Animal Source Intake

An animal ancient-DNA point begins with a project accession, a paper, and often several supplementary files—not with a finished map row. Bijux Pollenomics preserves that recovery chain so a published sample can be traced to the artifact and passage that support its identity, locality, and chronology.

The tracked collection is therefore broader than the atlas. A project may be important enough to curate while still lacking the evidence needed to place a sample on a map. Its absence from a map means not yet admissible at that resolution, not no evidence exists.

Collection Snapshot

The current recovery review records:

Measure	Current value	Interpretation
tracked archive projects	40	declared project inventory, not sample count
sample-foundation truth rows	894	curated source-row grounding and blocker population across 10 species
recovered final sample rows	868	extracted governed rows, not a complete source census
projects with exact expected counts	4	projects for which exact recovery completeness can be measured
projects with a minimum expected floor	22	projects for which recovery can be tested against a lower bound
projects with implausibly low recovery	8	source or extraction work still blocks a stronger claim
projects ready for publication review	8	projects that reached this lifecycle checkpoint
blocked projects	26	tracked projects held before that checkpoint

These measures are published together because no one of them is an honest summary of the collection. In particular, dividing the recovered sample total by the four exact-denominator projects would compare unrelated populations.

Denominator Confidence Governs Recovery Claims

Project recovery uses three denominator postures. They must not be combined into one portfolio-wide completion percentage.

Denominator posture	What can be claimed	What remains unknown
exact expected count	recovered, missing, duplicate, and unresolved identities can reconcile to a fixed source population	whether uncatalogued source material exists outside the declared authority
minimum expected floor	recovery can be shown to meet or fall below a supported lower bound	the true total and percentage complete
unknown expected count	recovered identities and their evidence can be reported	completeness, deficit, and percentage complete

flowchart LR
    Source["project evidence"] --> Denominator{"expected population posture"}
    Denominator --> Exact["exact reconciliation"]
    Denominator --> Floor["lower-bound comparison"]
    Denominator --> Unknown["recovered population only"]
    Exact --> Portfolio["posture-preserving portfolio summary"]
    Floor --> Portfolio
    Unknown --> Portfolio

A portfolio summary may count projects in each posture and may sum recovered identities after collision review. It may not divide the summed recovery by a mixture of exact, minimum, and unknown denominators. That calculation would manufacture precision from projects whose source populations are not known.

The Foundation Is A Preparation Ledger

data/adna/governance/animal_sample_foundation_truth.json records whether each foundation row has enough attributed evidence to support downstream curation:

Foundation posture	Rows	Intake meaning
fully grounded	502	identity and required evidence dimensions are attributable at the declared scope
partially grounded	256	useful evidence exists, but one or more dimensions remain materially limited
blocked: missing metadata	29	source metadata required to establish the governed row is absent
blocked: missing location detail	4	locality evidence cannot support the requested spatial claim
blocked: weak chronology	103	temporal evidence remains too weak for the stronger chronology claim

These 894 rows are not interchangeable with the 868 project sample-master identities. The foundation classifies preparation evidence; the sample master governs recovered identity. A row can be valuable in one population without having a one-to-one counterpart in the other.

flowchart LR
    SourceRows["captured source rows"] --> Foundation["894-row preparation ledger"]
    Foundation --> Grounded["fully or partly grounded"]
    Foundation --> Blocked["metadata, location, or chronology blocker"]
    Grounded --> Identity["project-owned sample identity where supported"]
    Blocked --> Recovery["named source-recovery work"]

The blocker class determines the next evidence action. Missing metadata calls for source or relation recovery; missing location detail calls for stronger sample-to-site evidence; weak chronology calls for a sample-owned temporal source or a narrower non-numeric posture. None is repaired by copying a value from a published point.

Intake And Point Populations Are Not The Same

The current point-evidence surface combines two identity postures:

Population	Count	Intake basis	Permitted description
final extracted samples with supplementary coordinates	233	directly extracted sample rows with final identity	sample-backed publication points
Wadi Halfa dromedary context	1	project SRP073444, paper-pinned place statement, provisional sample identity	qualified project-anchored context point

The second row is not part of the 868 recovered sample population merely because it appears in the 234-row point review. Its stable token anchors the project context, while sample_evidence_status: not_yet_recoverable and sample_identity_resolution: provisional preserve the missing sample-level link.

