Map Inputs

The atlas is assembled from governed evidence layers, geographic framing, and publication decisions. There is no single coordinate table that defines the map. Each visible feature retains a source-family role and a path back to the normalized or reviewed record that authorized it.

The domain language defines projection, evidence role, precision, publication member, and product scope.

The Six Governed Input Families

Input	Role in the atlas	Governing evidence surface	Interpretation limit
LandClim	primary pollen context	data/landclim/normalized/nordic_pollen_site_sequences.geojson	pollen sequences and grids, not direct human or animal evidence
Neotoma	primary pollen context	data/neotoma/normalized/nordic_pollen_sites.geojson	separate pollen inventory with uneven temporal support
SEAD	contextual archaeology	data/sead/normalized/nordic_environmental_sites.geojson	environmental archaeology context, not uniformly dated evidence
RAÄ	contextual archaeology	data/raa/normalized/sweden_archaeology_layer.json	Sweden-scoped density context, not Nordic-wide site coverage
Nordic boundaries	geographic framing	data/boundaries/normalized/nordic_country_boundaries.geojson	scope and clipping only; contributes no evidence score
Animal aDNA	sample-backed evidence plus visibly qualified project context and explicit refusals	data/adna/final/atlas/animal_atlas_point_candidates.json	point classes must remain distinct within an incomplete recovery program

Summary files and raw captures remain important refresh and review anchors, but they do not replace the normalized or admitted surface that governs a visible layer. This distinction matters when a raw inventory contains rows that normalization later excludes or qualifies.

SVAR lake records provide the authoritative candidate-lake anchors for the Sweden priority analysis. Human aDNA exports and additional report layers enter the products that declare them, but their presence does not change the role of the six audited atlas families above.

Evidence Role Is Part Of The Layer

flowchart LR
    Direct["direct sample evidence"] --> Product["declared publication"]
    Pollen["primary pollen context"] --> Product
    Archaeology["contextual archaeology"] --> Product
    Lakes["candidate-lake anchors"] --> Product
    Boundaries["geographic framing"] --> Product
    Product --> Reader{"reader interpretation"}
    Reader -->|claim| Direct
    Reader -->|environmental support| Pollen
    Reader -->|surrounding context| Archaeology
    Reader -->|selection unit| Lakes
    Reader -->|scope only| Boundaries

Layer co-location does not collapse these roles. A boundary can select a country but cannot support a scientific score. An archaeology layer can describe surrounding evidence without becoming a sample observation. A pollen sequence can supply palaeoenvironmental context without becoming direct aDNA evidence.

Assembly And Review

flowchart LR
    A[Source-family capture] --> B[Family normalization]
    B --> C[Coverage and posture review]
    C --> D{Eligible for this geography and product?}
    D -->|yes| E[Scoped layer export]
    D -->|no| F[Gap or refusal surface]
    E --> G[World map]
    E --> H[Regional map]
    E --> I[Country bundle]
    G --> J[Feature IDs and provenance links]
    H --> J
    I --> J

Eligibility is evaluated for a particular product. A record admitted to a world layer is not automatically Nordic evidence; a Sweden density layer is not automatically available for another country; and a tracked animal project is not automatically a mapped sample.

Layer Assembly Packet

Each emitted layer can be reconstructed from a bounded packet:

Packet member	What it controls
source population	exact normalized or reviewed input identities and version
relation contract	allowed joins from source record to evidence and geography
selection contract	product scope, evidence roles, point classes, and fitness predicates
field projection	properties retained for identity, provenance, space, time, role, and qualification
member inventory	stable feature IDs included in the layer
non-member inventory	excluded, refused, deferred, or outside-scope candidates and reasons

The layer count is derived from the member inventory. It cannot substitute for that inventory, and a browser filter cannot alter it. This makes a layer rebuild reviewable as an identity and decision diff rather than a visual map comparison.

Preserve Field Authority Through Projection

A public feature combines fields owned at different boundaries. The export is a projection of those authorities, not a new owner of every copied value.

