Method and trust layer

Trust comes from showing what the system will not claim yet.

Knowledge Prism is governed by a simple rule: the machinery supports the scholar. It can search farther, remember more, endure tedious sampling, and preserve provenance, but the research question and final judgment stay human.

Loading method facts...

Exoskeleton

Amplify the scholar's gestures.

The system follows the researcher's gait: asking, gathering, judging, connecting, interpreting, writing, revising, and remembering.

Not a replacement

The question stays human.

The tool can propose, rank, sample, and record. It does not decide what the research means.

Ledger as memory

The chain is not the subject.

The ledger keeps provenance and boundary-learning visible. It is memory for research judgment, not a spectacle.

Evidence ladder

A claim moves slowly on purpose.

Known fileThe file exists or is described. This is a clue, not scholarship. Technical label: metadata_only.
Catalogued sourceThe corpus inventory is stronger, but still not text-level verification. Technical label: metadata_manifest.
Candidate claimAn AI-generated or inferred claim waits for sampling. Technical label: hypothesis_only.
Sample checkedA bounded text sample supports, weakens, or blocks the claim. Technical label: sampled_text.
Evidence-backed research itemThe pilot supports the claim and the work belongs to the project boundary, but it is still not final. Technical label: sampled_text_supported_core_candidate.
Concept verifiedA concept has been seen and verified in text. Current count: 0. Technical label: concept_verified.
Final graph claimA settled graph claim. Current count: 0. Technical label: ontology_core.

Claim accuracy vs project relevance

A true claim can still be the wrong claim for this map.

A sampled claim can be textually accurate and still outside this project. Histology is the teaching case: true enough as a histology claim, not useful for this IR knowledge map.

Functional IR interpretation

Status is not meaning.

Every live item needs a functional reading: what IR function it performs, which interaction it illuminates, what layer it speaks to, what it can contribute, and what caution remains visible.

Boundary kinematics

Boundaries move on surprise and rest on confirmation.

Move on surprise

A case such as Histology or Afghan reportage can reveal that the current taxonomy needs a better boundary.

Rest on confirmation

The boundary should not refit itself to every shiny exception. Inertia keeps the domain coherent.

Log grey zones

Judgment is allowed, but hidden judgment is not. Every grey-zone call remains auditable and reversible.

Ledger as memory

The record protects the work; it is not the work.

The ledger is the project's notebook. It remembers what was tried, what was accepted, what was held back, and why. It should never become the main performance.

Machinery boundary

The machine may assist, but it may not take the pen.

Search, sampling, sorting, and audit trails are support functions. Research meaning, final relevance, and writing remain human-led.

Current pipeline

The live project state, read from static exports.

Open Expert Dashboard