This is one researcher and a public repository, and it needs your help.

Both papers are preprints seeking review. Tell us where a claim is overstated or an engineering choice is wrong, and point out any citation that does not hold up. Adversarial readings are welcome. The planned test and the rule that decides its result will be fixed before any live run, so a hard look at them now is especially useful.

The reference implementation runs on a consumer workstation: 32 GB or more of RAM, a GPU with about 12 GB of memory and a second with about 8 GB, all inference local. It also runs on one GPU or on the CPU alone. Run the offline suite from its seeds, help run the planned test of the workspace once live runs begin, and report what you find, null results included.

CAL v0.5 is legally novel and untested: a copyleft, source-available, ethical-source license with neurorights-derived covenants, copyfarleft commercial terms, and a guardianship pathway. We need counsel willing to review it and tell us where it breaks before someone else finds out in court.

Commit code to the reference implementation. The paper's future work runs along three lines. Module addition brings the held faculties into the competition one at a time: memory, a world model, a self-model, social cognition, a voice, and bounded planning. Learning through experience shapes components from the entity's own history, and embodiment connects the workspace to a body.