bijux-proteomics-knowledge

bijux-proteomics-knowledge is the scientific memory and grounding layer. It records claims, evidence, provenance, context, and contradictions, and it resolves proteomics results against biological entities without turning those records into a recommendation policy.

python -m pip install bijux-proteomics-knowledge

Keep grounding separate from action

A source record, a grounded claim, an evidence-sufficiency decision, and a recommendation are different artifacts with different owners.

Question	Owning decision	Result
was the source identified and represented correctly?	Knowledge ingestion and normalization	versioned evidence record or unresolved identity
what relationship does the source have to this precise claim and context?	Knowledge grounding and reconciliation	support, contradiction, qualification, context-only, or unresolved edge
does the assembled evidence meet the burden for this declared use?	Knowledge sufficiency policy	bounded review bundle, limitation, or knowledge gap
which candidate or action should be preferred under declared constraints?	Intelligence ranking and challenge	recommendation, condition, downgrade, escalation, hold, or refusal
did the authorized follow-up produce an accepted consequence?	Lab readiness, observation, and reconciliation	consequence record returned as evidence without rewriting the earlier claim

A “yes” in one row does not imply a “yes” in the next. In particular, correct identifier resolution cannot establish claim support, and sufficient evidence for description may remain insufficient for prioritization or experiment.

Evidence lifecycle

flowchart LR
    source["literature · ontology · database · run artifact"]
    normalize["normalize\nidentity · context · provenance"]
    resolved{"identity and context resolved?"}
    memory["evidence memory\nrecords · claims · bundles"]
    reconcile{"relationship to claim"}
    ground["ground\nprotein · pathway · feature · disease · drug"]
    assess{"use-specific burden met?"}
    brief["scientific review bundle"]
    gap["knowledge gap or unresolved record"]
    contradiction["contradiction or qualification edge"]
    source --> normalize --> resolved
    resolved -->|no| gap
    resolved -->|yes| memory --> reconcile
    reconcile -->|support| ground
    reconcile -->|contradict or qualify| contradiction --> ground
    reconcile -->|unresolved| gap
    ground --> assess
    assess -->|yes| brief
    assess -->|no| gap

The package preserves disagreement. Contradictory evidence is not averaged away, and missing context is not treated as negative evidence. Resolution decisions retain their source and rule so they can be revisited.

Evidence identity

A normalized evidence record keeps source identity separate from the claim it may support. At minimum, durable evidence preserves:

Dimension	Examples	Why it cannot be inferred later
source identity	DOI, accession, database record, run artifact	titles and labels are not stable keys
source version	release, revision, retrieval time, content digest	databases and documents change
biological context	organism, tissue, condition, perturbation, cohort	the same observation can reverse meaning across contexts
analytical context	workflow, instrument, threshold, model, comparator	technical assumptions bound the claim
relationship	supports, contradicts, qualifies, mentions	co-occurrence is not support
lineage	ingestion, normalization, resolution, review parents	a summary cannot reconstruct its derivation

Public biological grounding

The curated root API exposes typed resolution reports and functions for:

• protein identifiers;
• protein feature overlaps;
• pathway membership and coverage confidence;
• protein-complex membership;
• kinase–substrate relationships;
• disease terms;
• drug–target relationships;
• cross-species orthologs;
• knowledge coverage and schema compatibility.

Each resolution family can render reviewable TSV output in addition to its typed Python result. Root anchors EvidenceRecord, EvidenceClaim, EvidenceBundle, and KnowledgeDecisionBrief connect those biological resolutions to evidence memory and downstream review.

Evidence and reference workflows

The deeper references surface provides citation and context grounding, literature and ontology sources, curated corpora, benchmark ledgers, comparator positions and failures, contradiction dossiers, evidence-sufficiency checks, knowledge-deficit reports, scientific thresholds, reading packs, replay proof, and release-facing narratives.

These workflows preserve the identities and decisions needed to answer the grounding and sufficiency questions above without collapsing them into a source count or narrative confidence label.

Claim anatomy

Element	Required question
proposition	what exactly is asserted, at what granularity?
subject identity	which protein, peptide, feature, pathway, disease, drug, species, or experimental entity?
context	in which organism, tissue, condition, instrument, workflow, and time window?
source	where did the observation or assertion originate, and under which version?
support relationship	does the record support, contradict, qualify, or merely mention the proposition?
quality and limitations	what biases, coverage gaps, indirectness, and uncertainty constrain use?
intended use	is the evidence sufficient for description, prioritization, or an experimental decision?

A source can be authoritative yet irrelevant to the declared context. A correct identifier match can still ground the wrong biological entity or species. Knowledge records resolution and context separately so downstream consumers can distinguish these failures.

Classify each source-to-claim relationship

The relationship is evaluated against a precise proposition, not assigned to the source as a whole. The same paper, database release, or run artifact may support one claim and contradict, qualify, or merely contextualize another.

Relationship	Required test	Example disposition
supports	the observation bears directly on the proposition in a compatible context	add a support edge with scope and limitations
contradicts	a comparable observation is incompatible with the proposition	retain both records and add a contradiction edge
qualifies	the record narrows population, direction, magnitude, mechanism, or conditions	attach the limiting context; do not count it as unqualified support
contextual	the record explains background or plausibility without testing the proposition	retain citation value without adding claim support
unresolved	identity, context, or proposition alignment cannot be established	record a knowledge gap instead of guessing a relationship

flowchart LR
    source["identified source version"] --> context["resolve subject and context"]
    claim["precise claim"] --> relation{"relationship test"}
    context --> relation
    relation --> support["support"]
    relation --> contradict["contradiction"]
    relation --> qualify["qualification"]
    relation --> contextual["context only"]
    relation --> unresolved["knowledge gap"]

Source prestige, citation count, and identifier resolution may inform review, but none selects the relationship without proposition-level comparison.

