Bijux Pollenomics

bijux-pollenomics connects curated evidence to public maps and reports about pollen, palaeoenvironmental context, archaeology, hydrography, fieldwork, and ancient DNA. Its database preserves source identity, preparation, scientific decisions, publication membership, and gaps that prevent stronger claims.

Read the product guide Inspect the data system Open the report portal Inspect the repository

From Source To Public Claim

flowchart LR
    source["Release, paper,<br/>archive, registry, or API"] --> capture["Captured identity<br/>and material"]
    capture --> normalize["Normalized objects<br/>and relations"]
    normalize --> curate["Claim-specific review<br/>and decision"]
    curate --> manifest["Product manifest<br/>and member"]
    manifest --> view["Map, table, report,<br/>or field record"]
    view -. "trace backward" .-> normalize

The chain is reversible. A reader can move backward from a visible member to its admission decision, governed evidence, captured material, and upstream identity. A source correction moves forward through affected objects, decisions, manifests, and views.

Database Preparation Is Scientific Work

The tracked data system separates lifecycle stages because each stage has a different authority.

Stage	Decides	Must preserve
capture	what source material entered the repository	source family, release, locator, access context, and digest
normalization	how source-native records become typed objects	native identity, transformations, units, missingness, and loss
curation	which facts, relations, conflicts, and qualifications are accepted	decision owner, proposed use, evidence, reason, and recovery condition
review	whether an object is fit for a named claim or product	claim dimension, population, precision, conflicts, and exclusions
manifestation	which reviewed objects belong to a public product	manifest identity, member ID, role, geography, caveat, and revision
rendering	how product members appear in a map, table, or report	stable member identity and trace back to governed state

A normalized record is not automatically publication-ready. A map marker proves product membership, not every scientific statement shown beside it.

Identify The Result With Three Revisions

Pollenomics results join executable behavior, governed evidence, and a selected publication. Those surfaces can change independently, so a package version or map URL alone is not a reproducible identity.

Revision	Owns	Cannot identify alone
producer	collectors, schemas, normalization, validation, ranking, and publication behavior	which governed records or decisions were present
evidence	captured material, typed objects, relations, conflicts, reviews, exclusions, and recovery state	which eligible objects were selected for one product
product	manifest scope, member and non-member sets, roles, caveats, and rendered descendants	the behavior and evidence revision that produced them

A reusable citation binds all three, then identifies the stable product member or reported population. This distinction is especially important when a newer runtime can read an older evidence revision, or when several regional products select different members from the same governed database.

flowchart LR
    producer["Producer version"] --> run["Publication run"]
    evidence["Evidence revision"] --> run
    contract["Product contract"] --> run
    run --> manifest["Manifest identity"]
    manifest --> member["Member, non-member,<br/>or aggregate identity"]

The installed wheel supplies producer behavior. It does not contain the repository's evidence database or checked-in report products. Conversely, a tracked report remains a product of its recorded producer and evidence state; reopening it with a newer installation does not silently requalify it.

Propagate Corrections Through Ownership

A source correction should invalidate only the descendants that depend on the changed fact, while preserving the previous decision history and unaffected members.

flowchart LR
    correction["Corrected source identity<br/>or source-native fact"] --> affected["Affected governed objects"]
    affected --> decisions["Reopened decisions<br/>and conflicts"]
    decisions --> eligibility["Recomputed eligibility<br/>and ranking inputs"]
    eligibility --> manifests["Affected manifests<br/>and member sets"]
    manifests --> views["Regenerated maps,<br/>tables, and reports"]

At each edge, record the old identity, new identity, reason for change, and affected descendants. A renderer is not allowed to repair a source or curation problem locally. If evidence becomes insufficient, the durable outcome may be a qualified member, explicit non-member, exclusion, recovery item, or release refusal—not a fabricated replacement value.

Evidence Graph

Pollenomics needs several evidence dimensions that cannot substitute for one another:

• identity — which source, project, sample, site, taxon, lake, or product member is under discussion;
• locality and coordinates — what place is supported, by which geometry, method, precision, and conflict resolution;
• chronology — which source expression, interval, evidence class, and comparability posture are available;
• taxonomy — which accepted name, synonym, identification context, and uncertainty apply;
• lineage — how projects, samples, sites, sources, and products connect;
• curation — why evidence was admitted, qualified, excluded, or refused for a particular use.

Spatial proximity does not establish association, contemporaneity, or causation. Matching labels do not establish identity. A contextual period does not become a numeric temporal comparison without a shared basis and precision.

