Skip to content
Relay
12 findings

Margaret Ellison, advisor (synthetic persona)Senior advisor, founder and executive practice, New YorkAbout $820M across 183 households; four relationships above $50M; 7 of those households are in this prototypeWho's who

Complaints

Complaints. The complaints desk: identify a grievance in what the client wrote, log it, and start the clock the reporting rule runs on.

Agent read or preparedYou decide or sendClient said or holds

Complaints · ran day 0, 06:25

I read 5 captured messages on email for Margaret Ellison against 1 rule. Nothing fired and nothing fell under a confidence floor.

Read a policy into this desk

How I got there
  1. 1. AgentRead 5 captured messages on emailCadence: on every draft and captured message.
  2. 2. AgentApplied 1 rulesUnlogged client complaint
  3. 3. AgentRaised what fired and what I was not sure ofA rule under its confidence floor reaches a person even when it clears.
  4. 4. YouLeft every disposition to a personI never clear my own findings.

Read every inbound message for grievance language that no one logged, and prepare the log entry and the callback.

Reads
Every captured email and message, and every note before it goes to a client
Checks
Unlogged client complaint
Prepares
a task with an owner; a drafted note, which a person sends; a meeting to schedule
Runs
On every draft and captured message
Never
Never sends, never clears its own finding, and no advisor can switch it off or loosen it without a principal.
How
1 rule, each a stated condition over facts computed from the book; under a rule's confidence floor, or without the source it needs, it asks a person rather than clears.
Built on
FINRA Rule 4513, Records of Written Customer Complaints

Read a written policy into it

ClearRunningFINRA 4513FINRA 4530On every draft and captured message. Last run day 0, 06:25.

Rules this desk watches (1)

Each rule is data: a stated condition over facts computed from the book, the settings a layer may tighten, what the desk prepares when it fires, and the text it is built on with where that text stands in law today.

Unlogged client complaint

Flag for review

FINRA · FINRA Rule 4513 · firm-mandatory

Fires when
the text reads as a complaint and not: a complaint is logged
Prepares
a task with an owner; a drafted note, which a person sends; a meeting to schedule
Needs
Microsoft 365 mail and calendar
In law
In force since December 5, 2011.
Source
FINRA Rule 4513, Records of Written Customer Complaints
Today
Nothing fired.

Change this rule

Findings it raised

None open.

Tune this desk for Margaret Ellison

Editing asA lower layer can only tighten; a refused change stays in the log.

AgentComplaints

Complaints · The complaints desk: identify a grievance in what the client wrote, log it, and start the clock the reporting rule runs on.

Running

FINRA 4513FINRA 4530

On every draft and message. 1 rule, 1 evaluable with what is connected.

  • Refused at the Margaret Ellison layer: enabled to false. A lower layer cannot switch an agent off without a principal's approval. The desk runs for every advisor the firm runs it for.
Why the desks are data, and what an advisor layer may change
Each desk is one entry in the agent catalog: the human team it mirrors, the authorities it applies, the rules it watches and its cadence. Adding a desk is adding an entry. An advisor's layer can switch a desk on, run it more often or give it another rule of its scope, and every change is an entry in the same change log the rules use, replayable to any past moment. It cannot switch a desk off, slow it down or take a rule away: the same tighten-only invariant the rule set has, enforced in the resolver and shown on the desk when an attempt is refused.

Teach this desk a policy

Paste or drop a written supervisory procedure. The reader turns each sentence that carries an obligation into a candidate rule in the same shape the engine runs, cited to the sentence. You add the ones that say what the firm meant; each is in force on the next sweep and lands in the change log in your name.

What the reader does

  • Keeps only sentences with an obligation: must, may not, is required.
  • Matches the facts the engines compute, by the words a policy uses for them.
  • Reads the number and the comparator beside each fact.
  • Takes severity from the verb: a prohibition blocks, a review flags.
  • Cites the rule the sentence cites, or the paragraph.

This is the deterministic reader. In production a model reads the sentence more freely; it still only proposes, and what it proposes is still this data.