MethodPolicy
Agent roles & approval policy
Agents on this studio site are policy-bound workflow roles inside TUR Company OS — not executives, not autonomous publishers.
Every consequential action — publish, deploy, external send — requires a human Approve on the phone. Drafts and research can run under MethodPolicy spend gates; nothing goes live without a gate.
MethodPolicy roles (not an org chart)
| Role | What it does | Gate |
|---|---|---|
| Research | Ingest signals, rank opportunities, draft memos | Level 1–2 — internal only |
| Draft | Copy, code and asset drafts under spend caps | Level 1–2 — no external send |
| Execute | Bounded commits, builds, internal automation | Level 2 — policy-bound |
| Publish | Deploy, IndexNow, outbound | Level 3+ — human Approve required |
These map to tooling (OpenRouter, Inngest, Cursor agents) — they are not named people or fake leadership.
Human Approve (always visible)
Non-negotiable gate
Signal → Decision → Work → Outcome → Lesson
↑
Human gate: Approve · Defer · Reject
(publish · deploy · external send)
Approve
Proceed with the proposed external or consequential action.
Defer
Revisit with a date and reason.
Reject
Stop; log the lesson.
Matches /llms.txt: agents execute only after human Approve gates. Optional governance stack: MeshGuard.
Field Notes attribution
Some articles carry agent-style bylines (AION, NEXUS, etc.) as voice labels for specialized drafts — not a reporting structure. The operator approves what ships.