This distinction prevents two inverse errors: dropping useful paper-backed context because sample recovery is incomplete, and reporting that context as if a sample-bearing table had been recovered. A query that requires samples uses the first population. A map that accepts qualified project context may use both, but must expose the identity class.

From Project To Publishable Sample

flowchart LR
    A[Archive project] --> B[Paper linkage]
    B --> C[Article and supplement capture]
    C --> D[Sample-row recovery]
    D --> E[Stable sample identity]
    E --> F[Locality evidence]
    E --> G[Chronology evidence]
    F --> H[Coordinate review]
    G --> I[Temporal normalization]
    H --> J{Publication rules pass?}
    I --> J
    J -->|yes| K[Atlas and country layers]
    J -->|no| L[Tracked curation record]

Each transition has its own evidence requirement. A readable paper does not prove that its sample table was recovered; a recovered sample label does not prove an exact site; and a named site does not justify coordinates unless the coordinate source and resolution are explicit.

For project PRJEB90141, supplementary workbook Table S2 supplies four final goat sample rows—Direkli1-2, Blagotin3, Semnan3, and Acem2—with distinct archive identifiers, localities, coordinates, and chronology. The project accession is shared; the sample claims are not. For SRP073444, the paper supplies a named Wadi Halfa context while a recoverable sample-master row remains absent. These two projects therefore follow different branches of the intake contract.

flowchart LR
    Goat["PRJEB90141 + Table S2"] --> GoatRows["4 final extracted sample rows"]
    GoatRows --> GoatPoints["sample-backed points"]
    Camel["SRP073444 + paper place statement"] --> CamelGap["sample row not yet recoverable"]
    CamelGap --> CamelPoint["qualified project-context point"]

Governed Intake Surfaces

The source library separates cross-project inventory from project-owned evidence. This lets a reader locate the decision that controls each transition:

Question	Governing surface	What it establishes
Which projects and papers are tracked?	tracked_project_and_paper_inventory.json and paper_registry.json	archive and publication identity without implying sample recovery
Were supporting files acquired and classified?	supplement_acquisition_checklist.json and supplement_file_family_audit.json	fetch and file-family posture for sample-bearing material
What remains incomplete at intake?	source_intake_audit.json	project-level blocks, deficits, and next evidence boundary
How much sample material was expected and recovered?	project_sample_master_completeness.json and each project's sample_master.json	denominator-qualified recovery and stable sample identity
Which samples resolve to named sites?	project_sample_site_review.json and each project's sample_sites.json	sample-to-site relations without inventing coordinates
Which locality claims conflict or need normalization?	each project's locality_worksheet.json and sample_locality_evidence.json, plus sample_locality_conflict_ledger.json and site_name_normalization_dictionary.json	reported place text, normalized identity, precision, and unresolved disagreement
Which temporal claims are sample-owned?	project_sample_chronology_review.json and each project's sample_chronology.json	chronology coverage and source-owned temporal posture

The cross-project files live under data/adna/governance/source_library/. Project-owned files live under data/adna/governance/source_library/projects/<project-accession>/. A row in a cross-project review must resolve to the project file that supplies its sample, site, locality, or chronology evidence; the review cannot become a substitute authority.

