Collection Summary

data/collection_summary.json identifies the collected state from which publication begins. It records the collection version, generation date, source roots, acquisition metadata, and hashes for captured and normalized material.

The checked-in summary names collection version v66, generated on 2026-06-22, and seven collected families: AADR, boundaries, LandClim, Neotoma, RAÄ, SEAD, and SVAR. These are collection members. Animal ancient-DNA curation has its own project, paper, supplement, and sample governance under data/adna/ and is not made complete merely by this seven-family summary.

Recorded Contract

Field	Meaning	Reader use
generated_on	date the summary was assembled	distinguish collection time from publication time
version	shared collection label	join source and publication surfaces from the same state
collected_sources	source families included by the collector	establish collection breadth, not evidence completeness
source_output_roots	governed location of each family	locate the family tree without guessing
source_metadata	version, licence posture, retrieval date, method	interpret origin and reuse constraints
source_hashes	captured and normalized content digests	detect changed collection content
source_provenance	family name, role, roots, and digests	connect family identity to stored state
source_replacement_rules	family-specific refresh and replacement behavior	distinguish additive acquisition from governed replacement
source_traceability	source identity and stored artifact linkage	recover the acquisition basis for a family
contract_artifacts	machine-readable contracts derived for the collection	locate the semantics used by downstream consumers

The summary also carries family-scale counts such as LandClim sites and grid cells, Neotoma points, RAÄ heritage and total sites, SEAD points, and SVAR lakes. These counts identify the assembled state; their exact observation units and temporal semantics remain governed by the family contracts.

A Digest Can Identify Empty Content

A content digest proves identity of the bytes or logical member stream that was hashed; it does not prove that the stream contains governed records. The current AADR normalized_sha256 is e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855, the SHA-256 digest of an empty byte stream. The dedicated evidence-stage matrix accordingly reports the AADR normalized stage as missing.

This is not a hash failure. It is a meaningful receipt for absence at that boundary and must not be described as a populated normalized dataset.

Digest observation	Additional evidence required before interpretation
digest changed	member-identity and semantic diff
digest unchanged	proof that the same hashing scope and member ordering were used
digest identifies empty content	expected-member accounting and an explicit missing or valid-empty decision
digest is absent	acquisition or preparation outcome explaining why content identity is unavailable

A scientifically valid empty result is possible only when the expected source population, query or extraction scope, and zero-member outcome are themselves governed. Otherwise empty content is a missing preparation state, not evidence that the upstream population is empty.

Collection Header, Not Record Catalogue

The summary behaves as a collection-level transaction header. It identifies the source partitions and their captured and normalized content state, but it does not enumerate every governed object, relation, conflict, or publication member.

flowchart LR
    Header["collection identity and family digests"] --> Partition["family-owned record partitions"]
    Partition --> Claims["normalized claims and relations"]
    Claims --> Review["curation and product decisions"]
    Review --> Bundle["publication manifest and members"]

Question	Correct authority
Which collected source roots belong to this state?	collection summary
Which source-native or normalized records exist?	family-owned captured and normalized partitions
Which value owns a sample, locality, chronology, or coordinate fact?	fact-ownership registry and governing evidence record
Which records passed one product contract?	product admission and bundle manifest
Which visible rows and exclusions belong together?	publication bundle and review companions

Treating the summary as a record catalogue would make its family counts appear scientifically comparable and would hide animal curation that is governed outside the seven collector-managed roots.

Collection Membership Is Not Publication Membership

The v66 summary records seven collector-managed families:

Collection member	Collected root	Publication decision still required
AADR	data/aadr	release-row identity, deduplication, geography, and product scope
boundaries	data/boundaries	geometry version and named scope relation
LandClim	data/landclim	site-sequence versus grid-cell role and temporal posture
Neotoma	data/neotoma	site identity and temporal comparability class
RAÄ	data/raa	selected classification, aggregation grid, and Sweden-only role
SEAD	data/sead	four-country membership and context-only temporal posture
SVAR	data/svar	lake identity and candidate-product eligibility

Animal aDNA is governed separately under data/adna/. Its 40-project intake, 868 recovered rows, and 234 admitted point features must not be inferred from collected_sources. Conversely, the seven-family collection summary does not grant any member automatic map or report admission.

flowchart LR
    Summary["collection_summary.json / v66"] --> Families["7 collected source families"]
    Animal["separate animal source library"] --> Review["claim-specific review"]
    Families --> Review
    Review --> Product["world, regional, country, or lake product"]
    Review --> Refusal["qualified, excluded, or deferred evidence"]

The publication asks a narrower question than collection: which reviewed members, from which authorities, satisfy this named product? That question is answered by the product manifest and evidence rows, not by the collection list.

flowchart LR
    Capture["captured source state"] --> Hash["snapshot digest"]
    Capture --> Normalize["normalized family state"]
    Normalize --> NormalizedHash["normalized digest"]
    Hash --> Summary["collection_summary.json"]
    NormalizedHash --> Summary
    Summary --> Publication["versioned publication inputs"]

Interpreting Change

A changed snapshot digest means captured source content changed. A changed normalized digest means the normalized family state changed. A new report with unchanged collection hashes may reflect a new scope, contract, or presentation over the same collected state. These are different kinds of change and should not be described as equivalent data growth.

The summary does not measure sample admissibility, coordinate quality, chronology precision, or geographic completeness. An empty or small normalized digest can coexist with a tracked source family, and a collected family can be absent from a particular product because its publication rules do not pass.

Three Identities To Compare

Identity	Governing surface	Change it can reveal
collection identity	version, source metadata, and hashes in collection_summary.json	captured or normalized family state changed
evidence identity	stable normalized, sample, site, and review records	interpretation, precision, conflict, or curation changed
publication identity	scope, version, member inventory, and contract in the bundle	admission, geography, caveat, or presentation membership changed

The shared label v66 is necessary but not sufficient to prove that two files belong to one coherent publication. Their manifest membership, governing identifiers, and hashes must also resolve. A later product can legitimately reuse an unchanged collection while applying a revised review or scope rule; that is a publication change over the same captured state, not a new source release.

Verify A Collection Identity

Use the summary as a join point rather than as a narrative status badge:

1. match the publication's collection version to version;
2. confirm the family appears in collected_sources;
3. locate its governed tree through source_output_roots;
4. read its acquisition version, licence posture, date, and method in source_metadata;
5. compare the captured and normalized digests in source_hashes; and
6. follow the family contract for observation units, precision, and permitted publication roles.

A bundle that cites v66 but cannot resolve the relevant source root and digest has an incomplete collection link. A matching digest establishes content identity, not scientific fitness; admission and caveat surfaces provide that separate decision.

Use the source family matrix for evidence roles, the cross-domain matrix for comparability, and publication limits for what remains outside the visible products. The evidence database explains how collection identity relates to governed records and projections.