Knowable

A collection is not just a database. It becomes knowable when it can describe what it knows about itself and where its understanding is incomplete. Each entity in this collection declares what attributes constitute its knowledge: semantic embeddings, visual embeddings, color vectors, search indexes. The system tracks which are populated, which are outdated, and which are waiting for prerequisites that don't yet exist.

Knowledge arrives through streams of inquiry. A stream is not a pipeline. It is a discipline brought to bear on the collection. Art-critical analysis reads the work as a critic would, compressing title, medium, artist, and curatorial description into a semantic coordinate. Visual-formal analysis looks at what the eye sees, independent of art-historical context. Image perception asks what a vision model notices in the actual pixels, bypassing language entirely. These are different ways of knowing the same object.

What the collection does not know is as meaningful as what it does. An artwork without a visual description cannot have a visual-formal embedding. That gap is precisely stated, not hidden. The future streams below name specific inquiries the collection cannot yet conduct: conservation, provenance, critical reception, scientific analysis. The architecture holds space for knowledge it does not yet have. These named gaps are the collection's honest assessment of its own ignorance.

This is the enliteracy framework: not searchability conferred once, but attention that compounds. Each stream's artifacts can feed other streams. Visual perception enriches art-critical analysis. Conservation discoveries trigger re-evaluation. The knowledge graph is not a static map but a living record of inquiry, where depth at one point creates possibility at others.

16 streams of inquiry, each a different lens on the collection. Streams produce knowledge artifacts when they intersect with a knowable record. Some run as background jobs, some compute synchronously, and some exist only as named intentions, waiting for their first inquiry cycle.

These streams exist as named intentions. Their methodologies describe inquiries the collection cannot yet conduct. The architecture holds space for knowledge it does not yet have.

Per-attribute knowledge coverage across all four browsable models. Each row shows how many records have the attribute populated versus how many are applicable (prerequisites met). Outdated counts indicate records generated by a previous version.

Streams feed each other. An enrichment dependency means one stream's artifacts improve another's quality. A trigger dependency means new artifacts in the source automatically initiate a cycle in the dependent stream. The direction is intentional: perception feeds analysis, analysis feeds relationship extraction, and conservation discoveries can trigger complete re-evaluation.

The collection can speak about its own epistemic state. This is not a summary written by a person. It is generated from the metrics above, the collection describing what it has examined and where its understanding remains shallow.

Not every stream should run at every opportunity. The cadence evaluator tracks yield (what fraction of candidates produced new knowledge) and trend (whether yield is rising or falling) to recommend whether another cycle is worthwhile. Only the five active jobable streams are assessed. Sync streams compute on save. Future streams await their first inquiry.

Some knowledge lives above individual records. Topology clustering projects the entire collection into 2D and finds density regions, revealing structure invisible from any single artwork. These collective sources register with the Knowable system and report their own staleness.