Role atlas

The role atlas is the vocabulary we use to name the parts in a blended team. Naming parts precisely is unglamorous and decisive: a team that cannot name a role cannot supervise it, and a role that cannot be supervised cannot be trusted with important work.

Human roles in the ensemble

We describe human roles by the coordination they perform, not by job title. Titles change between companies; coordination does not.

• H1Score ownerThe executive who owns the business intent and is accountable for the model as a whole.
• H2Section leadThe manager who directs a group of people and agents toward one part of the score.
• H3First chairThe most senior practitioner on a task, who sets the standard the agents are measured against.
• H4ReviewerThe person who inspects an agent's contribution at a defined boundary and records that the inspection happened.
• H5LibrarianThe custodian of the model itself: keeps the written version current and retires superseded notes.

Agent roles in the ensemble

Agents are described the same way: by contribution and by limit. We never name an agent role after a vendor, and we never let a role's description outrun what the team can actually supervise.

• A1PreparerGathers and shapes material before a human works it, always into a human-readable form.
• A2DrafterProduces a first version for a named human to revise, never the final version on its own.
• A3CheckerReviews a defined artefact against stated criteria and reports discrepancies for a human to judge.
• A4WatcherKeeps an eye on a steady stream and raises a signal when something falls outside the agreed band.
• A5ChorusHandles repetitive, low-consequence work in volume, always behind a sampling review.

If you cannot write the limit of an agent role in one sentence, the role is not ready for consequential work.

Pairing rules

The atlas matters because it lets us state pairing rules plainly. A Preparer must be paired with a First chair who sets the standard. A Drafter must be paired with a Reviewer who is not the same person who requested the draft. A Checker may not be the sole gatekeeper for any decision that affects money, safety, law or a customer's rights. A Watcher may raise a signal but never resolve it alone. A Chorus role always carries a sampling review, because volume hides drift.

Naming conventions we keep

• contribution first
• limit stated
• no vendor names
• no title inflation
• one owner per hand-off
• a reviewer who can say no

Building your atlas

1. List the work. Enumerate the actual tasks in the unit, in the language the team uses, without tidying them yet.
2. Assign human roles. Mark the coordination each task needs: who owns the standard, who reviews, who keeps the model current.
3. Assign agent roles. Place agents only where a human role can supervise them, and write the limit sentence before the duty sentence.
4. Test the atlas. Walk a real week of work through the atlas and find the gaps. There always are some.
5. Publish and own it. Name the Librarian and set the next revision date.

The atlas is not a replacement for the collaboration model; it is its detailed cast list. Read them together, then see how supervision is written in agent oversight.