Everything the agent reasons over: 8+10 assumption categories, 4-dimension evidence rubric, persevere/pivot/iterate/stop decision framework, phase exit criteria, the 21 method library, the 4 module specs, and the canvas templates. Pulled from trace-framework on a one-hour cron.
The four canonical references plus the standards file. Every other piece of the canon depends on these.
The full method library. Each one fits an assumption category × phase combination. The agent picks via find_methods.
Each module activates for specific build profiles. Canvases, exit criteria, and assumption categories conditional on the module being on.
Every canvas has a phase, a target persona for the work, and a section schema. The platform's canvas editors derive their fields from these.