Clawser / MVP demo / recurring research brief ops

Agents do the work. Clawser runs the business.

This is the smallest real Clawser loop, 1 client, 1 runbook, 1 scheduled brief job, 1 exception flow, 1 receipt, 1 margin view. The goal is not a prettier report generator. The goal is to show the durable operating layer for recurring client work.

client portfolio runbook registry daily ops board exception queue receipt timeline margin visibility

Current phase

Open exceptions

Autopilot stays on until real judgment is needed.

Receipt status

Receipts are the durable record, not transcripts.

Gross contribution

Revenue minus agent cost and human intervention cost.

Live operating loop

Run the first workflow

This simulates the exact v1 path: schedule, source collection, draft creation, quality failure, exception handling, delivery, and receipt creation.

Client portfolio board

One live client service line

Why this wedge

Recurring research briefs force Clawser to own runbooks, exceptions, quality, receipts, and margin in one client-facing loop.

Runbook registry

Client-specific policy

Quality board

Completion is not quality

Exception queue

Only real judgment should page the operator

Receipt timeline

What was delivered and why it can be trusted

Margin board

Economics from day one

Work ledger

Append-only operational history

Object viewer

Durable records, not prompt soup

Build target

What this prototype now implements

One client account with one recurring research brief service line
One runbook with approved sources, exclusions, rubric, and delivery target
One scheduled brief job with a visible state machine
One quality-triggered exception requiring human judgment
One delivery receipt with source references and intervention history
One margin view showing agent cost, human cost, and gross contribution

Build-ready slice and acceptance criteria live in docs/specs/research-brief-mvp-v1.md.