Projected field	Governing authority	Conflict response
source and member identity	source-family capture and normalized or evidence member	stop if the exported identity cannot resolve uniquely
feature role	review and product contract	do not infer role from marker style or layer name
geometry and coordinate basis	governing normalized or sample-owned spatial evidence	retain qualification; never prefer display precision
chronology and comparability	temporal review or sample-owned chronology evidence	retain units, basis, caveat, and refusal posture
publication membership	scoped bundle manifest and admission decision	a source row or browser state cannot self-admit
display label and popup text	field projection under the publication contract	correct the projection when it conflicts with governing evidence

flowchart LR
    Identity["source and member identity"] --> Feature["public feature"]
    Review["role, place, and time review"] --> Feature
    Admission["scope and admission decision"] --> Feature
    Feature --> Manifest["bundle member inventory"]
    Feature --> Views["CSV, JSON, GeoJSON, Markdown, and HTML"]

When two public formats disagree, neither becomes authoritative because it is more detailed or easier to read. Resolve the field through its owning evidence or decision surface, correct the projection, and then regenerate every affected format from the same member inventory.

Geometry Semantics

Geometry is a claim with a basis, not merely a renderable value. Every mapped feature preserves which of these meanings applies:

Geometry class	Defensible reading	Required restraint
source-supplied point	the source reports this coordinate pair for the governed object at its declared precision	do not infer measurement accuracy from decimal places alone
sample-to-site point	sample evidence supports a relation to a governed site whose coordinates are separately owned	do not describe the coordinate as sample-supplied unless the source says so
named-place resolution	a documented resolver produced an approximate point from source wording	retain method, confidence, and approximate posture
representative regional point	geometry permits broad contextual display	never use for exact distance, containment, or sample-locality claims
centroid	a deterministic point summarizes a polygon or cell for display	retain the source geometry; the centroid is not an observed location
polygon or grid cell	geometry represents extent, scope, or aggregation	do not expand its count into synthetic point observations
withheld or unresolved	no defensible public geometry is available	retain the member in review or exclusion accounting rather than inventing a point

Coordinate reference system, axis order, transformation method, source geometry, and declared precision travel with the projected feature whenever they affect interpretation. A technically valid GeoJSON coordinate is not, by itself, evidence that the scientific object was observed there.

Materialized Input Contract

A layer is available for assembly only when the exact artifact named by its source-family contract exists with governed content. Related files do not satisfy that requirement:

Available artifact	Missing artifact	Result
source manifest	contracted raw capture	source identity known; raw stage still missing
normalized summary and count	contracted normalized member dataset	coverage may be described; layer cannot be rebuilt
evidence-stage matrix	source-specific review	review stage remains missing
retained published GeoJSON	its current normalized authority	product remains inspectable; current rebuild is blocked
empty directory or .gitkeep	any contracted evidence artifact	no stage evidence

This rule prevents a downstream map layer from being treated as a convenient backup of its own source. Recovery begins at the governing input or review surface and then regenerates the scoped export and bundle.

The current input scale is intentionally heterogeneous: 492 LandClim site sequences, 200 Neotoma sites, 2,172 normalized SEAD sites, a RAÄ density source covering 761,917 published Swedish sites, four Nordic boundary polygons, and 234 reviewed animal publication points. These counts describe different units and roles and must never be summed into one evidence total.

One Viewport, Different Scientific Objects

A LandClim marker, a Neotoma marker, a SEAD marker, an animal point, and a RAÄ density cell may overlap visually while answering different questions:

Visible object	Example meaning	Valid comparison	Invalid promotion
LandClim point	one dataset-specific pollen site sequence such as Aal Præstesø	compare declared sequence coverage and location	call the point a dated pollen event
Neotoma point	one database site such as Abborrtjärnen with nested collections and samples	compare site-level coverage under its temporal posture	count nested samples as independent map points
SEAD point	one captured environmental-archaeology site inventory row	report a declared spatial relation	infer same-period association without recovered chronology
animal point	one product-admitted animal evidence feature	follow sample or project identity and qualification	treat all admitted points as equally complete samples
RAÄ cell	aggregate count of selected registry records in one grid cell	compare declared density cells	reconstruct synthetic archaeological site coordinates

