Scale
Knowledge in a frame semantics graph is a topography, not a flat field and not a clean tree either. Frames compose two ways: structurally (a file holds sections, a section holds concepts) and argumentatively (a claim with its evidence and sources makes an argument). See graph construction for the composition and fractal composition for why it recurs.
The substrate runs across many scales. A rough ordering, large to small:
| Scale | What it is |
|---|---|
| Substrate | the whole shared store, everything every population reads and writes |
| Co-located region | a neighbourhood of the substrate where related work concentrates |
| Team / project membrane | a boundary around several agents and users working together |
| Active cohort | a membrane around several agents and their shared, currently active context |
| Personal base | a membrane around one user and their agents |
| Project context | the material scoped to a single project |
| Agent harness | one agent’s memory, tools, and instructions |
| Task context | what is active for a single task |
| File | one document — e.g. a file of agent instructions |
| Section | a part within that file |
| Concept | an individual idea mentioned in it |
Only the substrate sits cleanly above the rest. Everything below it is interconnected across levels, not neatly nested: a single agent’s active context, working inside a team, may pull instructions scattered across many files, projects, and other agents’ bases. A membrane is not a tier in this list — it is a boundary drawn at a moment over whatever is co-active, wherever it lives.
Which scale you read at sets what a reading means, including any metric computed over it. Pick the scale deliberately, and say which one.