Own Rules Guide

An own rule writes down a limit you hold the portfolio to — “no single security above 10 %”, “cash never below 5 %”, “bonds within 3 percentage points of their target” — so that it is checked every time the Risk page is read, instead of living in your head or in an agent’s prompt, where limits drift. The section Own rules at the top of Wealth → Risk lists the rules of the active view with what each one finds. This guide walks through writing a rule, reading its finding, changing it and ending it, with the labels the screen uses. The decision behind the model is recorded in ADR-0049; the reference description of the Risk page is in the product documentation, and the same rules as your agent reads and writes them are described in API and MCP.

All names and figures below are examples.

What a rule is

A rule reads one figure the app already shows, for one thing, and draws a line against it. It is one of three kinds:

• Cap — breached when the figure is strictly above the line. “No single security above 10 %” is a cap at 10.
• Floor — breached when the figure is strictly below the line. “Cash never below 5 %” is a floor at 5.
• Band — breached when the figure is outside the range from From to To. “Bonds within ±3 percentage points of their target” is a band from -3 to 3 on the drift.

“Strictly” matters: a figure exactly on the line is inside it, the same way the generic thresholds further down the Risk page read a line.

A rule belongs to the view it was written in. It is evaluated on that view’s steerable basis, and Risk lists it only while that view is active. A rule written under Everything is a portfolio-wide rule.

Which figures a rule can read

A rule computes nothing of its own: each figure is one the app already serves elsewhere, read in the rule’s view.

• The subject follows the measure. The Subject list offers only what the chosen measure can be read for, so the dialog cannot save a pair that has no figure.
• A security’s weight is its single-name weight, merged across depots — the figure the Largest single names table shows.
• A security’s drift needs one more choice, Plan of the classification: a position target lives in one classification’s plan, and the drift is only one figure once the plan is named.
• A bucket is capped through a view. “The speculative bucket stays under 5 % of everything” is a Weight cap whose subject is the view that selects that bucket, written while Everything is the active view.
• A drawdown reads negative (“-12.0 %”). “Never more than 20 % down” is therefore a Floor at -20. A positive line is refused, because a drawdown never rises above 0.

Writing a rule

1. Open Wealth → Risk. The page header names the active view (“Concentration and movement · view: Everything”). The new rule will belong to that view. To write a rule for another view, pick that view in the view switcher on Wealth → Holdings first; the choice holds on every Wealth tab.
2. Click New rule. The dialog’s first line repeats the view: “Evaluated in the view “Everything”, on its steerable basis.”
3. Give it a Name in your own words, for example Single names at most 10 %. The name is a label: it is never read for meaning, and two rules may share one.
4. Pick the Measure, then the Subject, and for a volatility or a drawdown the Window.
5. Pick the Kind. Type the Line in the unit its label shows (“Line (%)”, “Line (pp)”) — or From and To for a band — written the way the page writes numbers.
6. Pick the Severity: Warning or Hard.
7. In force from is today. A later date schedules the rule; an earlier date is refused, because a rule is never applied to a past it did not exist in.
8. Add a note if it helps, and click Save rule.

A worked example: a rule Bonds near target with Measure Drift, Subject the category Bonds of the Asset class tree, Kind Band, From -3 and To 3, Severity Warning. It stays met while the bond category sits within three percentage points of its target in the active plan, and it is undetermined while the view has no active plan.

Reading the findings

Every rule in force is evaluated each time Risk is read; a finding is never stored and never sent anywhere. The section’s heading counts them (“· 1 breached · 1 undetermined · 3 met”), and the table lists one row per rule: the Rule (its name, and under it its words — measure, window, subject, kind and severity), the Measured figure, the Line and the State. Breached rows come first, then undetermined, then met; within each, hard before warning.

• met — the figure was read and is on the right side of the line.
• breached — the figure was read and is strictly beyond the line. The badge carries the signed distance to the nearest line (“breached · +2.4 pp”). That is arithmetic, not an instruction.
• undetermined — the figure could not be read. The badge is dashed, and the reason stands under it: “12 of 20 observations” for a volatility without enough history, “no active plan” or “no target in the plan” for a drift with nothing to drift from, “nothing to weigh” for an empty view, “held, but not valued” for a position without a usable price or exchange rate, “the subject no longer exists”, “the figure is undefined” or “not measured”.

A rule set for a later date is listed under Scheduled rules with its start and its line until it begins. A change scheduled for a rule in force shows in the rule’s row (“From 2026-06-01: line 8.0 %”).

