Change Evidence

A changed map, count, ranking, or review posture has a cause. Bijux Pollenomics separates changes in source material, curation, product scope, analysis, and rendering so a visible difference can be interpreted instead of treated as an unexplained new result.

Causal Chain

flowchart LR
    Source["source capture"] --> Normalization["normalization"]
    Normalization --> Curation["curation decision"]
    Curation --> Admission["product admission"]
    Admission --> Analysis["comparison or ranking"]
    Analysis --> Rendering["presentation"]
    Admission --> Rendering
    Rendering --> Change["visible product change"]

Several causes can affect one product, but they should remain distinguishable in its manifests, evidence rows, reviews, and generated diff.

Change Classes

Class	Typical visible effect	Evidence needed to interpret it
source	added, removed, or revised upstream records	version, retrieval context, hashes, license posture, and source diff
normalization	changed identifiers, geometry, dates, or field representation	source-native value, normalization basis, and affected records
curation	changed linkage, ambiguity resolution, evidence class, or precision	governing decision, reason, and prior posture
admission	changed membership, qualification, or exclusion	named product rule, scope, and decision record
analysis	changed rank, score, sensitivity, or comparison	method identity, inputs, scenarios, and stability evidence
rendering	changed layout, labels, colors, or interaction	proof that structured membership and meaning stayed unchanged

Validation Layers

flowchart TB
    Shape["schema and format"] --> Relations["identity and relationships"]
    Relations --> Semantics["source-family meaning"]
    Semantics --> Fitness["evidence fitness"]
    Fitness --> Scope["product membership and scope"]
    Scope --> Presentation["cross-format presentation"]

Each layer depends on the preceding layer but answers a different question. A syntactically valid record can still point to the wrong source; a correctly linked record can still have unresolved chronology; a strong evidence record can still fall outside one product; and a correct manifest can still be misrepresented by a rendering.

Validation should therefore stop at the first failed layer and preserve its finding. Passing later presentation checks cannot compensate for an upstream identity or evidence failure.

Breadth Rule

Documentation breadth is semantic coverage, not page count. A change preserves breadth only when a reader can still discover the same governing boundaries, understand their meaning, operate them safely, and trace the evidence promised by the public interface. Combining pages is valid when the resulting page retains those capabilities; leaving a title or redirect without the governing content is not.

Documentation change	Acceptance evidence
page renamed or moved	navigation, inbound links, stable subject, and governing anchors resolve at the new path
pages merged	every distinct contract, procedure, limit, and proof path remains findable in the merged surface
page retired	a named replacement answers the same reader question, or the public capability is explicitly withdrawn
command or data contract changed	reference, examples, defaults, mutation boundary, and failure semantics agree with runtime behavior
evidence family changed	source role, lifecycle materialization, authority path, coverage, and publication consequences remain explained
generated report changed	its handwritten entry point still explains meaning, authority, interpretation limits, and how to inspect the governed state

A passing link check proves reachability; it does not prove breadth. Review the reader questions lost or gained, the authority paths affected, and the claims that would become impossible to audit. A smaller documentation tree is an improvement only when it removes duplication without removing meaning.

Counts Are Not Explanations

An increased count may reflect recovered evidence, broader scope, repaired deduplication, or a weakened admission rule. A decreased count may reflect upstream removal, corrected identity, narrower scope, or failed collection. The number alone cannot identify which happened.

Stable counts can also hide meaningful change: one member may replace another, coordinates may become less precise, chronology may be reclassified, or a direct-evidence row may become contextual.

Build A Semantic Diff Packet

A reviewable change packet compares identities before totals and meaning before rendering. For every affected population, retain:

Packet component	Required content	Question answered
boundary identity	source family, evidence surface, product, version, and scope	what population is being compared?
member sets	stable identifiers added, removed, retained, and replaced	did the population change even if its size did not?
property changes	source identity, role, geometry, chronology, precision, qualification, and fact owner	did retained members change meaning?
relationship changes	joins, aliases, sample–site links, comparison edges, and lineage independence	did the evidence graph change?
decision changes	admission, qualification, exclusion, refusal, and reason	did policy or evidence fitness change visibility?
presentation parity	structured formats and rendered views compared against the same manifest	is the visible change faithful to governed state?

flowchart LR
    Before["prior member identities"] --> Sets["added, removed, retained, replaced"]
    After["current member identities"] --> Sets
    Sets --> Meaning["property and relationship diff"]
    Meaning --> Decisions["admission and qualification diff"]
    Decisions --> Views["cross-format presentation diff"]
    Views --> Explanation["causal change statement"]

The packet should make a zero-count change visible. If one member disappears and another enters, or a retained member loses coordinate precision, the total may stay constant while the evidence population changes materially.

Reading A Product Difference

1. Identify the product and recorded scope.
2. Compare bundle membership and feature identifiers.
3. Separate additions, removals, and modified members.
4. Follow modified members to their admission and governing evidence records.
5. Compare source identity, curation reason, precision, role, warnings, and exclusions.
6. Treat rendering-only change as neutral only when structured meaning is demonstrably unchanged.

The final explanation should name both cause and consequence: for example, “two members were excluded after sample-owned locality review” or “labels changed while manifest membership and feature properties remained stable.” “The product was regenerated” describes an operation, not the change.

Evidence Required By Change Type

Changed surface	Minimum review evidence
source-family tree	source identity, version, capture diff, normalized identity diff, coverage, and removals
evidence or governance record	governing source chain, previous decision, new reason, precision, and affected consumers
product scope or admission	old and new rule, member-level diff, qualifications, exclusions, and affected bundles
analysis output	exact inputs, method identity, scenarios, sensitivity, and interpretation boundary
public bundle	manifest and member diff, traceability, cross-format consistency, warnings, and citations
narrative only	proof that no governed identity, role, scope, precision, or membership changed

These layers expose what a result changed and how that meaning can be traced. They do not by themselves establish historical correctness or fitness for a question outside the named contract.

Interpretation Outcomes

Outcome	Appropriate statement
stronger evidence recovered	the named records gained a stated source-backed property
qualification introduced	the records remain visible under a narrower claim
scope changed	membership changed because the product boundary changed
analysis changed	ranking or comparison changed under a named method or scenario
presentation changed only	the view changed while structured evidence and membership remained stable
cause unresolved	do not interpret the visible difference as scientific change

Change evidence protects against a common error: describing every regenerated product as new scientific evidence. Regeneration is an operation; the causal record determines what, if anything, changed scientifically.