Research Harness
Every name you follow becomes a case file that agents research to a set standard, and the work is checked and argued over before it reaches you.
One case file, from first look to a live position
Research Harness takes each company through six stages, all in one case file. AI agents do the work at every stage, and the calls along the way stay yours.
-
1
Scout
AgentsScout screens new names against your book and scores them, and the strongest wait for you as cards.
YouOpen a case from a card, or set one up yourself by choosing the company, analyst, agenda, thesis and run.
-
2
Thesis
MS0AgentsA framing step drafts the thesis, the view against consensus and what would prove it wrong, then files the key questions.
YouName the pillars, add any questions of your own, then approve the brief.
-
3
Research
MS1 → MS3AgentsResearch sessions work the agenda round by round. The Chief of Staff plans, the analyst researches and the gate checks every pass.
YouSettle any contested answers, and set the research level and the case's spending cap.
-
4
Pushback
MS4AgentsOnce coverage reaches 85%, a different analyst tries to break the research, and anything it weakens is reworked once.
YouIf the research breaks, you decide whether to pass or continue.
-
5
Call
Your turnAgentsThe call lays out the action, conviction, horizon and target, along with a p10 to p90 range and the exits.
YouApprove it, change it, ask for more, pass or defer. Every capital action waits for your click.
-
6
Live
MS5AgentsThe company is watched for catalysts, new filings, the quarterly results, street revisions and drawdowns, and re-rated after every pass.
YouRead each update. The digest arrives in your inbox, opening with the hook.
After each earnings print, the questions it touches are marked stale, and a refresh session reopens them.
Inside a research session
Each research session takes up to eight questions from the agenda and works through them in passes. Every seat has a single job, and the gate that checks the work is a set of rules, not a model.
Every answer opens with a hook, a single unhedged claim of 90 characters or less. It then states its edge against the street and cites its sources, and any answer that falls short goes back to the analyst.
The gate keeps partial answers, discards a pass when 60% or more of it is not public, and ends a session after two discards. You can set each of these rules for a case on the Canvas.
Sources are ranked with SEC filings at the top. A web source never stands on its own, and an answer can only call something "not public" after the filings or transcripts have been searched.
You can add a challenger, who argues the contested answers and recommends a side, and a CRO, who can file comments that force a rework.
See it at work
The screens your desk works in, shown with a turn in progress.
Every case on one board
The Harness tab shows every case in work, so you can see which turn is running, what the gate is judging and what is waiting on you.
- The Chief of Staff in the lead, with one lane for each case
- A single research lane that runs one turn at a time across the desk
- What the desk has spent today against its daily limit
One case, a turn in progress
Agent run follows a session as it happens, showing the Chief of Staff's plan, the analyst's questions, the gate's verdict and the map of the agenda.
- A station each for the Chief of Staff, analyst, gate, challenger, you and the CRO
- Use the Canvas to seat a challenger or a CRO and to set the gate's rules
- Each question lights up while the analyst is working on it
The agenda, measured
The agenda spans sixteen modules, from the business model through to valuation and catalysts. Coverage is weighted by how material each question is, and milestones show how far the research has come.
- MS1: the business model and the numbers
- MS2: every high-materiality question answered, with none contested
- MS3: coverage of 85% or more, and the case moves on to pushback
Contested answers come to you
When an answer rests only on web sources, contradicts the thesis or disagrees with figures that are already settled, it is held on a single sheet until you make the call.
- Accept the answer, or keep the earlier view
- Set it aside and research again, or find a stronger source
- During a run the work carries on, and the item appears in your daily digest
Runs on its own, inside your limits
A research run strings sessions together until the research reaches MS3, staying within a budget and a deadline. It pauses itself when the work stops paying off, and you can pause, resume or stop it whenever you like.
- $40 / 7 daysThe default budget and deadline for a run on Basic. On Pro, a run can spend $120 over ten days.
- 4 a dayUp to four sessions a day, with five minutes between them.
- 1 turnA single research lane runs one turn at a time for the whole desk.
- Basic / ProBasic covers the core agenda, while Pro works the whole agenda tree and writes the research paper.
A run also pauses itself if the framing step fails twice, the case's spending cap is used up, two errors come in a row, or three sessions end without keeping anything.
What a case leaves behind
- Company noteEvery pass opens with its hook, followed by the findings, the leads to follow up and up to four sources. During a run you also get a daily digest.
- Research paperThree to five pages organized by pillar, opening with the bottom line. Each published version is frozen as a PDF, and every earnings print adds a one-page update.
- Rating and the callEach pass re-rates the company across eight dimensions. The call becomes a recommendation, and once you approve it, a paper position in your portfolio.