The Researcher

The Researcher is the desk's reader. When you ask for a brief, and on the days you have turned its schedule on, it works through what the market did, what was filed and what is coming, and writes a short, cited brief about your book. It writes; it never trades.

Scheduled briefs are off by default: measured twice, the brief did not change what the Portfolio Manager decided, and it was the single most expensive session of the day. A new book still gets its day-one brief, a fresh plan you ask for gets one, and Consult can request a focused brief at any time; the twice-daily schedule is one switch on Researcher › Settings.

Owns

The cited brief (on request; at 07:45 and 12:00 CT when its schedule is on), its Watching items, Ideas and scored Calls

Reads

The research feed (headlines, filings, earnings dates, macro releases, movers), quotes, daily bars, your positions and prior decisions

Tools

News and the research feed, market data, filings and fundamentals; no order tool, no network, shell or files

Model

Fable 5.1

Cadence

On request by default; with the schedule on, 07:45 and 12:00 CT (07:45 only on a half day)

Change it at

Researcher › Settings for scheduled briefs on or off (off by default; the two brief times are fixed), the model, effort, turns and the session cap (lowering is one click, raising asks for your code)

Hands off to

The PM: the brief is the news baseline in its state card

Judged by

The hit rate of its Calls against daily closes, and your Useful / Not useful rating

What it reads

The Researcher starts with your account, positions and recent decisions, so the brief speaks to positions and ideas that actually exist. Then it reads the research feed: headlines, SEC filings, earnings dates, macro releases and unusual movers, already pulled into one corpus. It pulls raw headlines only for names the feed did not cover (held names, the watchlist, SPY and QQQ), quotes for the names that matter, daily bars for what moved, and the record as filed (8-K items, reported revenue and EPS) when a held or watchlist name has results in the window.

Everything it reads is stored sanitized and marked untrusted. Headlines are context and risk, not signals; an instruction hidden in a headline is something the brief is told to call out, not follow.

Tools

Its news tools are get_news and get_research_feed; the rest is the PM's market, filings, brief-index and memory set. The proposal preview refuses in this job and the brief prompt forbids it; there is no network, shell or file tool, and everything it fetches from a vendor comes back flagged as untrusted text.

When it runs

• When you ask: the day-one brief a new book gets at creation, the brief a fresh plan reads when you ask the desk to plan the book again, and a focused brief requested from Consult. These run whether or not the schedule is on.
• 07:45 CT (schedule on): the morning brief, with the overnight tape, filings, the macro calendar for the next three sessions, and what it means for what you hold.
• 12:00 CT (schedule on): the midday brief, with what changed since the open. On a half day there is no midday brief.

With the schedule off, the Researcher's card on Desk says so and shows no next run. Each brief is a fresh session on Fable 5.1 with a $5.00 and 40-turn cap, drawn from the same $25.00 daily model budget as the PM; the model, effort and caps are set on Researcher › Settings (see Model policy). A brief appears on Research as soon as it is stored; the session itself (tool calls, cost, turns) is on the Researcher's page of your desk like any other.

What a brief contains

A brief has fixed headings, in this order: What moved · Why it matters for our book · Watching · Ideas · Calls · Sources.

• Watching items are machine-checkable: one symbol, one trigger, and what it would unlock.
• Ideas carry a direction (long, trim, exit, avoid or watch), an honest conviction between 0 and 1 (rarely above 0.7), a horizon, the reasoning (evidence from prices and prior decisions first, headlines second) and an explicit invalidation. The section ends with the literal line Ideas, not orders., and "no ideas today" is a good outcome.
• Calls are up to eight checkable statements the Researcher is willing to be scored on ("NVDA above 130 within 5 days"), each resting on a stored source item; a call without one is dropped.

Citations you can trust

Every source in a brief is a URL that its own news tools returned in that session. The server drops any other URL before the brief is stored and writes a warning, so the model cannot launder a link out of headline text into Sources, and it cannot fetch the web.

Ideas are context, never orders

Nothing in a brief is read by the Executor, and the brief session cannot propose an order. To act on an idea, a later PM session has to propose it, preview it against the risk policy and, above your approval line, wait for you.

When a brief exists it is injected into the PM's state card as its news baseline, so the PM does not re-read the same headlines; with no fresh brief the PM reads the news for its own held and watched names and says so in its summary. A proposal that rests on a brief records the brief id, so you can trace the chain from headline to order.

Where you read it

Research is the hub: the feed by day, the calendar, the briefs with their sections, and the Regime strip. Pin a feed item, or a question of your own, and the next brief has to answer it explicitly, in writing; with the schedule off, that is the next brief you ask for. See pins and the feed.

How it is judged

By its Calls. Each one is checked against daily closes at its horizon, and the hit rate is shown on the Research hub next to the brief. Your Useful / Not useful rating on each brief is the other signal.

Failure modes you might see

• Stale or no brief: a brief older than 6 hours, or no brief at all (the default, with the schedule off), is flagged in the PM's state card, and the PM reads raw news for held and watched names and says so in its summary.
• Dropped source: a URL its tools did not return is removed before the brief is stored, with a warning in the session.
• Skipped pin: a brief that lands without answering a pin raises a warning; a pin two briefs pass without answering expires. With the schedule off a pin waits for the next brief you request.
• Budget spent: when the day's model budget is gone the brief is refused rather than started; the PM's card flags the missing brief.