Why undetermined is never met

A rule whose figure is missing has checked nothing. If it counted as met, a rule set would read green precisely when half its inputs were gone — the same silent drift the rules exist to replace, with a check mark on it. So an undetermined finding is never counted as met, is never hidden, and always says why. A security you do not hold is a different case: its weight is 0, which is a reading, so a floor on it can breach. Only a subject that no longer exists is undetermined.

What severity does — and what a finding does not

Severity changes how a breach looks and where it sorts: a hard breach carries the danger colour, a warning the warning colour. Nothing else depends on it. No rule blocks a booking, an import or any other change, and no finding says what to do: it carries no quantity, no trade and no suggestion. Deciding what a breach means is yours.

The section Largest single names below it is titled “generic thresholds, not policy rules”. Its lines (a single stock above 10 %, 7 % being the first line, an ETF above 25 %) are the app’s generic heuristic. Your rules do not change them, and they do not read your rules: a position can carry a breached rule of yours and a generic “above 10 %” at the same time, as two separate statements.

Changing a rule: a new version from a date

A rule’s name in the table is a link. It opens the dialog “Change rule — “…””, the one place a rule is changed. A change never overwrites: saving creates a new version from the date in In force from, and the dialog says so before you save — “Saving creates version 2. Version 1 (10.0 %) is in force since 2026-03-01, ends the day before the new one and stays readable.” The button reads Save new version. Under the fields, Versions lists every version of the rule with its period, its line, its severity and who wrote it.

• The new version starts today by default, or tomorrow when the version in force only began today: a version that has started is the standard of that day and keeps it. A later date schedules the change.
• A version that has been in force is never changed and never deleted; its period ends the day before its successor starts. A version that has not started yet can still be changed: the dialog opens on it, and saving replaces it.
• So “what was my limit in March?” has an answer you can read: the version list shows which line was in force on which days.

Renaming: no new version

The name is the rule’s label, not part of its standard. Change only the Name and the dialog says “Only the name changes: saving creates no new version. …”, and the button reads Save name. A rename creates no version: the new name applies to the rule with all its versions, and the audit journal keeps the previous name. Change the name and the line together, and the version note gains “The new name applies to the rule with all its versions.”; one save writes both or neither. A retired rule is renamed the same way.

Ending a rule

• Retire rule (in the dialog, confirmed first) ends the rule: it is no longer evaluated from today, and a version that only started today ends tonight. A change that was scheduled is dropped with it. The rule and all its versions stay readable under Show the retired rules, and a retired rule opens the same dialog. Saving a new version there takes the rule up again from that version’s date.
• Delete rule appears instead of Retire rule while none of the rule’s versions has been in force — a rule scheduled for a later date, say. Nothing was ever measured against it, so nothing of it needs to be kept.

What a rule protects

A security, a category, a classification or a view that a rule reads cannot be deleted while the rule exists. The refusal — in the page’s message band on the Views and Classifications pages, in the Cannot delete dialog on the securities page — names each rule with its status and its view, as in “Single names at most 10 %” (in force, view “Everything”), and each name links to Risk in the view the rule applies in. Following the link makes that view the active view on every Wealth tab, the way picking it in the view switcher does.

Retiring the rule stops its evaluation but does not free the object: a version that has been in force keeps its subject as the record of what the standard was. Only deleting a rule that was never in force frees it. A security a rule reads can still be retired, and it cannot be merged away into another security.

Your agent’s rules

Your agent writes rules over the API or MCP with its token, and they are the same rules you write on this page. They are in force like yours; nothing waits for your approval. What tells them apart is a word: a rule whose version in force the agent wrote ends its words with “· Agent”, and so do scheduled and retired rules whose version the agent wrote. In the dialog, every entry of Versions names its author, Operator or Agent. Your own rules carry no word, so a page without the agent’s rules reads exactly as before. The audit journal names the token that wrote each version.

An agent’s rule is changed, renamed and retired like any other. A version you save on this page is yours, and the word follows the version in force, not the name: a rename moves no word.

What rules do not do

• Nothing is pushed. A finding exists while it is read: on this page, or by your agent reading the breached findings. There is no notification and no mail.
• No look back. A rule is evaluated for today only, and never for a day before its version’s In force from — testing a rule against a past it did not exist in would be backtesting, which Portfolixir does not do.
• No advice. A rule is your standard, evaluated; it never proposes a trade, a size or an order.