Lab
Alignment Lab
An interactive Neurokinetic AI lab for exploring language residue, dense/sparse retrieval balance, abstention thresholds, concept IDs, side channels, and UAI-style render targets.
Static browser-only educational simulation
Inspect how retrieval evidence becomes a concept record.
The Alignment Lab is a secondary static browser-only educational simulation. It uses precomputed examples plus local custom-text normalization and heuristic matching in the browser. It does not send custom text to a hosted model endpoint, execute endpoints, mutate state, provide writable GET actions, certify safety, or claim medical/therapy authority.
Scenario library
The center is the concept registry; orbiting traces are mutable language surfaces.
Active stage
Normalize
Preserve the original expression while creating a comparison form with stable spacing, case, and Unicode normalization.
Risk controlled: encoding drift and confusable surfaces.
Candidate ranking
Resolution trace
Rendered envelope
Normalize
Preserve the original expression while creating a comparison form with stable spacing, case, and Unicode normalization.
Embed
Represent the expression in dense, sparse, or late-interaction neighborhoods where candidate concepts can be retrieved.
Neutralize
Separate language residue and side channels from the core invariant the handoff needs to preserve.
Resolve
Select a concept ID only when the evidence is strong enough; otherwise the system should abstain.
Render
Return the concept as the target surface required by the receiving person, system, or protocol.