Resolve Conflicts Without Erasing Sources

Two sources can disagree about coordinates, chronology, taxonomy, locality, or association while both source-native records remain valid captures. Curation decides what a named product may use; it must not rewrite the losing source as though the disagreement never existed.

flowchart TD
    assertions["Source-native assertions"] --> compare["Normalize for a declared dimension"]
    compare --> agree{"Compatible under the product rule?"}
    agree -->|yes| accept["Accept with retained provenance"]
    agree -->|no| conflict["Record conflict and affected claims"]
    conflict --> decision{"Evidence supports a decision?"}
    decision -->|yes| qualify["Select or qualify with reason"]
    decision -->|no| unresolved["Keep unresolved, exclude,<br/>or block release"]

A reviewable conflict record identifies the assertions, dimension, comparison rule, decision owner, evidence, product use, and condition that would reopen the decision. Source authority may differ by dimension: a registry can be preferred for administrative identity while a primary field record owns a sampling coordinate. Selecting one assertion for one use does not grant it universal authority over the object.

Propagate Spatial And Temporal Uncertainty

Coordinates and dates are measurements or interpretations with support, not decorative attributes. Precision lost during normalization or aggregation cannot be recovered by displaying more decimal places or a narrower chart bin.

Input state	Public representation	Prohibited inference
point with known precision	point plus method and precision	exact sampling location beyond that precision
polygon, locality, or administrative area	area or bounded locality	point identity at a centroid
conflicting coordinates	qualified selection or visible unresolved state	silent averaging into a new source fact
numeric date or interval	original basis, interval, uncertainty, and conversion	false precision after calendar or age-model conversion
contextual period	categorical context with its vocabulary and source	direct numeric comparability without a governed mapping
missing or withheld location	explicit missingness or access posture	zero coordinate or inferred public point

Spatial or temporal aggregation adds another claim. A map cell, region count, or time bin needs a membership rule for geometries that cross boundaries, uncertain dates that overlap bins, duplicates, qualified records, and missing values. The aggregate must retain its denominator and the uncertainty policy; otherwise its clean visual boundary can imply precision that no member owns.

Keep Map Semantics Auditable

A rendered feature may represent a source observation, a governed object, a product member, or an aggregate. Those roles require different hover text, counts, and trace routes. Clustering markers for presentation must not create a new scientific population, and proximity on the screen must not create a relation absent from the evidence graph.

Readers should be able to identify the feature role, product and evidence revision, member or aggregate identity, population rule, caveat, and route back to source and curation state. When privacy or source restrictions prevent that detail from being public, the feature needs a bounded access explanation—not a fabricated precise substitute.

Govern Sensitive Evidence Without Breaking Lineage

Some records can be scientifically relevant and still unsafe or unauthorized for unrestricted publication. Exact archaeological localities, vulnerable sites, contributor details, culturally restricted knowledge, licensed source material, and field observations may require different public and governed representations.

flowchart LR
    source["Governed source record"] --> classify["Rights, sensitivity,<br/>purpose, and audience review"]
    classify --> full["Restricted evidence object<br/>with complete lineage"]
    classify --> public["Public projection<br/>with bounded disclosure"]
    full --> decision["Claim-specific curation decision"]
    public --> product["Map, aggregate, or report"]
    decision --> product

Sensitive dimension	Governed record preserves	Public representation may use
vulnerable or protected locality	source geometry, precision, restriction basis, custodian, and authorized purpose	coarser area, withheld geometry, or aggregate whose uncertainty and disclosure rule are stated
personal or contributor context	identity only where required for consent, custody, or scientific interpretation	role, organization, or anonymized attribution when sufficient
licensed or non-redistributable material	source identity, access terms, digest, derived-use rule, and reviewer access path	permitted citation, metadata, and non-reconstructive derived evidence
culturally restricted knowledge	provenance, authority, permitted uses, review decision, and withdrawal condition	only the description and conclusion authorized for that audience
embargoed or under-review evidence	capture identity, embargo owner, scope, and expiry or review event	explicit unavailable or pending state rather than inferred content

Generalization must not create false scientific precision. A public centroid is not the hidden sampling point, an approximate date is not the restricted source chronology, and an aggregate must not permit trivial reconstruction of a protected member. The public product should retain a stable relation to the governed object, the reason for restriction, the interpretive limitation, and the authority that can revise or withdraw the projection.

Separate Permission From Scientific Acceptance

Permission to access, transform, or publish a source and scientific fitness for a claim are independent decisions. An openly licensed record may remain scientifically ineligible; a restricted record may be strong evidence for an authorized internal review but unavailable for public reproduction.

Every product decision should therefore retain both the rights posture and the claim-specific evidence verdict. Neither “public” nor “restricted” is a scientific quality grade.

Count The Right Population

Captured rows, normalized objects, reviewed claims, eligible candidates, published members, map features, and display aggregates are different populations.

