Valcore turns agent judgments into release signals

Field notes

The homepage lays out a three-step evaluation loop: define a standard, assemble labeled evidence, then compare versions and gate releases. Its example names concrete case types such as refund requests, missing context, and unsafe requests, while a sample run reports 94.2% agreement across 128 cases against a 90% threshold. Installation is offered through Homebrew, uv, and pip, and the project identifies itself as Apache 2.0 open source.