Bussjösjön and Bökesjön core comparison retains their full native count and age-claim coverage alongside published site metadata. Close source points and similar series-end numbers still do not establish historical basin radii or core identity.

The retained result verifies the unchanged relational snapshot through the production materialization reader. It includes 200 sites, 206 collection units, 209 datasets, 384 chronologies, 2,798 chronology controls, 12,388 samples, 23,281 age claims, 1,262 variables and 370,936 observations. The source retains 33 conflict records and 52 orphan records. The 33 include 32 chronology-control payload conflicts and one actual same-age-type default chronology conflict. The earlier materialization reported 144 conflicts because it treated multiple default chronologies of different age types as contradictory. Neotoma's chronology manual permits one default per age type; the corrected producer preserves those valid radiocarbon/calibrated pairs. The raw source and all other relational row counts are unchanged.

The reader checks parent existence and consistent site, collection unit, dataset and country ownership across the source graph, alongside manifested file hashes, row counts, stable ordering, country partitions and exact file membership. Its compact parent index does not retain the observation population. The atomic materializer uses the same reader before publishing its staging directory.

The current refresh review at data/neotoma/review/relational_refresh_review.md compares the present pinned candidate with its governed baseline. It is a zero-difference fixed point at the same baseline identity; a changed candidate requires a new source-specific drift explanation and review before replacement.

Unresolved chronology references remain explicit nulls; missing reference fields are refused. Source zeros remain unchanged, while corrected default conflict accounting is fully manifested. This validates relational transport, not chronological comparability, taxon review, propagation eligibility or measured vegetation accuracy.

Reproduce

Run from the repository root using the configured environment:

PYTHONDONTWRITEBYTECODE=1 .venv/bin/python \
  docs/report/source-relations/neotoma/verify_ownership.py \
  --source-root "$PWD/data/neotoma/relational"

The consumer validates source bytes directly rather than using this summary as input. Compare the emitted source/build/manifest/materialization identities with the retained result before reusing it. A changed source requires a new result.

Focused regressions cover missing parents, contradictory ancestry and countries, explicit null chronology, omitted chronology semantics, rejection before atomic publication, and rejection of a dangling sample link even after the modified part and complete materialization hashes are recomputed correctly. An existing country-partition regression alters a site without dependent rows so that it still reaches and tests country reconciliation independently of ownership.