Source Selection And Refresh

A source enters Bijux Pollenomics because it answers a declared question and can retain identity, use conditions, semantics, and limits through curation. Discovery is not admission, and a successful refresh is not evidence that the source's meaning stayed unchanged.

Selection Rules

Criterion	Required answer
scientific role	Is the source direct evidence, context, sampling context, or geographic framing?
identity	Can its dataset, release, accession, DOI, record, or other upstream identity be preserved?
access and use	Can retrieval context, license posture, and relevant restrictions be recorded?
recoverability	Can the required payload, table, supplement, or API response be captured reproducibly?
semantics	Can place, time, taxonomy, and identifiers be represented without false precision?
reviewability	Can ambiguity, missingness, conflict, and exclusions remain explicit?
publication role	Which products may consume it, and under which precision and scope?
sustainability	Can refresh and failure behavior be maintained without private untracked authority?

A source may be scientifically important while remaining in recovery or context-only posture. Admission records that distinction instead of forcing a binary “available” label.

Selection Record

A defensible selection records the question before the source. It names the target object and evidence role, explains why the source is suitable, identifies known alternatives or gaps, and states the conditions under which the source would be demoted or replaced. This prevents familiarity, convenient access, or large row counts from becoming unstated selection policy.

For literature-backed animal evidence, selection also distinguishes four levels that are often collapsed: a project is known; its paper is identified; its supporting material is recovered; and sample-owned locality and chronology are supported. Progress at one level does not imply completion of the next.

The selected family has a named evidence owner and a named preparation owner. The evidence owner is accountable for observation-unit meaning, authority, bias, and permitted claims. The preparation owner is accountable for capture, normalization, validation, and replacement behavior. One person or component may hold both responsibilities, but the decision record keeps them explicit so that operational success cannot stand in for scientific acceptance.

Selection Bias Remains Visible

The collected families are a governed evidence portfolio, not proof that every relevant source ecosystem has been captured. Selection records therefore retain the denominator and bias introduced by access, jurisdiction, language, digitization, publication practice, and source-specific preservation.

Selection condition	Potential bias	Required qualification
API or bulk access is available	easily retrievable sources may dominate	distinguish convenience from scientific priority
one national registry is mature	national density can look regionally representative	keep jurisdiction explicit
supplements are recoverable for some projects	recovered projects can look biologically dominant	report expected and recovered project/sample denominators
chronology is numeric in one family	numeric records can outrank relevant contextual evidence	preserve temporal role instead of rewarding format alone
stable coordinates are available	mappable evidence can displace important broad-area records	keep unresolved and non-point evidence visible

A new source is admitted because it fills a declared evidentiary role under a reviewable contract, not merely because it increases coverage or produces a denser map. Known alternatives and uncaptured ecosystems remain part of the selection boundary.

flowchart LR
    Question["declared evidence question"] --> Candidate["candidate source"]
    Candidate --> Role["evidence role and observation unit"]
    Role --> Recovery["recoverability and use conditions"]
    Recovery --> Semantics["identity, place, time, taxonomy"]
    Semantics --> Decision{"admission posture"}
    Decision -->|direct| Direct["direct evidence contract"]
    Decision -->|context| Context["context contract"]
    Decision -->|recover| Gap["tracked recovery gap"]
    Decision -->|refuse| Refusal["reasoned refusal"]

Refresh Lifecycle

flowchart LR
    Declare["declared source and version"] --> Stage["isolated capture"]
    Stage --> Normalize["family-aware normalization"]
    Normalize --> Validate["identity and contract checks"]
    Validate --> Compare["semantic and coverage comparison"]
    Compare --> Decision{"accept complete family state?"}
    Decision -->|yes| Replace["replace governed family tree"]
    Decision -->|no| Preserve["preserve prior governed tree"]
    Replace --> Reassess["reassess reviews and products"]

Staging protects the prior coherent tree from partial acquisition. Acceptance requires more than a nonzero row count: expected assets, identity, hashes, schema, semantics, and family contract must agree.

Replacement is atomic at the owned family-tree boundary. A candidate is built in an isolated sibling tree; the previous tree is moved to a recovery sibling; the candidate is promoted; and a failed promotion restores the previous tree. After successful promotion the recovery sibling is removed. This protects one owned tree from acquisition and promotion failure. It is not a transaction across several source families, the review database, and published products; those downstream surfaces are reassessed through explicit propagation records.

Refresh Acceptance Evidence

Layer	Evidence to compare
source	dataset identity, release, retrieval route, response context, license posture, and hashes
capture	expected versus recovered assets and any access or parsing refusal
normalization	stable member identities, field semantics, geometry, chronology, taxonomy, and missingness
coverage	denominators, added and removed members, conflicts, and precision classes
curation	changed linkage, governing fact owner, qualifications, and open recovery work
publication	affected admission decisions, bundle membership, warnings, and exclusions

An accepted refresh can legitimately produce no publication change. That outcome is meaningful when the record shows that changed source material did not alter normalized meaning or did not satisfy a product contract.

