Evidence Publication Capabilities¶
Bijux Pollenomics currently operates as an atlas builder and evidence- publication runtime with explicit limits. It collects source families, curates sample and context evidence, reviews claim fitness, ranks declared candidates, and publishes scoped products. The broader harmonization and interpretation engine is a project direction, not a current runtime claim.
That distinction does not reduce the implemented system to rendering. The current product owns material state transitions, evidence contracts, refusal paths, and public descendants. It does mean that readers should judge the runtime by those concrete capabilities instead of assigning it analyses that the public interfaces do not perform.
Operational Capabilities¶
| Capability | Governed inputs | Result |
|---|---|---|
| family-specific source collection | pinned source identities and family-specific acquisition rules | tracked captures with hashes, retrieval context, and replacement semantics where the family contract materializes them |
| evidence-database preparation | captured material, extraction rules, identity contracts, and null semantics | repository-owned records with preserved provenance, typed uncertainty, conflicts, and negative outcomes |
| animal evidence curation | archive projects, papers, supplements, and recovered samples | stable identity, locality, chronology, coordinate, conflict, and recovery records |
| cross-domain publication | admitted evidence plus declared source roles | world, Europe-plus, Nordic, and country bundles |
| geographic traceability | product manifests, feature IDs, and evidence owners | map members that resolve back to governed records and sources |
| lake decision support | SVAR identities, contextual evidence, ranking model, and sensitivity scenarios | ranked candidates with input roles and stability evidence |
| release accountability | coverage, drift, exclusion, and integrity reviews | passing checks, qualified claims, and explicit refusals |
These capabilities produce checked-in structured artifacts, not only prose or screenshots. Each state-changing operation has an owned input boundary, manifested output, and review surface. They constitute an evidence-publication platform; they are not evidence that the planned general engine already exists.
Operational here means that an owned contract exists and is exercised for the named result. It does not mean that every family has the same artifacts. The current lifecycle matrix records full capture, normalization, review, and publication materialization for Neotoma, SEAD, and animal ancient DNA. Other families retain narrower stage combinations: LandClim, RAÄ, and boundaries lack a materialized review stage, and SVAR retains capture and publication evidence without materialized normalized or review artifacts. AADR retains capture and publication evidence plus a compact, non-admitting source-accountability review; normalized membership and qualified scientific review remain unavailable. Those limits are explicit capability boundaries, not stages inferred from a successful downstream build.
Database Preparation Is Executable Evidence Work¶
Preparation is not a clerical prelude to analysis. It is where source-native observations become accountable repository evidence:
flowchart LR
Receipt["source receipt"] --> Parse["parse without semantic promotion"]
Parse --> Normalize["normalize identity, units, and nulls"]
Normalize --> Relate["resolve samples, projects, places, and citations"]
Relate --> Curate["record conflicts, exclusions, and uncertainty"]
Curate --> Admit["apply claim-specific admission"]
Admit --> Publish["manifested product"]The preparation capability is demonstrated by retained lineage and decisions, not merely by the existence of a table. A defensible record carries source identity, extraction context, normalization semantics, relationship evidence, and its review or blocking posture. Where one of those surfaces is not materialized for a family, the capability claim stops at the last observable stage.
Test A Capability Claim¶
A capability claim is complete only when a reader can resolve each part of the operation:
| Proof question | Required answer |
|---|---|
| who owns the behavior? | a canonical runtime module rather than a renderer, compatibility alias, or workflow wrapper |
| how is it invoked? | a supported command or Python entry point with typed or documented arguments |
| what state does it consume? | explicit source identities, versions, roots, schemas, and required prior decisions |
| what effect can it have? | read-only inspection or a bounded replacement of named governed roots |
| what result survives the process? | a structured return value plus manifests, records, decisions, or publication artifacts where the operation writes state |
| how can failure be interpreted? | execution errors remain distinct from qualification, exclusion, refusal, and release blocking |
| what verifies the contract? | focused checks over behavior, identity, membership, lineage, and semantic invariants |
This test distinguishes a callable helper from a product capability. It also distinguishes an operational producer from repository readiness: the code may own all seven answers while the selected evidence revision still lacks a required source, field, or decision.
Qualified Capabilities¶
Some real capabilities carry narrower claims because the evidence is uneven:
- animal source recovery tracks 40 projects, 1,455 recovered raw sample rows, and 1,450 final rows, while expected-sample denominators remain incomplete;
- animal point publication admits 170 sample-backed locality features representing 331 distinct samples without presenting locality and sample counts as one homogeneous population;
- Neotoma provides 170 numerically comparable site spans alongside five contextual-only and 25 unresolved sites;
- SEAD provides 2,069 assigned four-country site features plus 9,727 Swedish chronology-discovery features; its 25,109 claims partition into 14,264 comparable, 10,144 context-only, 60 explicit temporal-contract refusals, and 641 unresolved claims; and
- Sweden lake ranking supports prioritization, while field readiness remains dependent on bathymetry, access, permissions, and on-site verification.
