A research supervisor on call.
Drafts the concept note, you keep the judgement.
When a researcher is matched to a dossier, M11 opens a workspace with the dossier on the left and a Claude assistant on the right. The assistant is dossier-aware and discipline-aware — its system prompt encodes the Nick Golding methods register: favour mechanism plus statistics, flag the recurring traps, prefer reproducible workflows.
Dossier-aware
Loaded with the dossier context, linked Layer-0 documents and prior session notes. Asks better follow-up questions because it knows what you are working on.
Discipline-aware (the Nick lens)
Favours mechanism plus statistical residual. Flags traps — NegBin replacing the GP, case-only likelihood claiming identifiability. Channels the methods register of a real disease-modelling supervisor.
Concept-note sections
Drafts: background, methods landscape, data sources, expected outputs, ethics statement, timeline, budget justification. You edit; it learns; you ship.
Critique mode
Switch from drafting to critique — paste a section, get a structured review against the working-modeller adviser network's curated standards.
Citations on every turn
Every claim links back to a Layer-0 document, a peer-reviewed paper, or a labelled "this is the model's best guess — verify". No bare assertions.
Editor + adviser checkpoint
Output is always an AI-generated draft. Before any external submission the assigned working-modeller adviser signs off in a 30-minute review.
The assistant's system prompt is loaded with the dossier MDX, the linked Layer-0 documents, the methods landscape and the working-modeller adviser's notes. Every assistant turn cites its sources. The researcher's edits are versioned in Git. The whole conversation is exportable as a draft concept note in two hours of work — instead of two weeks. Editor + adviser review is mandatory before any external submission.
The live module in v0.0 is M10 · Document Triage.
A working demo of the same back-end architecture. Try it.