Requirements
Two lists. The requirements say what the system must do. The reality checks say how close it gets to real ERP software.
The reality map
Five levels, from the shell agents should handle quickly to the end to end consistency that makes an ERP trustworthy. Published before the first run. Cells change color as tests are run.
Community driven. F11, B8 and the whole fifth level came from comments on the first post. The catalog stays open.
Requirements catalog
Filter by group or priority. Specs and tests reference these IDs, so every change can be traced back to why it exists.