Skip to content

Source Families

Pollenomics combines eight contracted source families. Their records can share a publication, but the source contract preserves what each family measures, where it applies, how it was acquired, and which claims it can support.

The portfolio and the collection command have deliberately different boundaries. Seven families have collectors and appear in source-support: AADR, boundaries, LandClim, Neotoma, RAÄ, SEAD, and SVAR. Animal ancient-DNA evidence is the eighth contracted family, but it enters through a literature-and-archive intake workflow because there is no single upstream release to collect. A support-matrix row therefore means that a collection adapter exists; it does not mean that a capture is present, complete, current, or admitted to a product.

Contracted Sources

Family Evidence role Suitable questions Principal limit
LandClim primary pollen context pollen sequences and modelled vegetation context model grids and observed sites have different semantics
Neotoma primary pollen context palaeoecological site and pollen comparison temporal coverage and resolution vary by site
SEAD environmental archaeology context archaeological and environmental context access, normalization, and chronology require explicit review
RAÄ Swedish archaeology context heritage and archaeological context in Sweden national coverage cannot be generalized beyond Sweden
Boundaries geographic framing country selection, clipping, and map extent boundaries are not scientific evidence
SMHI SVAR lake and hydrography context registered Swedish waters and lake-oriented filtering registry presence does not establish sampling suitability
AADR human ancient-DNA evidence release-versioned sample metadata current scope excludes genotype processing
Animal source intake sample-owned animal aDNA evidence project, paper, supplement, sample, locality, and chronology recovery source completeness varies by project and sample

Choose By Question And Observation Unit

The most convenient layer is not necessarily the right evidence. Select a family by the object it observes and the claim it is allowed to support.

Intended question Appropriate starting family Observation unit Required qualification
What pollen or vegetation context is represented? LandClim or Neotoma sequence, site, or model grid state whether the value is observed or modelled and retain its time basis
What archaeological context is registered nearby? SEAD or RAÄ environmental-archaeology site or Swedish heritage record preserve national reach, registration bias, and chronology limits
Which geography contains a record? boundaries administrative polygon use only for framing or selection, never evidential weight
Which registered lake is being considered? SVAR water-body registry record distinguish registry identity from sampling suitability
Which human aDNA metadata record is represented? AADR release-versioned sample row retain release and metadata precision; genotype analysis is separate
Which animal specimen supports a claim? animal source library project-owned sample with paper lineage require sample-owned place, time, and coordinate evidence for exact publication

If no family observes the required unit, combining nearby layers does not repair the gap. The valid outcome is a contextual statement, a recovery item, or a refusal.

Source Identity

Every collected family is bound to repository evidence that includes:

  • a source key and display name;
  • selected version and retrieval date;
  • acquisition method and output root;
  • source-specific license posture;
  • hashes for captured and normalized content;
  • provenance linking the source to its repository representation; and
  • replacement rules describing how refreshes affect tracked data.

The current bindings are recorded in data/collection_summary.json. A hash identifies bytes; it does not certify scientific completeness, temporal comparability, or publication fitness.

A Family Writes A Bounded Database Partition

Collection does not pour unlike records into one undifferentiated table. Each family owns a partition whose keys, observation unit, native fields, spatial and temporal meaning, and evidence role remain visible after normalization.

flowchart LR
    Release["identified source release"] --> Native["source-native members"]
    Native --> Partition["family-owned normalized partition"]
    Partition --> Relations["typed cross-family relations"]
    Relations --> Decision["claim-specific product decision"]

The partition boundary prevents a common field name from becoming common scientific meaning. A site_id in Neotoma and a site_id in SEAD belong to different identity domains. Coordinates from either family can participate in a declared proximity relation, but do not authorize an identity join.

Database responsibility Family contract must preserve
member identity source-native key, repository key, release, and collision posture
observation unit site, sequence, grid cell, registry record, sample, or polygon
field meaning native value, units, nulls, parsing or normalization rule, and precision
lifecycle roots captured, normalized, reviewed, and published surfaces
evidence role direct, primary context, contextual, sampling, or framing
replacement behavior staged root, final root, failure preservation, and semantic review

Minimum Capture Packet

A captured member is usable evidence only when the repository can recover the object and the context in which it was acquired:

