Research records / Questions with consequences
Let the comparison
change our minds.
Selected findings, failures and simpler explanations. A research record is useful when you can see what it does and does not establish.
The science, in the open
A finding is stronger
with its alternatives.
Three isolated studies. What changed, what did not, and what a simpler explanation can still account for.
What can be learned without the answer?
Retained history helped an observation-trained recurrent model. A structured estimator still did better.
Scoped positive resultExamine the comparison ↗When should a belief reopen?
A failed short-horizon study exposed a lockout: avoiding a source also prevented discovering its recovery.
Failed precursor · scoped follow-up passExamine the comparison ↗Can consequences change the next choice?
Learned outcomes supported adaptation. A complete policy cache matched the planner, limiting the claim.
Functional effect · stronger interpretation not establishedExamine the comparison ↗Public methods summaries from internal research snapshots reviewed 16 September 2026. Reported results, not independently reproduced for this publication. Full run archives are not yet public.
The earlier record
stays readable.
The journal based on September 14 records preserves its own publication dates and scope. New evidence does not rewrite the earlier story.
A project beyond the session.
What changes when unfinished work can survive a restart without becoming an instruction to act?
Memory that matters to what happens next.
The important question is not how much a system remembers. It is what remembering changes.
Sometimes the instrument is wrong.
An uninterpretable endpoint is not a failed mind. Why we preserve the mistake in the measurement.