The animal database also contains different governed populations: 1,455 recovered raw project sample-master rows, 1,450 final sample-foundation rows, 331 distinct admitted samples, and 170 point-publication locality features. These are distinct contracts rather than a single attrition funnel. The public layer contains 117 domesticated-core and 53 wild/progenitor-context features, all with admitted sample backing.
flowchart TB
Inventory["collected and curated population"] --> Review{"claim-specific evidence review"}
Review -->|supported| Admit["admitted product member"]
Review -->|limited| Qualify["qualified role or precision"]
Review -->|insufficient| Exclude["exclusion or recovery record"]
Admit --> Product["manifested publication"]
Qualify --> Product
Exclude --> Account["accountability surface"]The engine is credible because all three branches are durable outputs.
Capability Ledger¶
Operational status is claim-specific. The same domain can be operational for one output, qualified for another, and outside scope for a stronger analysis.
| Domain question | State | Governed result | Claim ceiling |
|---|---|---|---|
| Which AADR v66 rows belong to a country bundle? | operational | release-resolved members, manifest, table, and GeoJSON | geographic publication of metadata, not genotype analysis |
| Which recovered animal samples meet the point contract? | operational for 331 samples represented by 170 locality features | final sample identity, locality, coordinate, chronology posture, and traceability | admitted subset, not complete project recovery |
| May the Wadi Halfa dromedary context appear spatially? | not published | readiness evidence retained with no_admitted_sample_backed_locality_candidate |
context remains outside the point product |
| Are current SEAD sites contemporaneous with nearby aDNA? | qualified per record | 25,109 chronology claims include 14,264 comparable claims, while context-only, refused, and unresolved claims remain ineligible | interval overlap is contextual evidence, not proof of association |
| Which Swedish lakes rank under declared scenarios? | qualified decision support | ranking, sensitivity, and fieldwork-preparation packets | prioritization, not sampling readiness |
| What population-genetic process produced a pattern? | outside scope | no governed capability | requires a new analysis and evidence contract |
This ledger prevents a mature command or attractive visualization from lending its status to a stronger question. Capability state follows the claim being asked, not the package or source family as a whole.
flowchart LR
Question["declared scientific or product question"] --> Inputs["owned inputs"]
Inputs --> Contract{"governed capability contract?"}
Contract -->|no| Outside["outside scope"]
Contract -->|yes| Fitness{"evidence fitness"}
Fitness -->|complete for claim| Operational["operational result"]
Fitness -->|bounded| Qualified["qualified result"]
Fitness -->|insufficient| Refused["refusal and recovery evidence"]Planned Engine Surfaces¶
The public surface contract identifies three planned capabilities:
| Planned surface | Missing current contract |
|---|---|
| multi-evidence harmonization runtime | no general observation-unit alignment, cross-family transformation, and governed harmonized output |
| evidence-aware scoring and interpretation engine | no general inference contract connecting domain evidence to scientific interpretation |
| workflow replay and diff execution | no public runtime that replays arbitrary governed stages and explains semantic differences between runs |
Existing commands may inspect, rebuild, or compare particular governed surfaces. Those focused operations should not be generalized into these broader engine claims.
Outside Current Scope¶
No current contract supports:
- AADR genotype processing or population-genetic inference;
- automatic causal inference across pollen, archaeology, and ancient DNA;
- synthetic chronology for records whose source does not own numeric time;
- exact geolocation from broad locality or project geography;
- autonomous field-site selection or coring instructions;
- continent-wide claims from Sweden-specific RAÄ context; or
- a single composite evidence score that erases family roles and uncertainty.
An output that appears to provide one of these results would be outside the declared product, even if it could be computed from nearby columns or map geometry.
Extension Contract¶
A new source family or analytical capability becomes part of the current product only when it has:
- stable upstream identity, version, licence, and acquisition lineage;
- a declared observation unit and normalized schema;
- fact ownership, null semantics, and conflict behavior;
- spatial and temporal posture;
- a distinct evidence role and product question;
- admission, qualification, and exclusion rules;
- manifests and traceability for every public descendant; and
- focused verification that detects semantic drift.
This contract allows the system to grow without describing intention as implementation. Until all eight relations exist, the proposed surface remains external, exploratory, or planned rather than a published runtime capability.
Continue to runtime scope and ownership, publication scope, and the data system.