Refresh Proves Replacement, Not Equivalence

Acceptance establishes that the candidate can replace the governed family state. Equivalence is a narrower conclusion that must be demonstrated for the dimension in which it is claimed. The same source label can identify different bytes; identical row counts can conceal member replacement; and unchanged members can acquire different locality, chronology, taxonomy, or use posture.

A comparison packet separates those possibilities:

Comparison	Evidence	Conclusion it permits
acquisition	upstream identity, release, retrieval route, response context, and digest	whether the captured object is the same object
population	stable member identities plus added, removed, merged, and split sets	whether the governed population is equivalent
semantics	field mappings, nulls, units, precision, and normalized value diffs	whether retained members mean the same thing
authority	license, source role, fact ownership, and provenance-edge changes	whether the same uses remain defensible
consequence	review, admission, manifest, exclusion, and presentation diffs	whether publication posture is unchanged

flowchart LR
    Candidate["candidate capture"] --> Identity{"same upstream identity?"}
    Identity --> Members{"same member identities?"}
    Members --> Meaning{"same normalized meaning?"}
    Meaning --> Authority{"same authority and use posture?"}
    Authority --> Consequence["explain publication consequences"]
    Identity -. no .-> Reconcile["record replacement or refuse"]
    Members -. no .-> Reconcile
    Meaning -. no .-> Reassess["reassess evidence and products"]
    Authority -. no .-> Reassess

An accepted refresh may therefore be non-equivalent and still correct: new members, repaired identities, or stronger evidence can justify replacement when their consequences are explicit. Conversely, matching checksums for a repackaged derivative do not establish semantic equivalence unless the derivation and governed member population are also reconciled.

Classify The Change

Observed change	Required interpretation
byte or packaging only	demonstrate unchanged normalized meaning before calling it neutral
upstream version or retrieval route	update provenance and assess comparability
new records	review denominators, coverage, conflicts, and product membership
removed records	distinguish upstream removal, corrected duplication, scope change, and collection failure
changed locality or chronology	rerun evidence fitness and point-admission review
changed taxonomy or identifiers	review joins, species ownership, and downstream traceability
changed license or access posture	reassess continued capture and publication eligibility
newly recovered supplement	revisit sample identity, completeness, place, time, and affected products

Deletion is evidence that requires a reason. A lower count cannot be accepted as harmless merely because the new files pass structural validation.

Refuse A Refresh When Meaning Cannot Be Reconciled

Keep the last governed snapshot when a candidate refresh cannot answer any of these questions:

Unresolved condition	Why replacement is unsafe
upstream members cannot be matched to retained identities	additions, removals, merges, and splits cannot be distinguished
schema change has no semantic mapping	field continuity, null meaning, units, or precision are unknown
retrieval identity or terms are incomplete	origin, reproducibility, or permitted reuse cannot be established
normalization succeeds only by dropping rejected rows silently	the source population and missingness denominator are lost
review inputs changed but decisions were not re-evaluated	publication posture may refer to superseded evidence
product membership changed without causal attribution	evidence, boundary, rule, and rendering changes are conflated

A refresh refusal is not a claim that the upstream release is defective. It means the repository cannot yet replace its governed state without losing an authority or explanatory link.

Propagation

flowchart TB
    Source["source change"] --> Normalized["normalized diff"]
    Normalized --> Review["review posture"]
    Review --> Membership["publication membership"]
    Membership --> Product["map, report, or table"]
    Source --> Provenance["version, retrieval, license, hashes"]
    Provenance --> Review

The absence of a downstream diff is itself a result to explain. It may mean the changed records remain outside product scope, were excluded by the same rule, or did not alter normalized meaning.

Demotion And Retirement

A source can move to a narrower role when access, semantics, or maintenance no longer support its current use. Existing provenance and review history remain visible; public products are reassessed; and replacement sources do not silently inherit the retired source's authority.

Refresh cadence therefore does not determine maturity. A recently collected context layer may support less than an older, well-curated direct-evidence record, and an unavailable supplement may remain the decisive gap in an otherwise current project.

Replacement never erases provenance. The superseding source receives its own identity and selection record; affected normalized records and products retain the chain needed to distinguish a source substitution from newly discovered scientific evidence.

Migration Pressure

A source schema, access route, identifier system, licence posture, or observation model can change without announcing a new scientific meaning. Migration pressure is the set of differences that the existing preparation contract can no longer absorb safely.

Pressure	Required response
renamed or split fields	map source expressions explicitly and review null, precision, and unit semantics
changed native identifiers	publish equivalence, merge, split, and collision decisions before replacing stable identities
new endpoint or archive packaging	preserve the new retrieval identity and prove member equivalence
changed spatial or temporal model	revise the family contract and re-evaluate comparison and publication fitness
discontinued access or restrictive terms	retain the last governed snapshot and narrow refresh or redistribution claims
replacement source	create a new source identity and selection decision; never inherit authority silently

A migration is complete only when source, normalized, review, and publication populations reconcile by identity. Matching totals or a structurally valid schema does not prove unchanged meaning.