Permanent public chronology
Research Releases document reportable changes as they occur. A release may report a gain, a failure, a correction, a replication, or a newly frozen artifact.
Public state transfer, not hype. Releases preserve evidence status, limitations, and backing documents so later work can correct them without rewriting their history.
Research Release Sep 15, 2026
Consequence -> retained distinction -> changed future selectivity. Two adversarial rounds have defined enough structure to build the smallest Formation Kernel that can prove the candidate mechanism wrong.
Research Release Sep 13, 2026
A frozen moving/carry mechanism rose from 44/64 transported-reuse seeds at x1 to 64/64 at x2 and x3, while normalized work per agent-tick declined.
A result is reportable when an outside person could use, test, critique, or continue what was learned. It does not have to be a win. Old releases remain addressable; corrections and supersession are linked rather than overwritten.