Every Recovery Edge Has A Receipt

The source library is a graph of attributed relations, not a folder of papers beside a table of samples. Each edge is independently recoverable:

Relation	Receipt required
project to paper	accession, DOI or stable paper identity, relation basis, and governing registry row
paper to artifact	canonical source URL, logical and physical path, media type, size, fetch outcome, and content identity
artifact to sample	table, sheet, row, archive member, or excerpt locator plus source-native sample label
sample to site	project-owned sample and site keys, native relation or explicit curation basis, and ambiguity posture
sample to chronology	verbatim claim, source locator, ownership class, dating basis, normalized result, and precision
site to coordinate	reported locality, coordinate source, resolution method, confidence, and disagreement outcome

flowchart LR
    Project["project accession"] -->|registry relation| Paper["paper identity"]
    Paper -->|capture receipt| Artifact["source artifact"]
    Artifact -->|row locator| Sample["stable sample"]
    Sample -->|project-owned relation| Site["governed site"]
    Sample -->|claim lineage| Time["chronology evidence"]
    Site -->|spatial provenance| Coordinate["coordinate claim"]

An orphaned edge blocks only the claim that depends on it. A sample can retain its identity when its locality is unresolved, and a site can retain its name when no defensible coordinate exists. Conversely, a complete-looking endpoint cannot repair a missing edge: an exact coordinate is not sample evidence when the sample-to-site relation is absent.

One Intake Produces Several Governed Decisions

Extracting a sample row does not write one universal “curated” status. It creates independent decisions that can mature at different rates:

Decision record	Candidate input	Durable outcome
sample identity	native label, accession, table locator, and project relation	final identity, unresolved candidates, or refusal to merge
sample-to-site relation	sample label, site label, table relation, and project context	direct relation, documented substitution, ambiguity, or missing link
locality claim	verbatim place text, site evidence, hierarchy, and conflicts	resolved locality with precision or unresolved claim
chronology claim	verbatim date, basis, sample ownership, and competing wording	numeric interval, textual posture, contextual range, conflict, or unknown
coordinate claim	supplied coordinate or named-place resolution with provenance	exact, approximate, substituted, region-only, withheld, or unresolved
publication admission	the preceding decisions plus product scope	admitted, qualified, excluded, or deferred member

flowchart TB
    Row["recovered sample row"] --> Identity["identity decision"]
    Identity --> Site["sample-to-site decision"]
    Site --> Locality["locality decision"]
    Identity --> Time["chronology decision"]
    Locality --> Coordinate["coordinate decision"]
    Coordinate --> Admission["product admission"]
    Time --> Admission

This fan-out is the central curation operation. A final identity can coexist with unresolved chronology; an exact coordinate can coexist with text-only time; and a project-context feature can remain visibly distinct from a final sample-backed member.

What Is Captured

Evidence unit	Preserved information	Why it matters
Project	archive accession, species scope, project URL, intake status	keeps archive identity separate from later interpretation
Paper	DOI, canonical URL, title, journal, year, linked projects	establishes the publication anchor
Source artifact	source URL, logical path, storage path, content type, size, fetch status	identifies the exact acquired object
Supplement	file family, archive member, parse status, linked paper	exposes whether the usable sample evidence was actually recovered
Sample	source-native label, stable repository identifier, source locator and excerpt	prevents project-level evidence from being presented as sample-level evidence
Locality	reported place text, site assignment, resolution, provenance, conflicts	controls how precisely a sample may be mapped
Chronology	reported date text, normalized interval, basis, precision, provenance	controls whether temporal comparison is defensible

HTML article and archive captures may be stored as compressed .html.gz payloads. Their logical article.html or archive_metadata.html identity stays stable, while companion metadata records the physical path, byte size, and encoding. Storage optimization therefore does not break provenance locators.

Evidence Hierarchy

Different artifacts can support different claims about the same sample. The curation record keeps those authorities separate:

Claim	Preferred authority	Acceptable narrower fallback	Never sufficient by itself
sample identity	sample-bearing supplement or archive row with stable locator	explicit paper table or project metadata linked to the sample	project title or species list
project membership	archive accession and sample-to-project relation	paper statement naming both project and sample	DOI proximity alone
locality	sample row, sample-specific table, or explicit sample-to-site link	documented named site or region with matching precision	project country or map-centroid inference
chronology	sample-owned reported value and dating basis	explicit sample group interval carried as contextual or broad	paper year or undifferentiated project period
coordinate	source-supplied sample or verified site coordinate with provenance	documented approximate named-site resolution	arbitrary representative point for a region

