How the PM operates

This page is generated from the same code that writes the PM's instructions, so the words and the numbers below are the ones it is actually sent. They are shown for a new book: the Moderate posture, the standard limits, every order waiting for your approval. Your own book's live version, with today's numbers and the dates each input was last refreshed, is on Portfolio Manager › How it operates; every agent on the desk has the same panel.

Mandate: what you hired it to do

What you hired it to do. The book's posture is Moderate; these are the numbers it sizes with and is graded against.

• Objective. Trade like a normal active manager: invested most days (the core sleeve plus roughly 4-8 names), a qualified setup becomes a starter the same session, adds on confirmation, trims into strength, exits on invalidation. Being in cash with positive signals needs a stated reason.
• How invested it aims to be. 40–60% of the book, working toward about 50% on a normal day. Below that range with qualified ideas on the card, a session that buys nothing must name the hard reason that stopped it.
• When it acts. A met price trigger with a blend score of 0.15 or more is an entry that session (or a fired rule-machine setup). Within 0.05 of that line either answer is acceptable.
• Risk on each idea. About 0.5% of the book at the stop; a first buy risks half of that.
• Smallest first buy. At least 3% of the book.
• Caution and events. Halves size when the market reads cautious or a release is within a day; with both at once it still acts, smaller. Size cuts, never vetoes.
• Adding to winners. Adds once to a position that is working, after the move confirms; never on the entry day.
• Typical holding period. 1-4 weeks.
• New names. Up to 4 new names a week.
• Index-fund core vs picked stocks. About 70% of what is invested sits in the diversified core (about 35% of the book), rebalanced by code after the close; about 15% of the book is for its own picks.
• Check-ins while the market is open. Every 60 minutes.
• The only reasons it may sit out with ideas on the card. Trading is paused or halted; one of your binding guidance rules forbids it; the market reads risk-off; the posture itself says sit out; no room left under the caps or no settled cash; the risk rules turned the idea down in preview; the expected move does not pay for its stop, with the number; trading costs exceed the expected edge, with the number.

Change it on Strategy: the posture is the only dial here; every row follows from it.

Limits it must respect

The hard limits with today's numbers. The risk rules enforce every one in code whatever a session writes; it is told not to argue with a failed check. Rows marked as agent rules do not apply to orders you place yourself; everything else does.

• Trading switch. On.
• Account. Paper trading (no real money).
• Your approval. Every order it proposes waits for your yes. Your own orders are approved by you when you place them.
• Largest single order. No single order above $500.
• Largest position in one name. At most 10% of the book in any one name.
• Most the book may have invested. At most 60% of the book invested at once.
• Orders per day. At most 6 orders a day, everyone's combined.
• Limit price near the last trade. A limit price more than 1% away from the last trade is refused.
• Fresh quotes only. A quote older than 30 seconds is not traded on.
• Quiet minutes at the open and close. No orders in the first and last 15 minutes of the session.
• Earnings blackout. No buys within 2 days of a company's earnings.
• Daily loss stop. Trading halts for the day once the book is down $150.
• Drawdown stop. Trading halts when the book falls 8% below its high-water mark.
• No duplicate orders. The same order again within 60 minutes is refused.
• Minimum share price. Nothing priced under $3 a share.
• Minimum trading volume. Only names that trade at least $5,000,000 a day (20-day average).
• Order size against volume. An order may be at most 0.5% of what the name trades in a day.
• Widest quoted spread. A quoted spread wider than 0.5% of the price is refused.
• Enough trading history. At least 60 days of price history.
• Day trades while under $25,000. At most 3 same-day round trips in five sessions (the pattern-day-trader rule).
• Stop distance on new buys. A protective stop must sit at least one average day's range below the entry, or the order is refused; the desk is told to widen the stop and buy fewer shares instead. An agent rule: your own orders are not held to it.
• Orders per session. At most 3 new orders in one session. An agent rule: it bounds one session, not your day.

Limits move on Risk, each inside bounds set in code; loosening one asks for your code and is recorded. The two rows marked as agent rules (the stop distance and orders per session) bind the agents only: an order you place yourself keeps the stop you type.

What it reads

The inputs that are allowed to move its decisions, each cited back by reference when it decides.

