Model Birth Observatory and Cognitive Birth
2026A research program for recording model initialization, tracing checkpoint-level development, and measuring eight distinct cognitive axes under frozen and auditable decision contracts.
Research
Questions we are working on, methods we use, and results we can defend.
Counfield keeps a narrow scope. What is written here is written to be defended, not to be published quickly. Every number carries the authority of the evidence behind it, negative results are kept as visible as positive ones, and anything that was not measured is labelled unmeasured rather than filled in.
STOP-v1 Forensic Audit of Model-0
2026A 26.65M-parameter model produced replicated signal on M, I, and A, but the frozen eight-axis gate correctly retained a STOP decision after difficulty-scale, format, multiplicity, and independence audits.
An Eight-Checkpoint Developmental Profile of Local-70M
2026Matched diagnostic measurements from initialization to 531.99M training tokens show early A emergence, later I growth, non-monotonic axis trajectories, and a divergence between validation loss and the measured cognitive profile.
Phenotype Shift During Continued Pretraining
2026Continuing the same 69.97M-parameter model to 1.00B cumulative tokens improved validation loss while measured I and A declined and easy R increased, providing an exploratory example of non-uniform developmental change.
These four entries do not share one authority. The technical report carries the official result and the forensic audit; the two developmental entries are explicitly exploratory / diagnostic.
Authority is separate from content type. A research note and a technical report can sit side by side while resting on completely different classes of evidence, so every entry above carries its evidence authority next to its type.
Frameworks
A framework is a way of classifying things, not a result. It carries a framework status rather than an evidence authority, because the evidence-authority vocabulary describes measurements and there is no measurement here to describe.
Artificial Intelligence Evolution Index (AEI)
2026A 31-stage taxonomy of developmental and evolutionary regimes, from a deterministic machine to an open-ended universal regime, with a dated editorial placement of current frontier systems inside it.