How the discipline layer works

The discipline layer is the thing that took the longest to build and is the least interesting to talk about at parties. Nobody signs up for Sentio because we have a great post-loss cooldown timer. But it's the part of the product that keeps accounts alive, so it deserves a full explanation.

Here's how a rule you write in plain English ends up as a gate that blocks a trade six hours later.

Step 1: You describe the rule

Open the discipline page. There's a text field. You type:

Stop me after 3 trades in a day. If I take two losses in a row, halve my size on the next one. No Unicorn between 2pm and 4pm ET — my win rate is terrible then.

That's it. No dropdowns, no form fields, no boolean matrix. You describe the rules the way you'd describe them to a trader friend over coffee.

Step 2: Sentio AI parses it

The text gets sent to a Claude Sonnet call with a system prompt that knows the shape of the DisciplineSettings object. It returns a structured proposal — a set of concrete parameter changes, not the abstract intent:

• max_trades_per_day: 3
• size_halve_after_losses: 2
• per_model_mute_after_hour: { unicorn: 14 }
• per_model_mute_before_hour: { unicorn: 16 }

You see the diff before you save. If the parser misread you — "no that's after 2pm, not before" — you say so and it re-parses. The intent stays in your words; the compilation just makes it exact.

Step 3: The rules live in your account

Once saved, the rules are properties of your user row in the database. Every scan tick, every alert, every setup check runs through them. They don't live in a config file you have to remember to load — they're the account's state.

Discipline that requires you to remember to apply it isn't discipline. It's another checklist item you'll skip when things get busy. Rules in Sentio are ambient — always checked, always enforced, invisible when they're not tripping.

Step 4: The scanner runs the gates

Here's the sequence when a Forever Model setup finishes forming on NQ 5m:

1. Model detects the structure — sweep, reclaim, IFVG, CISD.
2. Setup passes the model's own quality gates (unraided swings, displacement magnitude, etc).
3. Setup is verified against fresh candles (three-lens verifier — correctness, currency, realness).
4. Setup goes through your discipline rules:
   • Trade count for today ≥ your daily cap? → veto, log, done.
   • Session in a per-model mute window? → veto.
   • Last realized loss inside your cooldown window? → veto.
   • Bias lock says long-only and this is a short? → veto.
5. Only if every gate passes does the alert fire + the setup appear in your scanner list.

A vetoed setup gets logged to your activity feed with the reason. You can see what would have surfaced — useful for calibrating your rules — without it ever reaching the alert bus or the decision dialog.

Step 5: The journal closes the loop

Every trade you take gets journaled — automatically if you use the decision dialog, manually if you log the trade after the fact. Realized P&L feeds back into the rule state. So:

---

Why it works

Every rule in the layer is something you already know you should follow. You're not learning new discipline principles from Sentio — you're outsourcing the enforcement of principles you already believe in but don't consistently apply.

That's the whole product in one sentence. Not we know better than you. Not trust our signals. Just: the rules you wrote at 9am get to survive the 2:47pm version of you.

See the discipline page in the app for the full list of primitives. Or read about the four scenarios the layer is designed to prevent.