The correct cross-layer join is therefore a typed relation between two named members, not a merge on marker position. Its result records the two source identities, relation rule, distance or interval calculation, scope, and the weaker input posture.

flowchart LR
    View["shared viewport"] --> LandClim["site sequence"]
    View --> Neotoma["database site"]
    View --> SEAD["site inventory row"]
    View --> Animal["admitted evidence feature"]
    View --> RAA["aggregate density cell"]
    LandClim --> Relation["typed spatial or temporal relation"]
    Neotoma --> Relation
    SEAD --> Relation
    Animal --> Relation
    RAA --> Relation
    Relation --> Claim["qualified cross-domain statement"]

Layer Acceptance Contract

Each exported layer answers five questions before it enters a bundle:

Contract field	Required answer
identity	what stable feature and source-family identifiers survive export?
role	is the feature direct evidence, primary context, contextual archaeology, a candidate anchor, or framing?
admission	which normalized or reviewed surface authorized inclusion?
qualification	which spatial, temporal, and citation limits travel with the feature?
scope	which world, regional, or country product selected it?

If a layer cannot answer all five, it may be useful source material but is not ready to function as a governed publication input.

flowchart LR
    Family["source family"] --> Normalize["normalized identity"]
    Normalize --> Review["role and qualification"]
    Review --> Admit{"product contract passes?"}
    Admit -->|yes| Export["scoped layer"]
    Admit -->|no| Refusal["gap or refusal surface"]
    Export --> Bundle["versioned publication bundle"]

The refusal branch is part of the evidence architecture. It keeps a visually clean layer from concealing records that were captured but could not support the product's claims.

What A Scoped Export Must Preserve

A public layer must retain enough structure to answer:

• which source family supplied the feature;
• whether the feature is direct evidence, context, or framing;
• which geography admitted it and why;
• which normalized identity and source record it came from;
• whether its coordinates are direct, resolved, or withheld;
• whether its temporal semantics permit numeric comparison;
• which rows were excluded or refused rather than silently dropped.

This is also why visual filters are not the publication boundary. A feature must already belong to the scoped export before the browser can show or hide it. Client-side controls cannot authorize an otherwise ineligible record.

Geographic Products Are Subsets

World, regional, and country outputs have explicit publication contracts. A regional bundle must be a defensible subset of its upstream evidence, and a country bundle must preserve sample, locality, chronology, coordinate, and citation linkage for every included direct-evidence feature.

Boundary polygons frame those subsets but never increase evidence strength. Context layers may explain what surrounds a sample or lake, yet proximity does not turn them into sample-owned proof.

Use The Correct Denominator

Layer counts have meaning only against the population from which they were selected. For animal points, distinguish tracked projects, recovered samples, locality-qualified samples, and admitted point rows. For pollen and archaeology layers, retain the applicable source inventory and temporal posture. Reporting only the number of visible markers makes conservative admission look like collection completeness and makes heterogeneous layers look comparable when they are not.

Tracing A Visible Feature

Use the feature's layer and stable identifier to follow this route:

1. identify the product contract under docs/report/world/, docs/report/regions/<region>/, or the country publication root;
2. locate the feature in the corresponding GeoJSON or evidence-row export;
3. follow its source-family, sample, site, or source identifiers into the normalized data root;
4. inspect coordinate and temporal semantics before comparing it with another layer;
5. check the refusal and coverage surfaces when expected evidence is absent.

The trace should resolve two independent paths: evidence lineage from feature to source, and publication lineage from feature to scoped manifest and parent product. Evidence lineage explains what the member means; publication lineage explains why that member is present in this particular map. One path cannot substitute for the other.

The repository atlas input audit summarizes refresh anchors and tracked metrics for the active input families. The source-fact ownership registry identifies the governing evidence surfaces. The cross-domain evidence matrix shows their responsibilities, while point publication rules define the direct-evidence admission boundary.