Runtime Scope And Ownership¶
bijux-pollenomics is the canonical runtime that turns explicit source and
repository inputs into governed data, review, ranking, and publication
artifacts. Each family materializes only the stages its current contract owns;
a downstream product does not prove that an absent normalized or review
artifact exists. Ownership follows scientific meaning: collection owns
acquisition, evidence modules own normalized claims, review owns fitness,
product contracts own membership, and reporting owns presentation.
Domain Language¶
| Term | Meaning in this runtime | Common misreading |
|---|---|---|
| capture | retained upstream material plus retrieval identity | evidence already fit for a claim |
| governed record | repository-owned identity and claim representation with source lineage | a generic row whose columns are self-authoritative |
| review | claim- and use-specific evaluation of fitness, precision, conflict, or recovery | a universal quality score |
| admission | decision that one governed record belongs to one product scope and role | proof that every claim on the record is analytically comparable |
| publication | manifested members, warnings, traceability, and renderings | a new source of scientific facts |
| refusal | durable result that preserves why a stronger claim or representation is unsupported | missing work that may be discarded |
These terms name ownership transitions. “Collected,” “reviewed,” and “published” are not interchangeable maturity badges, and a later state never retroactively supplies an earlier missing authority.
Lifecycle And Surface Map¶
flowchart LR
Command["CLI or Python request"] --> Collect["source collection"]
Collect --> Stage{"family contract and materialized stage"}
Stage --> Normalize["owned evidence records"]
Stage --> Direct["retained capture or family-specific descendant"]
Normalize --> Review["fitness, conflict, and comparability"]
Review --> Product["scope, role, and admission"]
Direct --> Product
Product --> Publish["structured bundles and presentation"]
Publish --> Verify["traceability, subset, and release checks"]
Verify -. correction .-> Normalize
The arrow into publication carries admitted evidence and declared roles. It does not transfer upstream fact ownership to the report writer. A correction starts at the earliest governing record and then regenerates its descendants. The direct path is not a shortcut around scientific review; it makes the current family-specific lifecycle observable. Claims that require a missing stage remain unavailable until that stage is governed and materialized.
Ownership Boundary And Map¶
| Runtime boundary | Owns | Durable result |
|---|---|---|
data_downloader and adna.sources |
retrieval, decoding, source identity, and captured artifacts | versioned family trees and animal source dossiers |
adna.projects and normalization modules |
stable samples, sites, chronology, coordinates, and family records | repository-owned evidence state |
evidence and analysis.review |
ambiguity, precision, comparability, recovery, and refusal | review ledgers and fitness outcomes |
foundation and analysis modules |
product scope, evidence role, admission, ranking, and sensitivity policy | machine-readable product and decision contracts |
reporting assembly |
member relationships, bundles, and structured public rows | world, regional, country, and specialized product state |
reporting presentation |
Markdown, HTML, tables, maps, and reader navigation | derived human-facing artifacts |
The tracked data/ tree is governed evidence state. docs/report/ contains
derived publication and accountability state. Public guides explain how to
read those artifacts; they do not become a third database.
An ownership boundary is crossed only through a declared contract. Repeating a chronology in a popup does not transfer chronology ownership to the popup; calling a collector from the CLI does not transfer source-family semantics to the command parser. Corrections begin at the earliest authority and flow forward through typed descendants.
Effect Authority And Fact Authority¶
The component allowed to write a file is not necessarily the component that owns the scientific meaning inside it:
| Authority | Governs | Example |
|---|---|---|
| effect authority | whether an operation may create, replace, or publish a bounded tree | the publication assembly owns replacement of its staging and output roots |
| fact authority | identity, place, time, taxonomy, precision, and source role | the evidence record and its source-family contract own claims repeated in a bundle |
| decision authority | fitness, admission, ranking, qualification, or refusal for one declared use | a review or product contract owns why a record appears or remains absent |
| presentation authority | faithful labels, grouping, links, and visual encoding | a renderer owns display shape but cannot strengthen the underlying claim |
A single operation may exercise all four through typed boundaries, but they remain independently auditable. If a published chronology is wrong, changing only the popup would create disagreement; the correction belongs at the fact authority, followed by review, admission, and publication regeneration.
