Evidence foundations¶
Knowledge owns the durable relationship between a claim and the evidence that supports, contradicts, qualifies, or fails to resolve it. Its memory preserves source identity, biological context, provenance, confidence, and history before Intelligence applies decision policy or Lab adds experimental consequence.
flowchart LR
S["source or observation"] --> P["provenance"]
P --> G["identity and context grounding"]
G --> C["claim relationship"]
C --> R["reconciliation"]
R --> B["versioned review bundle"]
R -. contradiction .-> X["contradiction ledger"]
R -. missing support .-> D["knowledge deficit"]
Evidence routes¶
| Question | Guide | Governing evidence |
|---|---|---|
| What does Knowledge own? | Package overview | memory, grounding, reconciliation, and review bundles |
| Does a record belong here? | Ownership boundary | evidence state rather than downstream policy |
| Which biological and review capabilities exist? | Capability map | typed resolvers, evidence memory, references, and audits |
| What must remain outside? | Scope and non-goals | algorithm ownership, ranking, execution, and lab authority |
| Is public language grounded? | Workflow claim grounding | exact source, context, support, and contradiction chain |
| Is the scientific backdrop current? | Workflow literature audits | bibliography, corpus, gap, and freshness record |
This package does not own resolves boundary cases where evidence state is being confused with a recommendation or an operational result.
Evidence protocol¶
- Identify the source, version, retrieval context, and applicable license.
- Normalize the record without losing the original source identity.
- Resolve biological identifiers and contextual qualifiers explicitly.
- Attach the record to a precise proposition as support, contradiction, qualification, mention, or unresolved material.
- Reconcile duplicates and conflicts without discarding independent lineage.
- Assess coverage, directness, quality, freshness, and intended-use sufficiency.
- Produce a versioned review bundle that retains support, contradiction, and deficits.
sequenceDiagram
participant S as Source
participant K as Knowledge
participant I as Intelligence
participant L as Lab
S->>K: evidence with provenance and context
K->>K: ground, reconcile, retain contradictions
K->>I: immutable review bundle
I->>L: recommendation referencing bundle
L-->>K: observation as new evidence
The loop appends knowledge. It never allows a later outcome to rewrite the source and evidence available to an earlier decision.
Grounding dimensions¶
Biological grounding is multidimensional. The package exposes resolution for protein identifiers, features, pathways, complexes, kinase–substrate relationships, diseases, drug targets, and orthologs. A successful match in one dimension does not settle the others.
For example, an identifier may resolve while the species is wrong; a feature may overlap while the experimental condition differs; a pathway membership may be correct while coverage is too sparse for the intended conclusion. Review bundles retain these distinctions instead of flattening them into one score.
Contradiction and sufficiency¶
Contradiction is a first-class relationship, not a failed ingestion. A review can contain strong support and strong contradiction simultaneously. Confidence describes the current evidence posture under declared context; it is not a ranking score and does not authorize an action.
Evidence sufficiency is always relative to intended use:
| Intended use | Typical burden |
|---|---|
| descriptive context | resolved identity, source provenance, and relevant context |
| scientific interpretation | multiple relevant records, limitations, contradictions, and coverage assessment |
| candidate prioritization | versioned review bundle suitable for declared decision policy |
| experimental action | decision record plus Lab feasibility, risk, and readiness review |
Missing evidence remains a knowledge deficit. It is not automatically negative evidence, and it cannot be converted into support by confident prose.
Evolution rules¶
The domain language stabilizes claim, evidence, source, context, grounding, contradiction, and sufficiency. The change principles preserve source lineage and review history when schemas or resolution policy evolve. Dependencies and adjacencies and repository fit keep the package independent of Intelligence, Runtime, and Lab, while lifecycle overview defines how evidence moves from ingestion to reviewed memory.