An independent research publication
Chronomata studies how an agent’s recorded experience can shape its later actions—and, in a separate workstream, how compact construction rules might produce useful computational systems.
Working apparatus. Bounded experiments. Open questions.
No established learning advantage or claim about subjective experience.
01 / The questions
The persistent-agent apparatus separates model calls, external memory, a small simulated world, and an independent evaluator. The questions concern retrieval, continuity across interruptions, source fidelity, and measurable task performance under finite resources.
Developmental computation asks a different question: whether construction or growth rules offer useful transfer beyond a conventional modular builder. That assessment has draft protocols and a small engineering witness; its learning experiments remain unexecuted.
Explore the two workstreams →02 / The record
Reports distinguish intended tests from executed work, preserve failures and invalid attempts, and separate measurements from interpretations. Historical local registrations are never presented as public preregistrations.
13 October 2026 / Completed calibration
A bounded 4B model calibration met the delivery deadline but exhausted its generation allowance before producing a valid decision. It establishes neither a memory benefit nor broader functional reliability.
Read the experiment record →