Overview Shipped
A team lead, a roster of specialists, a lens per specialty, and a set of actors whose authority is deliberately unequal — the one that finds a problem is never the one that decides what it is worth.
Nothing here is "an agent" in the undifferentiated sense. Each actor has a defined field of view and a defined authority, and the two are what make the design work: an actor that can see everything ends up deciding everything, and an actor that can decide about its own output ends up grading its own homework.
What each actor can see, and what it may do
| Sees the diff | Sees the roster | Sees all findings | Can raise | Can score | Can delete | |
|---|---|---|---|---|---|---|
| Team lead | yes | yes | no | no | no | no |
| Worker (a specialist) | yes | no | no | yes | no | no |
| Verifier | yes | no | no | yes | no | no |
| Consultant | yes | no | yes | no | yes | no |
| Report writer | no | no | yes | no | no | no |
Nobody can delete a finding. That column is empty on purpose and is asserted by a test: the pipeline can down-rank a finding to the bottom of the report, but it has no path that removes one.
The vocabulary, in one place
A team is a bundle of data. A specialist is one member of it. A specialty is one narrow lens that specialist reviews through — a specialist typically holds several. A worker is a single dispatch of one specialist through one lens. A verifier is a second dispatch that checks whether the worker covered its declared ground. A consultant scores what everyone raised. Open any of them below.