Memory integrity

Evidence memory uses explicit claim and evidence models, normalized ingestion, reconciliation, and graph-integrity checks. Provenance links the normalized record back to its source; contradiction handling records incompatible claims without discarding either; review briefs summarize the current state without becoming the canonical evidence store.

flowchart TD
    N["new evidence"] --> M["append to memory"]
    M --> R["resolve identity and context"]
    R --> C["reconcile with active claims"]
    C --> S{"relationship"}
    S -->|supports| U["support edge"]
    S -->|contradicts| X["contradiction edge"]
    S -->|qualifies| Q["scope or limitation edge"]
    S -->|unresolved| G["knowledge gap"]
    U --> B["versioned review bundle"]
    X --> B
    Q --> B
    G --> B

New evidence appends to memory and may change the current review. It does not erase the evidence or review state available to an earlier recommendation.

Reconciliation outcomes

Reconciliation does not force one winner. It can identify exact duplicates, retain context-specific variants, mark unresolved identity, connect support and contradiction edges, or supersede a record while preserving its history.

flowchart TD
    A["incoming record"] --> I{"identity resolved?"}
    I -->|no| U["unresolved evidence"]
    I -->|yes| D{"same content and context?"}
    D -->|yes| X["duplicate lineage"]
    D -->|no| R{"relationship to active claim"}
    R --> S["support"]
    R --> C["contradiction"]
    R --> Q["qualification"]
    R --> N["not relevant in context"]

Retrieval success, identifier resolution, and source reputation are necessary inputs to grounding; none alone makes a claim true.

Evidence sufficiency is use-specific

The same evidence bundle can be adequate for description and inadequate for prioritization or experimental action. Sufficiency must therefore name the intended use and apply its burden explicitly.

Intended use	Minimum review burden	Stop condition
describe an observation	resolved subject, source, context, and faithful proposition	identity or context remains unresolved
summarize a body of evidence	source diversity, deduplication, support and contradiction edges, coverage limits	the summary hides conflict or systematic gaps
prioritize a candidate	relevant support, active contradictions, comparator context, and evidence-quality assessment	ranking would depend on missing or indirect evidence
support an experimental decision	candidate burden plus falsifiers, uncertainty, feasibility-relevant context, and explicit review scope	consequential uncertainty has no owner or response
revise an earlier conclusion	new evidence identity, previous bundle identity, reconciliation result, and change explanation	history would need to be overwritten to explain the change

flowchart TD
    EB["versioned evidence bundle"] --> IU{"intended use"}
    IU --> DS["descriptive burden"]
    IU --> PS["prioritization burden"]
    IU --> AS["action-support burden"]
    DS --> RV{"burden met?"}
    PS --> RV
    AS --> RV
    RV -->|yes| SB["sufficiency record with limits"]
    RV -->|no| KG["knowledge deficit or refusal"]

Sufficiency is not stored as an unqualified property of a source or claim. It is a versioned assessment of a named bundle against a named use.

Ownership boundary

Knowledge depends on foundation contracts and core scientific semantics. It does not depend on runtime, intelligence, or lab. Runtime can provide a run bundle as an input artifact; intelligence consumes the resulting scientific review bundle; lab observations can be ingested as new evidence through an explicit handoff. None of those consumers may mutate evidence history through their own policy or operational state.

Reconcile Conflicting Evidence

Conflicting observations are not necessarily inconsistent records. They may refer to different species, tissues, perturbations, analytical policies, time points, or claim granularity. Reconcile identity and context before attempting to reconcile conclusions.

Audit step	Question	Preserve even after resolution
source custody	can each source version and payload be reopened?	source identifiers, retrieval time, digest, and lineage
subject resolution	do both records refer to the same biological entity and granularity?	original identifiers and resolution rule
context alignment	are organism, tissue, condition, cohort, workflow, and comparator compatible?	every mismatched or missing dimension
proposition alignment	do direction, relation, magnitude, and time window express the same claim?	original proposition and normalized claim
quality review	are indirectness, bias, coverage, and analytical uncertainty comparable?	source-specific limitations
disposition	support, contradiction, qualification, context-specific variant, or unresolved?	both records and the rule that produced the edge

flowchart TD
    a["evidence record A"] --> custody["verify source custody"]
    b["evidence record B"] --> custody
    custody --> identity{"same resolved subject?"}
    identity -->|no| unresolved["retain unresolved identity"]
    identity -->|yes| context{"contexts comparable?"}
    context -->|no| variant["retain context-specific variants"]
    context -->|yes| proposition{"same proposition?"}
    proposition -->|no| qualify["record qualification or distinct claim"]
    proposition -->|yes| relation["record support or contradiction edge"]

Use Workflow Consequence Maps to see how the resolved evidence constrains action, What Changed The Recommendation for decision attribution, and Decision Support only after the review bundle has a declared sufficiency posture.

Continue By Evidence Question

Need	Read next	Review is complete when
understand memory, grounding, and review ownership	package overview	source, normalized evidence, claim relationship, review bundle, and consumer authority remain distinct
trace a public claim to support and contradiction	workflow claim grounding	every relationship resolves to source version, subject, context, proposition, and limitation
evaluate curated literature pressure	workflow literature audits	freshness, bibliography identity, benchmark gaps, and comparator gaps have explicit dispositions
inspect memory and reconciliation boundaries	architecture	new evidence appends relationships and review versions without overwriting history
choose Python, data, or artifact contracts	interfaces	an independent consumer can reopen the evidence identity, provenance, relationship, and sufficiency record
review source, coverage, and inference limits	known limitations	missing context, contradiction, indirectness, and coverage gaps remain visible to downstream consumers