Packet field Required meaning
source identity family, upstream owner, dataset or archive name, and release or accession
member identity source-native key and repository key without relying on row order
acquisition canonical locator, retrieval method, retrieval time, and access outcome
content identity physical or logical artifact path, media type, size, and digest
reuse posture licence or terms evidence and any redistribution limit
intended role observation unit, geographic and temporal reach, and permitted evidence role
preparation lineage parser or extraction rule, normalized destination, and unresolved fields 
flowchart LR
    Upstream["upstream object"] --> Capture["capture packet"]
    Capture --> Bytes["content identity"]
    Capture --> Member["source-native member identity"]
    Capture --> Role["declared evidence role"]
    Member --> Prepared["normalized record"]
    Bytes --> Prepared
    Role --> Prepared

A DOI without the recovered table, an API URL without a release or response identity, or a copied row without its source-native key is an incomplete capture packet. Such material can remain in recovery state, but cannot silently enter the normalized population.

Source Contract As Scientific Interface

A family contract defines the stable boundary between an upstream ecosystem and every repository consumer. It is more than an acquisition recipe.

Contract dimension Evidence retained Promise to consumers
identity owner, dataset or archive identity, release, and source locator records can be attributed to the intended upstream object
observation unit site, sequence, sample, registry record, polygon, or project relation counts and joins preserve what was actually observed
semantics source-native fields, units, nulls, geometry, and time meaning normalization does not silently strengthen the source
acquisition retrieval method, date, license posture, payload identity, and replacement rule a capture can be reproduced or challenged
role direct evidence, primary context, contextual domain, sampling context, or framing publication cannot promote context into direct proof
fitness coverage, precision, conflict, and product-specific review consumers can distinguish captured data from admitted evidence

A source-name match is not enough to preserve this interface. A release that changes observation unit, schema meaning, licensing, geographic reach, or time semantics requires renewed interpretation before old publication assumptions can be reused.

Contract, Member Schema, And Review Have Different Authority

The family contract is intentionally not a universal row schema. It identifies the family's role, primary question, lifecycle roots, expected artifacts, and coverage metrics. The materialized family records define source-native and normalized member fields; review records define claim fitness.

Surface Owns Example mistake when overextended
family contract family identity, role, lifecycle paths, required artifacts, and metric keys treating a declared normalized path as proof that valid members exist
captured member upstream identity, native fields, retrieval context, and content locator treating raw field presence as normalized meaning
normalized member schema repository identity, represented values, null semantics, and transformations treating a normalized coordinate as publication admission
review record claim dimension, evidence basis, precision, conflict, and fitness treating one product's decision as universal acceptance
product manifest scope, admitted members, roles, qualifications, and bundle identity treating membership as source completeness

This separation lets the family contract remain stable while an upstream schema changes, and lets a record's scientific posture change without rewriting its captured identity. A consumer needs the contract and the member schema; neither can stand in for the other.

Source Admission

flowchart LR
    Candidate["candidate source"] --> Identity{"stable identity and owner?"}
    Identity -->|no| Defer["defer with reason"]
    Identity -->|yes| Access{"recoverable and permitted?"}
    Access -->|no| Defer
    Access -->|yes| Semantics{"observation unit and role understood?"}
    Semantics -->|no| Review["retain for source review"]
    Semantics -->|yes| Contract["admit source-family contract"]
    Contract --> Capture["versioned capture and normalization"]

Admission requires more than topical relevance. The system needs a stable source identity, an acquisition route, a reuse posture, a defined observation unit, a geographic and temporal interpretation, and an explicit role in the publication model. A source that fails one requirement may remain documented for recovery without being presented as a collected evidence family.

An admission decision is auditable only when its evidence can be separated from the source itself. The decision packet names the proposed family owner, upstream authority, selected release or accession, acquisition route, reuse posture, observation unit, evidence role, known biases, normalization owner, and the products allowed to consume the result. It also records rejected alternatives and the condition that would trigger demotion or renewed review. This makes selection policy inspectable instead of hiding it in downloader code or repository history.

Refresh does not repeat admission blindly. A new release can change schema, licence terms, endpoints, coverage, or semantics; those changes require review even when the source name remains stable.

Contract, Capability, State, And Fitness

Four surfaces answer different questions and must not be read as substitutes:

Surface Question answered What it does not prove
source-family contract What kind of evidence is this family allowed to contribute? that software can currently acquire it
source-support Which collector adapters and declared country scopes exist? that any bytes were captured or validated
collection and evidence-state matrices What governed material is present and which checks passed? that a record is fit for a particular product
product admission ledger Which members satisfy one named publication contract, and why? that excluded or unresolved evidence is scientifically irrelevant

The separation matters most for partial recovery. A family can be contracted and supported while its current capture is incomplete; a complete capture can still contain unresolved members; and a resolved member can remain outside a product because its place, time, role, or precision does not meet that product's policy.

