Walk Through One Atlas Result¶

Start with the product guide, then open the Nordic Evidence Atlas. The current product is an atlas_builder: it collects and normalizes source evidence, publishes atlas and country bundles, and offers bounded exploratory methods. It is not a scientific decision engine or an accepted vegetation-accuracy result. This walkthrough follows one published LandClim II modeled-context row from the atlas to its source-bound summary and reproduces that summary without changing the published files.

Three identities matter. The runtime is the installed command and its version. The evidence revision is the checkout and the input hashes in the result receipt. The publication product is the selected Nordic atlas and its governed files. A working command establishes runtime behavior; it does not by itself establish that a source is complete or a result scientifically validated.

Install and inspect the scope¶

Use a source checkout with Python 3.11 or later and uv. From its root, follow the locked setup instructions and run:

make install
artifacts/root/check-venv/bin/bijux-pollenomics --version
git rev-parse HEAD
git status --short
artifacts/root/check-venv/bin/bijux-pollenomics product-scope --json
artifacts/root/check-venv/bin/bijux-pollenomics source-support --json
artifacts/root/check-venv/bin/bijux-pollenomics validate-collection-summary \
  --summary-path data/collection_summary.json

The inspection commands do not collect sources or publish a map. In the checked checkout, product-scope reports atlas_builder and lists scientifically accepted LRA and independent vegetation accuracy among unsupported claims. source-support reports implemented adapters for seven source families; validate-collection-summary confirms the existing seven-source summary's structure. “Implemented” and “validated summary” do not mean that every source record is admitted, current, or fit for a particular scientific claim. The operator workflows explain which later commands write data or publications.

Find the atlas layer and its source row¶

From the Nordic atlas page, open the interactive map. Open Show controls if the filter panel is closed, then expand Modeled context. The controls panel scrolls independently on a small screen. Choose Show modeled metric to activate the layer. Select the source land-cover type family, Open land, and the published 0–100 BP window. Changing the window updates the active layer; the toggle then reads Hide modeled metric. “BP” is the source's years-before-present window label, not a newly dated pollen sample. The map displays individual grid cells from a published REVEALS vegetation reconstruction. Its visible-frame download retains the selected cell values and source window. Do not read a cell as an observed vegetation measurement or a continuous interpolation between windows.

For one cell, the normalized source grid data/landclim/normalized/nordic_reveals_temporal_grid_cells.geojson has record_id=937075:1030:0-100-bp. Its popup identifies Sweden, grid cell 1030, the source DOI, and the 0-100 BP window. Read that exact record from the checkout:

artifacts/root/check-venv/bin/python - <<'PY'
import json
from pathlib import Path

path = Path('data/landclim/normalized/nordic_reveals_temporal_grid_cells.geojson')
features = json.loads(path.read_text(encoding='utf-8'))['features']
cell = next(feature['properties'] for feature in features
            if feature['properties']['record_id'] == '937075:1030:0-100-bp')
print(cell['record_id'], cell['country'], cell['time_label'])
print(cell['reconstruction_values']['OL'], cell['standard_errors']['OL'])
PY

It reads 29.0510048529897 percent modeled open-land cover with source standard error 8.98674192646519 percentage points. Its provenance points to LANDCLIMII.RV.means.JUN2021/TW1.RV.estimates.jun21.csv inside the retained LandClim II archive. That is a model result for one grid cell, not a country estimate or measured vegetation.

The companion panel CSV provides a compact source-to-result check. Find the row with country=Sweden, source_window=0-100 BP, and metric_key=OL:

Field	Published companion-row value	Interpretation
`cell_count`	26	source-grid cells assigned to Sweden in this window
`modeled_count`	24	cells with usable modeled values
`high_quality_count` / `low_quality_count`	11 / 13	source quality labels among modeled cells
`no_pollen_data_count`	2	placeholders excluded from numeric ranges
`minimum_estimate_percent` / `maximum_estimate_percent`	15.506724107 / 89.8378153552	extrema of individual-cell modeled open-land percentages
`minimum_standard_error_percent` / `maximum_standard_error_percent`	2.07094958371 / 19.4680580393	extrema of individual-cell source standard errors, in percentage points

These extrema do not describe a country mean. The standard errors are not pooled or converted to confidence intervals. The panel explanation and receipt identify the 400 country × window × metric rows (four countries, 25 windows, four metrics), the 1,875-row input temporal grid, input hashes, quality counts, and PANGAEA 937075 source. Open land (OL) is a land-cover type; the agricultural-land/cereals (AL) panel is a plant functional type component within it, not another disjoint land-cover class.

