Object And Relation Model

Pollenomics preserves scientific meaning by assigning every identifier to a specific object type and every join to a typed relation. A shared label, coordinate, place name, or publication is evidence to investigate a relation; it is not a universal key.

See domain language for the distinction between a source object, governed object, claim, decision, member, and projection.

Governed Object Types

Object	Stable identity	Facts it may own
source family and release	family key plus release, accession, or captured snapshot identity	upstream owner, version, acquisition, licence posture, and source role
captured artifact	content identity plus path or source locator	retrieved bytes or response, media type, size, digest, and acquisition outcome
archive project	project accession within its archive	project metadata and captured project surface
paper	DOI or another stable publication identity	publication identity and supporting-material inventory
sample	project-owned sample key with source locator	sample identity and sample-owned evidence links
site or locality	stable site key or governed locality claim	place wording, locality class, relation to samples, and supported precision
chronology claim	claim key bound to a governed object and evidence locator	reported time, normalized interval, basis, precision, and conflict posture
product	product key, scope, and version	admission rules, membership, roles, warnings, and exclusions
publication member	product identity plus governed evidence identity	the role and posture of one object in one product

Identifiers remain typed even when an export places several of them in one row. A paper DOI does not identify a sample. A project accession does not identify a site. A map feature identifier does not become the sample key.

Identity, Evidence, And Projection Are Separate

The same real-world subject can participate in several database objects without those objects collapsing into one record:

flowchart LR
    Source["captured source row"] -->|supports| Sample["governed sample"]
    Source -->|reports| LocalityClaim["locality claim"]
    Source -->|reports| TimeClaim["chronology claim"]
    Sample -->|has candidate| LocalityClaim
    Sample -->|has candidate| TimeClaim
    LocalityClaim -->|evaluated for| Point["product point member"]
    TimeClaim -->|qualifies| Point
    Point -->|belongs to| Product["versioned product"]

The sample owns its identity; the evidence surfaces own their reported facts; the curation decision owns which claim is accepted; and the product manifest owns membership. This separation permits one locality correction to propagate to every dependent product without rewriting the captured source or changing the sample key.

Cardinality Is Scientific Meaning

flowchart TB
    Paper["paper"] -->|describes| Project["archive project"]
    Paper -->|provides| Supplement["supporting material"]
    Project -->|contains| Sample["sample"]
    Sample -->|linked by evidence| Site["site or locality"]
    Sample -->|supported by| Chronology["chronology claim"]
    Sample -->|evaluated for| Product["product admission"]
    Product -->|manifests| Member["publication member"]

One paper may describe several projects. One project may contain many samples. Several samples may share a site while retaining different dates. One sample may carry competing locality or chronology claims. One governed object may enter several products under separate admission decisions.

Flattening those relations can create false facts. Copying one project place or age into every sample is not normalization; it is an unsupported change in fact ownership and cardinality.

Relation Contract

A defensible relation retains:

Field	Required meaning
left and right identities	typed governed objects on both sides
relation type	identity, ownership, citation, membership, containment, proximity, temporal overlap, or another declared predicate
evidence locator	source row, supplement location, registry relation, or deterministic derivation supporting the link
method	direct capture, curated resolution, geometric predicate, temporal rule, or declared normalization
precision and scope	where the relation is valid and how strongly it may be interpreted
posture	accepted, qualified, conflicted, unresolved, refused, or outside scope
revision identity	database state under which the relation was evaluated

Proximity and containment are derived relations. They can be reproducible and useful without proving biological association, shared identity, or contemporaneity.

Relation Direction Is Part Of Meaning

Relations are read from a declared subject to a declared object. Reversing an edge may produce a different claim or no valid claim at all.

Directed relation	Valid reading	Invalid reversal
project contains sample	the source project population includes the sample	the sample owns or exhaustively represents the project
source row supports chronology claim	the located row provides evidence for that claim	the normalized claim is the source row
sample linked by evidence to locality	a governed relation connects this sample to this place	every sample at the place shares that sample's evidence
product admits governed object	the object satisfies one product contract	the object's existence authorizes every product
point within boundary	geometry satisfies a versioned containment predicate	the boundary supplies the point's locality evidence
interval overlaps interval	two compatible intervals meet one declared rule	the underlying observations are associated or independent

A relation key therefore includes its typed endpoints, predicate, method, scope, posture, and revision. A pair of object identifiers without the edge contract is not enough to reproduce the join.

Observation Lineage And Independence

Independence is a property of evidence lineage, not of file count or marker count. Two publication members are independent support only when their governing observations and evidence locators are distinct at the level needed by the claim.

flowchart TD
    Artifact["one supporting artifact"] --> Claim["one governed claim"]
    Claim --> World["world member"]
    Claim --> Nordic["Nordic member"]
    Claim --> Country["country member"]
    World --> Copies["three projections, one observation lineage"]
    Nordic --> Copies
    Country --> Copies

Apparent multiplicity	Independence reading
one sample repeated in world, regional, and country products	one governed observation projected into three scopes
one site exported as CSV, GeoJSON, and HTML	one product member represented in three formats
several samples supported by the same project-level locality sentence	distinct sample identities with shared, non-sample-owned locality context
two samples with separate source rows and sample-owned evidence	potentially independent observations, subject to the claim and source design
two datasets derived from the same underlying measurement	related descendants whose shared lineage must remain visible

An independence claim therefore records the observation unit, source locator, derivation path, and relationship between the supporting records. Without those fields, use “multiple records” or “multiple projections,” not “independent evidence.”

