Guidance: standing and for now

The PM starts every session with a blank memory and one page of context, the state card, built from three layers: hard rules the Executor enforces (layer 1: posture, limits, the tradable universe), your guidance in words (this article), and its own memory of positions, theses and lessons. This article is about the words.

One box, one list, two lifetimes

Strategy › Guidance has one box to write in and one list of everything in force. Whatever you write is read back as a card, and the card carries the choices; Telling your desk what you want walks through it. Every note you confirm has a lifetime:

• Standing. Your house style in your own words: "prefer large, boring companies", "no averaging down", "half-Kelly at most". It applies to every session until you end it, and it renders first on the state card under Standing guidance. The PM reads it; it does not acknowledge it note by note (if it cannot follow one today it says so, with a reason). This is what the Playbook used to be; your old Playbook text is now your first standing note.
• Until a date or until an event. A nudge for now: "no new NVDA buys until the print", "keep 40% cash this week". A dated note expires (1 day, 3 days, 1 week or 2 weeks, or the day the desk read from your words); a note until an event ("until the Globex IPO") waits for that event instead: the platform watches the news for it, wakes the PM when a matching story arrives, and asks you monthly whether it still matters (it lapses after six months if nothing happens). Both render under Guidance for now (acknowledge every one). Every session while it is active must say how it understood it: applied, not applicable, or complied but disagrees (with a reason). You see that trail on the note's row. This is what Steering used to be.

Notes are about the whole book or one symbol, in plain language. A note kept in your exact words can run to 4,000 characters and is quoted verbatim.

Soft or enforced: decided on the card, never by accident

You do not pick "soft" or "hard" up front. When the desk reads your words as a rule the Executor can enforce, that piece of the card carries a Rule choice:

1. 1Enforce saves the rule as a hard constraint: the Executor rejects any buy that breaks it from its next loop (sells always pass), whatever the PM thinks. Confirming asks for your authenticator code. The rule keeps the lifetime you chose.
2. 2Keep as guidance keeps the words soft: the PM weighs them and must answer for them, the Executor does not enforce them.

Nothing becomes a rule without your click, and the audit log records your words and what you chose. With Use my words exactly the same question is asked when you press Save guidance: "This reads as 2 enforceable rules + guidance for now", with Enforce and Keep as guidance per rule, and the rest of what you wrote (the thesis, the reasons) is saved as guidance either way. That reading is done by fixed text patterns ("no new buys in X", "don't buy X or Y", "keep gross under N%", "keep N% cash") and, on hosts where it is available, by a small fixed classifier that may only answer in that same closed set of rules; anything else it says is discarded. Build the rule yourself, on the same card, spells a rule out by hand.

To tell the desk something:

1. 1Open Strategy › Guidance and write in the box, in plain language.
2. 2Press Read it back to me. On the card, check what it covers and how long it lasts (Standing, Until a date, Until an event), and answer Enforce or Keep as guidance if it asks.
3. 3Press Confirm. The row shows at once in the list: standing notes as In force, notes for now as Awaits the next session until a session acknowledges them.

Guidance starts in two more places: Guidance on a symbol page or an approval card opens the box with the sentence started for that symbol, and Consult can put your words on the same card in the thread; it uses exactly the same route, bounds and code check as this screen.

Two look-alikes are kept apart on purpose: the lessons the PM writes for itself are weighed separately, and feedback from people you share the book with lands under Inbox › Feedback for you, never in front of the PM.

Standing guidance

Write how you like to trade, once, and refine it as you learn: what to prefer, what to avoid, how big a starter is, how long you hold, what makes you sell. A standing note has no version history of its own: the audit log keeps every change. Reword on its row puts the words back in the box; confirm the new one, then press End now on the old one. If you keep re-writing the same note for now, make it standing.

Guidance for now

For this week, this event or this symbol: it reacts to something happening now, it expires on its own, and it is acknowledged so you can see it landed. An Until an event note ("until the Globex IPO", "until NVDA earnings", "until the Fed cuts") names the event in words a headline would use. Saving it creates an if-then watch on that event (you can see it under Research › Standing watches and on the note itself: what it is waiting for, how long it has waited, what kind of story would trigger it). When a matching story is ingested the PM is woken, reads the story, and either confirms the event (the note has done its job and ends) or answers "not yet" (the note keeps waiting). Every 30 days the note asks Still relevant?: Keep watching or End it. It lapses after six months if the event never comes, so nothing guides the desk forever by accident. A rule you Enforce from such a note still lapses after two weeks: what the Executor enforces always ends on a date.

When a note kept in your exact words reads as a rule the Executor could enforce but the text patterns did not catch it, a small fixed classifier reads it on the desk a few seconds after you save; the note's row then shows Reads as an enforceable rule with Enforce (asks for your code) and Keep as guidance. Nothing is enforced without your click.

The Watchlist on Strategy › Watchlist is a seed, not a gate. Names you switch on lead each session's reading and appear first on the state card.

Algorithm weights and modes (Signals), cadence and models (Desk › Portfolio Manager › Settings) and each agent's instruction modules, on that agent's Settings tab on Desk (instruction modules· Desk › <agent> › Settings › Instruction modules), shape sessions too; each has its own home and Strategy only links to it.

What the strip and the preview show

The strip beside Strategy's title says the book's three settings in a few words on every tab (the posture in force, when the desk must ask you, how the money is split) and, when the desk has suggestions waiting, their count with a link to the Inbox, where they are answered. Each chip opens its panel on Profile; none of them edits anything. On the Guidance tab the Next session panel beside the list says what the coming session will read in one sentence ("It will look at 18 names and everything you have told it, on Sonnet with a $2.00 budget") with the item-by-item list behind a fold. Everything else that guides the PM (Risk limits, Signals, the agents' settings, Research pins) has its own screen and is linked from where it matters, not repeated here.

Which lifetime

Either can be enforced: a standing "never buy TSLA" and a two-week "no new NVDA buys" both become Executor rules when you choose Enforce and confirm with your code.