Before reusing a count, retain:

• observation unit and stable identity namespace;
• source, database, and product revision;
• geographic, temporal, taxonomic, and publication scope;
• eligibility, exclusion, unresolved, and missingness rules;
• the manifest or review surface that owns the denominator.

The repository preserves negative evidence when a visible product member lacks one accountability dimension. Removing the member would hide collected evidence; presenting it as fully supported would overstate the record.

Product manifests must therefore preserve more than displayed members. Their review surface includes eligible members, qualified members, explicit non-members, exclusions, unresolved candidates, and the rule that partitions those populations. Otherwise, a smaller map could be mistaken for a more complete evidence base rather than a stricter publication decision.

Separate Biological Absence From Observation Failure

A source with no reported pollen occurrence may represent true absence, no sampling, inadequate effort, preservation loss, detection below a threshold, taxonomic ambiguity, inaccessible evidence, or an unpublished negative result. Those states must not collapse into the same map value.

Evidence state	Strongest safe interpretation
surveyed with declared method and valid non-detection	not detected under that effort, material, method, place, and time
sampled but assay or preservation failed	occurrence remains unknown for the failed observation
source reviewed with no eligible record	absent from the governed source population, not necessarily biologically absent
taxon unresolved or grouped	evidence applies only to the reported identification level
locality or chronology unusable	occurrence may exist but cannot enter the spatial or temporal claim
no source or sampling evidence	no observation claim is available

An absence claim needs a target taxon, opportunity to observe, detection model or limit, eligible negative population, and uncertainty. Map styling and counts should distinguish non-detection, missing evidence, and excluded evidence so visual emptiness does not become a biological conclusion.

Expose The Sampling Process Behind Map Density

Dense occurrence maps can reflect collection access, research priorities, preservation, publication, database coverage, and digitization effort as much as biological prevalence. Record density is therefore an observation-process result before it is an ecological pattern.

Sampling pressure	Evidence needed for interpretation
spatial access	surveyed areas, accessibility constraints, site-selection rule, and unsampled regions
temporal coverage	collection periods, effort changes, chronology precision, and publication lag
source inclusion	searched sources, languages, repositories, unavailable material, and deduplication
preservation and recovery	material context, assay or identification opportunity, failures, and detection limits
taxonomic attention	identification resolution, synonym policy, specialist review, and grouped records
public disclosure	withheld sensitive sites, aggregation, licensing limits, and reconstructability risk

Effort correction is itself a model and must expose its denominator, covariates, assumptions, residual bias, and sensitivity. Normalizing by the number of available records cannot recover unrecorded sampling opportunity. Publish raw governed counts alongside any effort-adjusted surface so readers can distinguish observed evidence from modeled prevalence.

Current Product Boundary

The implemented runtime is an atlas builder and evidence-publication system. It supports named source collection, source-preserving preparation, governed objects and decisions, declared ranking models, sensitivity outputs, and manifested regional and fieldwork products.

It is not yet a general cross-domain harmonization or causal-inference engine. Unlike observation units are not automatically reconciled, and product membership does not authorize workflow-wide scientific interpretation.

Public Surfaces

Surface	Answers	Does not answer
source families	what entered, under which identity and access conditions	record-level publication fitness
evidence database	objects, relations, fact ownership, revisions, and coherent state	whether every object belongs in a product
curation records	conflicts, decisions, recovery, admission, and refusal	new source-native facts
product manifests	versioned scope, members, non-members, roles, and caveats	stronger evidence than the database contains
Nordic atlas	role-aware spatial comparison and traceability	causation or contemporaneity from proximity
fieldwork records	dated visits, locations, media, and bounded observations	lake-wide conditions or sampling readiness

Current Scientific Limits

The public products remain deliberately smaller than the collected evidence:

• animal sample, locality, chronology, coordinate, and source-recovery gaps can remain qualified, excluded, or release-blocking;
• SEAD supports inventory and spatial context, not general numeric temporal comparison;
• RAÄ coverage is Sweden-specific and does not provide an equivalent Nordic registry;
• modern administrative boundaries frame publication scope without adding scientific weight;
• lake rankings express evidence richness and decision support, not field readiness or coring-site selection;
• field visits document bounded observations without validating nearby data layers.

These limits are product facts and remain attached to the relevant evidence and report surfaces.

Reproduce Or Challenge A Result

1. Name the product manifest and stable member.
2. Recover the admission decision and governed evidence identities.
3. Inspect source, locality, coordinate, chronology, taxonomy, role, and caveat.
4. Confirm database, runtime, and product revisions.
5. Recompute only through the owner of the disputed transition.
6. Compare identities, semantics, decisions, populations, and manifested descendants—not only files or rendered appearance.

Continue with the data guide for source and curation authority, the database model for coherent evidence state, or the report portal for checked-in publication products.