When authorities conflict, the intake does not select the most precise value. It records the competing claims, their locators, and the decision basis. A less precise but directly supported value outranks an exact-looking value with unclear ownership.

flowchart TB
    Artifact["captured paper, supplement, or archive row"] --> Locator["stable source locator"]
    Locator --> Claim["identity, locality, chronology, or coordinate claim"]
    Claim --> Conflict{"conflicting support?"}
    Conflict -->|no| Governed["governed sample evidence"]
    Conflict -->|yes| Ledger["conflict and ambiguity record"]
    Ledger --> Decision["qualified resolution or refusal"]
    Governed --> Admission["product-specific admission"]
    Decision --> Admission

Recovery States Are Evidence, Too

The intake registry distinguishes incomplete acquisition from incomplete extraction. These conditions have different remedies and different scientific meaning:

• paper capture blocked — the publication anchor is not readable locally;
• supplement capture blocked — the paper is known, but its sample-bearing files are unavailable;
• sample extraction blocked — readable material exists, but defensible sample rows have not been recovered;
• locality or chronology unresolved — the sample exists, but a public spatial or temporal claim would exceed its evidence;
• publication ready — sample identity and the fields used by the output satisfy the applicable admission rules.

Expected sample counts are also provenance-bearing claims. When the available paper or archive surface is too weak, the registry keeps the count unknown rather than turning an estimate into an apparent fact.

Foundation blockers and intake blockers remain separately addressable. A weak chronology posture can block a temporal claim even when the paper, supplement, sample row, and locality are fully recovered. Conversely, an unavailable supplement can block sample extraction before locality or chronology can be evaluated at all. Reporting both levels prevents “blocked” from becoming an opaque catch-all.

Recovery completeness must therefore name its denominator. “All recovered samples reviewed” describes the extracted rows; it does not mean every sample expected from every paper or project was recovered. Where a trustworthy expected count is unavailable, completeness remains unknown even if every known row is curated.

Recovery And Publication Are Separate Ledgers

The intake surfaces record whether source material was found, captured, parsed, and converted into stable sample evidence. Publication surfaces record whether a governed record satisfies a particular output contract. Keeping these decisions separate preserves important cases:

Recovery state	Publication consequence
source known, supplement unavailable	retain project and paper identity; do not invent sample rows
supplement captured, sample table unextracted	expose recovery work; do not equate capture with sample coverage
sample recovered, locality unresolved	retain the sample; refuse exact-point publication
sample and site resolved, chronology contextual	publish only where the product accepts non-numeric temporal posture
point admitted, project under-recovered	publish the supported member; retain the project-level completeness warning

This is the central collection rule: evidence already fit for a narrow product does not erase known recovery debt, and recovery debt does not erase the evidence that has been defensibly curated.

How To Audit A Sample

Start with the project registry under data/adna/governance/source_library/project_registry.json, then follow the project's source_bundle_path. The bundle connects project and paper records to captured artifacts and supplements. Project-owned recovery records under data/adna/governance/source_library/projects/<project-accession>/ then preserve the source claims:

1. sample_master.json establishes stable sample identity and source lineage;
2. sample_locality_evidence.json and sample_sites.json preserve reported locality, resolution, and site linkage;
3. sample_chronology_evidence.json and sample_chronology_provenance.json preserve temporal claims and their source locators;
4. the species-owned normalized/ directory under data/adna/species/<species-slug>/ exposes sample_records.json, site_evidence.json, and coordinate_provenance.json for downstream products.

Cross-project audits expose missing captures, ambiguous identities, locality conflicts, chronology gaps, and manual-curation work without promoting those records into public points.

An audit should also verify that every published descendant points back to the same governing sample identity. Species views and report rows are projections; they cannot silently split one sample, merge distinct samples, or replace the project-owned authority.

Reading A Visible Point Correctly

A visible point means that the repository can defend the sample at the published spatial and temporal resolution. It does not mean every source field was exact, that all samples from the project were recovered, or that every tracked project reached the same maturity.

For the field-level contracts, continue with sample records, locality evidence, and chronology evidence. The final map admission rules are documented in point publication rules.