Capability Map¶
| Capability | Current owner | Current result | Claim ceiling |
|---|---|---|---|
| source collection | family collectors and animal source intake | versioned captures, metadata, hashes, and source dossiers | capture presence, not scientific fitness |
| evidence preparation | family normalization and animal evidence modules | stable records, relations, provenance, and unresolved states | represented meaning, not automatic product admission |
| scientific review | evidence and analysis review modules | conflicts, precision, comparability, recovery, and refusal | the declared claim and use only |
| product assembly | product contracts and reporting assembly | world, regional, country, lake, and fieldwork membership | manifested members within one scope |
| presentation | reporting renderers | maps, tables, Markdown, and HTML | faithful rendering of admitted state |
| repository verification | maintainer package and focused tests | contract findings and release stops | agreement for the checked boundary, not new scientific authority |
The map describes implemented owners. General multi-evidence harmonization, evidence-aware interpretation, and workflow replay remain outside the current runtime capability boundary even where the repository contains design direction for them.
Command Semantics¶
Commands fall into two operational classes:
| Class | Representative commands | Contract |
|---|---|---|
| orientation | product-scope, surface-map, ownership-map, source-support, adna-species |
read declared capabilities without rebuilding governed state |
| state-changing | collect-data, refresh-animal-adna-foundation, publish-reports |
consume explicit roots, write governed artifacts, and require semantic review |
A successful process exit means software execution completed. It does not mean every requested scientific claim passed. State-changing operations can legitimately emit qualified results, empty admitted subsets, exclusions, or a release refusal while returning successful execution.
Four Success Conditions¶
| Condition | Question answered | What failure requires |
|---|---|---|
| execution | did the command complete its software contract? | diagnose invocation, input, or runtime failure |
| evidence fitness | do identity, place, time, coordinates, and role support the claim? | recover, qualify, or refuse evidence |
| admission | does the record satisfy the declared product rule? | retain an accountable exclusion or change the governed rule |
| publication integrity | do manifest, evidence rows, traceability, and presentation agree? | block release and regenerate descendants |
flowchart TD
Run["successful execution"] --> Fit{"evidence fit for claim?"}
Fit -->|no| Account["qualification, exclusion, or recovery"]
Fit -->|yes| Admit{"admitted by product contract?"}
Admit -->|no| Account
Admit -->|yes| Integrity{"publication descendants agree?"}
Integrity -->|no| Block["block publication"]
Integrity -->|yes| Release["governed product"]
No later condition upgrades an earlier one. A map build can be technically successful while a row remains scientifically unresolved, and a valid evidence record can remain outside a particular publication scope.
Operation Evidence Packet¶
A reproducible runtime result includes:
- command or API identity and explicit arguments;
- governed input roots and source or product version;
- software outcome and diagnostics;
- written manifest and artifact membership;
- semantic changes to identities, fields, precision, or membership;
- admitted, qualified, excluded, and unresolved outcomes; and
- focused verification against the affected contract.
Terminal output alone cannot prove what entered a product. Generated files without invocation and input identity cannot prove how they were produced.
Dependencies And Adjacencies¶
| Distribution | Responsibility | Boundary |
|---|---|---|
bijux-pollenomics |
canonical scientific and publication runtime | owns schemas, commands, evidence rules, and outputs |
pollenomics |
short executable and import compatibility | forwards to the canonical runtime; owns no scientific fork |
bijux-pollenomics-dev |
repository-only documentation, validation, and release support | must not own public runtime behavior or scientific truth |
Repository workflow policy can decide when to run the runtime, but it cannot change evidence meaning. Public documentation can explain an output, but it cannot admit a record. The compatibility package can rename an entry point, but it cannot produce a distinct schema or scientific result.
Dependencies used for coordinate transformation, safe XML processing, serialization, or rendering supply mechanics. They do not own the repository's source roles, claim precision, or admission policy. An adjacency becomes a product dependency only through a named interface and an explicit transfer of typed data—not because both systems can read the same file format.
Change Principles¶
- Change a fact at its earliest governing record, then regenerate declared descendants.
- Preserve source-native identity and wording before normalizing shape.
- Keep collection, fitness, admission, and rendering as separate decisions.
- Prefer a qualified or refused result to invented precision or silent loss.
- Require explicit roots for operations that read or replace governed state.
- Treat manifests, exclusions, warnings, and traceability as one publication contract.
- Expand the runtime only through a durable owner, typed boundary, and focused proof.
These principles keep a correction local to its authority while still making its publication impact visible. They also prevent compatibility adapters, workflow routers, and renderers from becoming accidental scientific owners.
Explicit Limits¶
The runtime does not:
- process AADR genotype data;
- infer missing sample coordinates or chronology from nearby records;
- treat map proximity as temporal overlap or causation;
- turn archaeology density, boundaries, or lake identity into direct evidence;
- certify access, bathymetry, permits, or coring suitability; or
- make every tracked source record eligible for publication.
Continue to evidence engine capabilities, repository scope and limits, and the runtime architecture.