Reproduce the bounded summary¶

The installed repository environment includes the development producer. Use an absent output directory under artifacts/; this command writes only that new directory. It reads the checked-in normalized LandClim grid and parameter table, validates the modeled-context manifest, and emits a CSV and JSON receipt:

PYTHONDONTWRITEBYTECODE=1 artifacts/root/check-venv/bin/python -m \
  bijux_pollenomics_dev.source_completion.landclim_modeled_panels \
  --repository . --output artifacts/jobs/landclim-panel-review
cmp artifacts/jobs/landclim-panel-review/panels.csv \
  docs/report/regions/nordic/landclim-modeled-panels/panels.csv
cmp artifacts/jobs/landclim-panel-review/report.json \
  docs/report/regions/nordic/landclim-modeled-panels/report.json

At the checked-in evidence revision used for this tutorial, both comparisons matched byte for byte. A different source revision may produce a difference; inspect the input SHA-256 values and source labels in the two receipts before interpreting it. The producer creates a descriptive summary of an existing model output. It does not rerun the historical LandClim reconstruction or establish an independent vegetation reference.

Check a missing-input refusal¶

The panel summary does not create an LRA request. It describes published modeled cover; it supplies no measured pollen counts, basin geometry or chronology for a new reconstruction. After the two-file reproduction above, check that boundary with the supported, read-only validator:

artifacts/root/check-venv/bin/bijux-pollenomics validate-lra-input \
  artifacts/jobs/landclim-panel-review/lra-request.json --input-root data

The named request is deliberately absent from the panel output. Expect exit status 2, JSON status=refused, and reason_code=lra_request_io_or_encoding. The detail names the missing request. This is a request-file refusal, not a numerical result or a finding that every scientific prerequisite has been checked. The validator writes no product and runs no reconstruction.

For your own reconstruction, replace that path with an actual lra-request.v1 file and keep --input-root pointed at the evidence root that owns its declared relative files. The request needs original pollen counts and taxa, input hashes, supported pollen productivity and transport parameters, measured basin/ring geometry, typed age support, time-bin assumptions, a fixed seed, exclusions and uncertainty declarations. The source prerequisites and LRA command contract explain those requirements. A map cell's modeled percentages and source standard errors cannot supply the missing observations.

A present request can still be refused for missing or changed source bytes, unsupported chronology or parameters, or an unavailable compatible numerical extension. Read the emitted reason and detail before continuing. For lra_extension_missing, return to the locked workspace setup: it includes the separately owned bijux-pollenomics-lra extension alongside the host, development producer and compatibility distribution. The engine integration boundary explains the optional extension. Keep the request, input identities and refusal when sharing the issue; validation success only means the declared exploratory input passed its checks, with scientific acceptance still not_evaluated.

Decide what the result supports¶

The modeled layer is environmental context. The atlas also contains admitted, source-linked direct-evidence features, reference geography, and other context layers; their roles and units remain distinct in the Nordic atlas inventory. A source-support adapter, a map point, and a modeled cell are not interchangeable admissions. Nearby features are not necessarily linked in time or causal process.

For a vegetation-accuracy claim, a separate measured vegetation target would need matched site, time, taxon response, and spatial support. The current vegetation-reference decision admits zero measured pairs among 16 nearby EMPD2–NFI candidates, so this walkthrough yields no accuracy score. Exploratory REVEALS/LOVE commands also require their own complete count, chronology, basin, pollen productivity, and transport inputs; this atlas row does not supply those inputs. If the source row or required inputs are missing, retain the missing-input or refusal result and follow the source prerequisites rather than filling gaps with invented collections.