Worked Relation Graph

The Direkli Cave goat record shows why typed relations matter. The public feature resolves through a product member to sample capra_hircus:sample:prjeb90141:samea4453841. That sample is namespaced by project PRJEB90141 and supported by supplementary workbook Table S2, row 2. Separate relations connect it to the Direkli Cave locality, supplied coordinates, and sample-owned chronology.

flowchart TB
    Workbook["Table S2, row 2"] -->|identifies| Sample["SAMEA4453841"]
    Project["PRJEB90141"] -->|contains| Sample
    Sample -->|reported at| Site["Direkli Cave"]
    Site -->|has supplied pair| Coordinate["coordinate claim"]
    Sample -->|has sample date| Time["chronology claim"]
    Sample -->|evaluated for| Decision["animal point admission"]
    Coordinate --> Decision
    Time --> Decision
    Decision --> Member["world product member"]

None of the arrows can be replaced by label similarity. The paper label Direkli1-2, archive identity SAMEA4453841, and product feature token are different identifiers connected by evidence. Keeping them distinct allows a coordinate or chronology correction without changing the specimen identity.

Fact Ownership

When a fact is repeated, the database does not resolve disagreement by taking the newest file or the majority value. The fact-ownership registry names the governing surface. Supporting surfaces are derived consumers that must be regenerated when the authority changes.

Fact	Governing level	Typical dependent views
animal project inventory	cross-project project registry	species summaries and coverage dashboards
animal sample identity	project sample master	species-normalized records and atlas candidates
sample-to-site linkage	project sample-site evidence	locality summaries and point admission
chronology	sample chronology evidence	temporal comparison, popup wording, and product eligibility
pollen context	normalized family record	regional pollen layers and comparison packets
country framing	normalized boundary geometry	country membership, counts, maps, and reports

Relations Are Superseded, Not Rewritten In Place

When stronger evidence changes a join, retain enough history to explain both the former and current relation. The stable endpoint identities may remain unchanged while predicate, evidence locator, precision, or posture changes.

Change	Stable element	Superseded element
project context replaced by sample-owned locality	sample identity	locality relation, owner, and evidence locator
approximate site anchor replaced by supplied coordinates	site and sample identities	coordinate relation, basis, precision, and point decision
ambiguous sample label resolved	source artifact and candidate identities	ambiguity relation and selected identity decision
product geography changes	governed evidence object	containment relation and product membership decision

Deleting the prior relation makes historical products and decisions unexplainable. Keeping two relations as simultaneously current creates a false conflict. Supersession names which relation is current, why it replaced the previous posture, and which descendant decisions require reevaluation.

Join Safety

Candidate bridge	Safe only when	Refuse when
sample labels	source locators or curated aliases establish the same sample	spelling or position is the only match
site names	stable site identity or explicit sample-site evidence exists	the name is broad, reused, or project-level
coordinates	basis, precision, method, and governed objects are retained	rounded equality is treated as identity
temporal overlap	both claims have compatible bases, intervals, and a declared overlap rule	one side is textual, contextual, or incomparable
country membership	governed point evidence and boundary version support the predicate	a representative regional point creates false locality precision

The correct result of an unsafe join is an unresolved or refused relation, not a fabricated key.

Anti-Joins Are Accountability Queries

The database must also support questions about missing relations. A sample without a site link, a site without defensible coordinates, or an admitted candidate absent from a product manifest is not discarded by an inner join. It belongs in an unresolved, excluded, or integrity population with a named reason.

Anti-join	Meaning to investigate
recovered sample without locality claim	locality evidence has not been recovered or related
locality claim without mappable coordinate	the place may be broad, ambiguous, or intentionally withheld
reviewed candidate without product membership	scope, qualification, exclusion, or manifest defect must explain the absence
product member without governing evidence relation	database-integrity failure; publication must be blocked

This is how the file-backed database preserves denominators. A query that silently drops unmatched objects cannot support a completeness statement.

Relation Evidence Must Survive Export

A compact export may denormalize labels and values for use, but it must retain enough information to recover the governed relation. At minimum that means the typed object keys, relation or claim key, evidence locator, method, posture, and relevant revision or product identity. If those fields are omitted, the export is a presentation extract: it may be convenient for display but cannot independently support the database claim.

This distinction is especially important for GeoJSON. Geometry proves only that a feature can be drawn. The associated properties must still distinguish reported coordinates, representative points, derived centroids, and governed sample-locality evidence.