How the desk is built
An agentic research desk. Official closes first. Missing fields stay blank.
Specialists with tools, not a prompt that writes the book
After each session, a weekday orchestration starts a set of research agents. Each agent has one job and a constrained tool: pull the close, value the universe, draft the note, or publish the book. They retry on a failed pull. They archive the prior book. They do not invent a number to finish a sentence.
That constraint is the advance. Fluency is cheap. An agent that will leave a field blank when the print is missing is the research standard.
How the desk runs after the close
Four specialists write one book. The desk only reads what they publish.
The agents
Four specialists. One coverage book. The UI only reads what they publish.
Close agent
Pulls the official session print for every S&P 100 name after the close. A name without a close stays blank. Pre-market prints are unused.
Valuation agent
Computes implied NTM EPS versus the sector-median forward P/E. Street target is labeled Street. DCF runs only when FCF, beta, and shares are in the book.
Notes agent
Writes bull / base / bear from the sector-median scenarios and the close. Same ≥15% Overweight / ≤−5% Underweight / else Hold bands on every ticker.
Publish agent
Opens coverage when the book is one close. Refresh can update prints. It cannot invent a price, a house EPS, or a 10-K story, and it cannot overwrite a published note.
Orchestration
- 1. Wake after the close. The daily cycle starts once the session is official. It does not trade on a pre-market print.
- 2. Hand work to specialists. Close, valuation, notes, and publish run as separate agents with their own tools and retries. A failed close stays blank.
- 3. Gate the book. Coverage publishes when the universe is one close. Street consensus stays labeled Street. Desk recommendations use one sector-median rule.
- 4. Archive, then refresh. The prior book is stored. The next session can update prints. It cannot rewrite a published note.
Guardrails
No invented price, multiple, house EPS, or 10-K story. Missing stays blank.
Durable cycle
Weekday post-close run with retries. Prior sessions stay in history.
One book
S&P 100, notes, earnings, and company pages read the same published book.
What sits on the desk
- Official close, price history, and street consensus
- Recommendation versus the sector-median NTM multiple
- Annual statements, estimate revisions, and earnings surprise when in the book
- Close recap: breadth, leaders, laggards, desk versus street