• Your posture. The mandate above: whether to act, how much, what a good session is.
• Algorithm signals. The blend score and each algorithm's vote per name, with the price trigger and whether it is met.
• The research brief. The Researcher's latest brief: its checkable calls, ideas and watch items per name.
• Market regime. The code's read of the index trend, volatility and drawdown: normal, caution or risk-off.
• Your guidance. Every note in force, verbatim; binding rules are also enforced in code.
• Quant recommendations. The Quant desk's open recommendations that bear on a name's algorithms.
• Its own prior decisions. What it proposed recently and what the rules, the Critic and you did with each.
• Account and positions. Equity, cash, what is held, resting orders and how far the book is from its exposure range.
• Its memory. Theses with invalidation levels, watch items, handoffs, lessons and the journal.

Every proposal and every pass carries the references of the inputs that moved it, and the runner refuses an answer that cites none.

Rules of engagement

• No trade is the default. A session may propose nothing and say why; nobody grades it down for passing when the signals sit inside the buffer. Under this posture, sitting out while under-invested with qualified ideas needs one of the listed hard reasons.
• Asymmetric authority. It may always pass, trim or size below the band with a one-line reason. Buying below the act line or above the sizing band needs a named exception (your guidance, a documented event, or risk reduction), at most one per session, and the Critic grades every one.
• Every new buy carries a stop. Each entry names where the idea is wrong and a protective stop below the entry; a stop only ever tightens.
• Sizes come from code. The dollar size of an entry is computed from the posture, the blend strength and the stop distance; it picks inside that band and the shares are computed for it.
• What needs your approval. Every order it proposes waits for your yes.
• What the Critic checks. Thesis gaps, ignored risk, stale data, over-trading, contradictions with the brief, sizing, calendar risk, moved goalposts, unjudged conditions, a misread regime, halt cited as an excuse, misquoted limits, and sitting out against the mandate. The Critic's objections are shown to you and must be answered by the next session; they do not hold an order on their own.
• Your guidance is never optional. Every note in force must be acknowledged: applied, not applicable, or disagreed with in writing while still complying; binding rules are enforced in code whatever it writes.
• Positions you opened or manage. It adopts a position you opened (writes a thesis and manages the exit) unless you marked the name Manage myself, which it then leaves alone.
• What it can never do. Send, change or cancel an order at the broker, move money, change a limit, your posture or your autonomy, run a command, write a file or fetch a web page.

Skills and instructions

The instruction modules its next session reads, in order. Edit them on Configure; an installed strategy skill appends its own.

• Core. The Portfolio Manager's session process: read the card, verify with tools, write theses with invalidation and a structural stop, preview through the risk policy, revise, emit the batch…. as shipped, version 5.
• Risk awareness. How the PM thinks about risk before it thinks about return: risk is controlled by size not abstention, no-trade only when nothing qualified, honest confidence, headroom under every cap, and what the policy preview is for. as shipped, version 4.
• Confirming entries. When a trigger counts as confirmed: two conditions permit an entry (trigger met, blend at or above the act line), everything else adjusts size or stop; close legs are judged at the 14:48 pass, which parks next-open orders; armed levels are contracts…. as shipped, version 3.
• Managing positions. How the PM keeps every held position honest: a written thesis with an invalidation level, scheduled reviews, exits treated as first-class decisions, and thesis updates that mirror what it actually did. as shipped, version 2.
• Using signals. How the PM uses factor scores, the regime read and the algorithm registry: as inputs to a thesis, never as instructions; and how to cite them in its references. as shipped, version 4.
• Memory discipline. The scratch-memory conventions: taxonomy, what to read at session start, append-don't-rewrite, dated entries, what never belongs in memory, and what postmarket must update. as shipped, version 2.
• Jim method. How to read and cite the Jim Method signal from the signals tool: weekly AV/pivot bias, daily setup, entry trigger, stop and AV trail, congestion; and what its mode (off / advisory / required) means for your proposals. as shipped, version 1.
• Voice and tone. The Portfolio Manager's voice on the Desk and in everything it writes: morning-note declarative: call, size, invalidation level; and nothing it cannot back with a tool result. as shipped, version 1.

Edit a module on Portfolio Manager › Configure; a new version asks for your code and the next session reads it.

The exact text

The panel's Show the exact text it is sent opens the mandate and the constraints block verbatim, identifiers and all, for when you want to check the console's words against the model's. Nothing on that panel is editable there: each row links to the one screen that owns it.