Registered research
The full record, not a highlight reel.
Every preregistered Claim is accounted for. Status comes from canon Decisions plus a typed, hash-bound public status source—not editorial inference.
Memory systems comparison v1
A vendor-comparison pre-registration retained as an inspectable withdrawal. It was never executed.
Memory systems control arm
A control-arm pre-registration created before observation and withdrawn with the unauthorized vendor route.
Artifact coherence transfer study v1
A prospective study invalidated before measurement after a protocol deviation and an independently identified detector defect.
Before registration
Cross-domain conjectures
These are directions for criticism, not Claims, Evidence, Decisions, or findings. Their public scores route attention; they do not confer truth.
Leave-one-variable-out tests may falsify diagnostic system models
Withholding one telemetry variable or dependency edge can test whether a proposed failure account predicts the missing relation.
Closure pressure may collapse the hypothesis portfolio
An autonomous research loop optimizing closure alone may repeatedly select near-term tractable conjectures and narrow its source and hypothesis distribution.
Research-loop capability must be measured by stage, not prose quality
A strong synthesis may conceal failures in retrieval, registration, execution, interpretation, or publication.
Expert interventions should enter as typed graph constraints
Principal corrections may be more reliable when represented as scoped, reviewable constraints instead of unstructured prompt residue.