Operating evidence workflows

Knowledge operations preserve the history needed to answer not only “what is believed?” but “which sources, contexts, conflicts, and policies produced that posture?” A normal workflow appends evidence, validates graph integrity, grounds identities, reconciles relationships, and emits a versioned review.

flowchart LR
    S["register source"] --> I["ingest and normalize"]
    I --> V["validate memory graph"]
    V --> G["ground identity and context"]
    G --> R["reconcile support and conflict"]
    R --> A["assess coverage and sufficiency"]
    A --> B["publish review bundle"]

Common operating routes

Work	Required review
ingest literature or a database extract	source identity, retrieval context, license, normalization, duplicates
ingest a Core or Runtime artifact	artifact digest, producer, schema, run provenance, computed versus inferred content
ingest a Lab outcome	assay and batch identity, observation, QC, deviation, reconciliation context
refresh an annotation pack	source version, changed mappings, unresolved and newly ambiguous identifiers
change reconciliation policy	before-and-after conflicts, selected actions, holds, downstream review differences
assemble a decision brief	fixed memory revision, intended use, support, contradiction, deficits, freshness

Common workflows gives the detailed sequences and installation and setup covers optional reference and file dependencies.

Incremental evidence updates

Do not rebuild history by overwriting records. Register the new source or observation, link superseding material, rerun affected grounding and reconciliation, and publish a new review revision. Preserve the previous bundle for decisions that cite it.

sequenceDiagram
    participant O as Old review
    participant M as Evidence memory
    participant N as New source
    participant R as New review
    N->>M: append record and provenance
    M->>M: validate, ground, reconcile
    M->>R: assemble new revision
    O-->>R: historical lineage, not replacement

Diagnose evidence state

Symptom	Inspect first	Likely class
expected claim disappeared	source ingestion, normalization, graph edges	missing or invalid lineage
one identifier maps unexpectedly	source version, alias and organism context	ambiguous or changed grounding
support count increased after refresh	duplicate and derivation relationships	non-independent evidence
contradiction vanished	reconciliation action and context split	destructive conflict handling
review looks more certain than memory	sufficiency policy and unresolved deficits	review assembly error
consumer sees different state	memory revision and schema identity	stale or incompatible artifact

Observability and diagnostics maps these symptoms to reports. Failure recovery covers repair while preserving lineage.

Security and source custody

Treat external files and reference payloads as untrusted input. Record source and license metadata, validate schema and size before ingestion, and avoid embedding credentials or restricted material in persisted public artifacts. See security and safety and deployment boundaries.

Scale and determinism

Batch ingestion, graph partitioning, caching, and parallel resolution must preserve record identity, independent lineage, conflict classification, stable ordering where promised, and the same review result as the supported serial path. Performance and scaling defines the equivalence burden.

Release boundary

Changes to source policy, status enums, grounding, reconciliation, sufficiency, or review schemas can alter downstream decisions even when imports remain stable. Release and versioning requires explicit compatibility and before-and-after review evidence for those changes.