DEVİM RESEARCH PLATFORM
Research
Public research notes, reports, methodology work and theoretical outputs from DEVİM.
336M Gate-B: capacity effect not established; architecture-design gate opened
The matched 336M Gate-B evaluation is complete. Scale improved FORM and positive-control readouts but did not establish the preregistered capacity effect.
Where is DEVİM now?
Verified current state of the DEVİM Model and Computational Dialectics research programs as of September 15, 2026.
DEVİM research timeline
From the 43M reference and 110M diagnosis through FORM/CONTENT separation to the final 336M Gate-B decision.
Computational Dialectics: where is D1-B Gate B?
The implementation passed a 22/22 preflight and an independent review returned PROCEED, yet no optimizer step is authorized until preregistration is frozen.
CONTENT: why relation binding was not enough
How controlled relation-binding success led to a diagnosis of shortcut-dependent learning with weak transfer.
Grounded Turkish Data: not just more data, but more traceable data
DEVİM's evidence-linked Turkish supervision program: provenance, sequestering and answer-present/answer-absent structure.
What went wrong in DEVİM?
We track progress through failed gates, evaluator confounds, overfitting and rejected diagnoses—not only through successful experiments.
Open research, controlled release
DEVIM publication policy for sharing research evidence with deliberate release controls.
Evidence before claims
The publication and validation approach that determines which research results become visible on devim.org.
Why is Computational Dialectics a separate research track?
A scope note explaining why Computational Dialectics and Devim model training are not treated as one project phase.
Why Turkish-first?
A short framing of why Devim treats Turkish as a starting condition of model research rather than a localization layer added later.