Risk Register

Knowledge failures often preserve valid-looking records while corrupting what readers are allowed to infer. The risk surface spans identity, source custody, relationship direction, uncertainty, reconciliation, coverage, and downstream interpretation.

flowchart LR
    S["source and version"] --> E["evidence record"]
    E --> C["claim"]
    C --> G["relationship graph"]
    G --> H["review handoff"]
    X["contradiction or ambiguity"] --> G

Scientific And Custody Risks

Risk	Observable failure	Required control
identity ambiguity	one symbol resolves to the wrong protein, species, or isoform	namespace, organism, version, and ambiguity status
source drift	a database or ontology changes without a retained version	source release and retrieval provenance
citation decay	source is retracted, moved, or no longer supports the claim	citation context, freshness review, explicit invalidation
relationship inversion	supports/refutes, drug/target, kinase/substrate, or parent/child direction reverses	typed directed edges and asymmetric tests
evidence laundering	derived interpretation is stored as primary evidence	evidence class and derivation lineage
contradiction erasure	reconciliation drops a dissenting record	retain competing evidence and resolution rationale
confidence conflation	source confidence becomes claim or decision confidence	separate confidence dimensions and owners
reconciliation instability	input order changes the resolved record	deterministic resolution and retained alternatives
graph corruption	dangling, duplicate, or unreachable evidence enters memory	graph-integrity validation
coverage overclaim	“not represented” is read as “biologically absent”	denominator, source scope, and unresolved set
schema drift	old persisted evidence changes meaning on load	compatibility evaluation and migration evidence
review-state divergence	a brief no longer matches canonical stored evidence	content fingerprint or regeneration check
redistribution error	external content is shipped beyond its license	source-specific licensing and distribution boundary
downstream reinterpretation	consumers assign new meaning to canonical fields	typed handoff contracts and cross-package tests

Review Priority

Prioritize failures that change meaning without breaking serialization: namespace defaults, relationship direction, source versions, uncertainty coercion, and reconciliation policy. Test contradictory evidence, unresolved identifiers, stale sources, duplicate relationships, empty coverage, and old serialized records.

A trustworthy handoff preserves not only the accepted record but also why it was accepted, what competed with it, and which uncertainty remains.