Admission Has Three Separate Decisions

Source-family admission, record capture, and product admission answer different questions. Conflating them makes a large collection look more publishable than its evidence warrants.

Decision Question Durable outcome
family contract Is the upstream object identifiable, recoverable, interpretable, and legally usable for a declared role? admitted family or documented deferral
record capture Which source-native members were actually retrieved and normalized? captured member, rejected row, or unresolved member with reason
product admission Does this reviewed member satisfy one named product's identity, spatial, temporal, and role requirements? admitted, qualified, excluded, or deferred membership

A family may be fully contracted while some records remain unresolved. A record may be captured correctly yet excluded from every public product. A member admitted to a spatial inventory may remain ineligible for time-aware analysis. Each state is meaningful and remains queryable.

Source Substitution Is Forbidden

A source may fill only the claim dimension it owns. Boundaries can decide spatial membership but cannot validate a sample coordinate. SEAD can supply environmental archaeology context but cannot date a nearby specimen. SVAR can identify a registered water body but cannot establish coring feasibility. AADR cannot resolve animal sample identity, and animal project metadata cannot stand in for sample-owned locality evidence.

Cross-domain products therefore join governed claims; they do not merge source authority. When one required dimension is absent, the product must qualify, defer, or refuse the claim instead of borrowing certainty from another family.

Change Propagation

flowchart LR
    Release["new release or recovered artifact"] --> Capture["capture identity and diff"]
    Capture --> Normalize["member and semantic diff"]
    Normalize --> Review["coverage, conflict, and precision review"]
    Review --> Decision["admission impact"]
    Decision --> Product["affected product membership"]
    Decision --> None["no public change, with reason"]

The last branch is important: collection and curation are products even when a record remains outside public scope. A refresh is trustworthy when its lack of publication impact is explained, not merely when regenerated maps happen to look unchanged.

Portfolio Breadth Is Not Claim Coverage

Eight contracted families do not imply eight independent witnesses for every question. Coverage is evaluated per claim dimension and observation unit:

Proposed claim Families that may contribute Families that cannot fill the decisive gap
pollen context around a locality LandClim and Neotoma boundaries can frame scope but cannot supply pollen evidence
sample-owned animal chronology animal literature and archive evidence nearby SEAD, RAÄ, or AADR records cannot date the animal sample
country publication membership governed evidence geometry and boundaries country labels cannot repair unresolved coordinate precision
lake decision support SVAR identity and geometry plus role-aware context layers high context density cannot establish bathymetry, access, or coring suitability
temporally aligned cross-domain comparison families with compatible reviewed intervals contextual labels and unresolved time cannot be promoted into numeric overlap

Count families only when the question is portfolio inventory. For scientific support, count independent governed observations that are eligible for the specific claim and retain each source role.

Direct Evidence, Context, And Framing

flowchart TB
    Direct["direct evidence\nAADR and sample-owned animal aDNA"]
    Primary["primary pollen context\nLandClim and Neotoma"]
    Context["contextual domains\nSEAD and RAÄ"]
    Sampling["sampling context\nSVAR"]
    Framing["geographic framing\nboundaries"]
    Publication["cross-domain publication"]
    Direct --> Publication
    Primary --> Publication
    Context --> Publication
    Sampling --> Publication
    Framing --> Publication

Co-publication does not make these roles equivalent. For example, a boundary can determine whether a point appears in a country view, but it cannot validate the point's date or coordinate. A lake record can support candidate discovery, but not a coring recommendation without further evidence.

Animal Source Recovery

Animal aDNA is not acquired as one uniform release. The source library links papers, archive projects, supplements, sample identifiers, sites, locality statements, and chronology claims while retaining their separate provenance.

A project can remain tracked even when it is not publishable. Missing supplements, ambiguous identifiers, unresolved localities, and conflicting dates enter recovery queues and review ledgers. This preserves the difference between “no evidence exists” and “the necessary evidence has not yet been recovered.”

Infrastructure Is Not Evidence

PalaeOpen and similar collaboration or interoperability networks can help align metadata and reuse practices. They are not direct source families unless a specific governed dataset is acquired, versioned, licensed, and admitted through the source contract.

Compare And Inspect