Workflow

Paper trading agent risk control workflow

Use this workflow when an AI paper trading agent needs a repeatable risk review before another sample, prompt change, or rule change. The goal is to make simulated size, exposure, drawdown, skips, and pause rules visible before paper PnL can distract the review.

Start free Open AI agent risk controls

Paper risk controls are review gates

Trading Boy does not execute live trades, hold funds, or provide financial advice. A paper trading agent risk control workflow reviews simulated practice evidence only. It should never be used as a live bot approval process, buy or sell signal, or guarantee that a strategy will work with real fills.

Risk-control workflow

Run these gates in order. A paper winner that breaks a risk rule should fail the workflow because the review is about process quality, not the outcome of one simulated trade.

Gate	What to inspect	Evidence to keep	Stop condition
Version freeze	Prompt, persona, watchlist, risk cap, and output format.	A version label tied to the sample window.	The sample mixes hidden prompt or rule changes.
Size cap	Paper position size, stop distance, and risk per idea.	Planned size from the position size calculator.	The agent increases size because confidence is high.
Exposure check	Open simulated positions, correlated ideas, sector overlap, and repeated entries.	A row that names each related paper position.	The agent stacks similar trades without a written reason.
Drawdown gate	Current simulated drawdown and maximum allowed loss for the review window.	A drawdown note linked to max drawdown.	The paper agent keeps acting after a pause rule should trigger.
Skip discipline	No-trade decisions when setup, data, spread, invalidation, or risk is unclear.	A skip reason in the agent journal.	Only entries are recorded, so discipline is invisible.
Pause rule	Rules for stopping the version after repeated risk breaks or unclear data.	A pause note and next review action.	The team keeps testing because a recent paper trade won.
Next change	The single risk behavior that should change before the next sample.	A prompt-versioning or rule-change note.	The review rewrites size, setup, watchlist, and output at once.

Example risk-control review

Version: A simulated AI paper agent runs a breakout workflow on a small crypto watchlist. The version uses a 1 percent paper risk cap per idea, a rule that blocks highly correlated entries, and a pause rule after two size violations in one week.

Risk event: The agent logs three potential entries during the same market move. The first idea is eligible. The second and third share the same driver, so the workflow treats them as correlated exposure rather than independent opportunities.

Review: The first simulated entry stays in the sample. The second and third are logged as skips with the blocked condition. Paper PnL is not used to override the exposure rule because the outcome is not known when the decision is made.

Next action: Keep the version active only if the next sample continues to log correlated skips. If the agent keeps suggesting stacked entries, move to prompt versioning and add a stricter exposure field to the output format.

How this differs from a generic risk page

The broader AI agent risk controls page defines the controls that should exist: size, drawdown, exposure, invalidation, frequency, and pause rules. This workflow shows how to inspect those controls after real paper decisions have been collected.

That distinction matters because a rule can look complete in a template and still fail during review. The agent might name a risk cap but ignore it after a strong signal. It might describe exposure but still log several related entries. The workflow catches those behavioral gaps.

Where it fits in the AI agent cluster

Use AI trading agent rules to write the system, AI trading agent prompt template to express it, this page to review risk behavior, and the evaluation checklist to decide whether the paper sample is clean enough to keep studying.

If the agent output is too loose to review, tighten it with the AI trading agent output format template before collecting another sample.

Risk notes to keep in every sample

A usable paper sample should show the planned size, actual simulated size, stop or invalidation, related open ideas, drawdown state, skip reason, and any pause trigger. These fields let a reviewer see whether the agent respected the rules before the outcome was known. Without them, the review can drift into hindsight: winning oversize decisions look smart, and losing oversize decisions look reckless, even though both broke the same control.

Keep risk notes boring and structured. The best paper risk review does not need dramatic language. It needs enough evidence for another reviewer to understand what happened, whether the agent followed the written system, and which single change should happen next. If the page is being used with the AI paper trading agent workflow, archive the risk review beside the version record so the next sample can be compared cleanly.

Common failure modes

One common failure is confidence-based sizing. The agent says the setup is strong and increases simulated size even though the written rule does not allow it. Another failure is missing skip evidence. If only entries are logged, the paper record cannot show whether the agent avoided weak conditions. A third failure is drawdown amnesia: after a losing sequence, the agent keeps creating entries because each individual trade looks acceptable in isolation.

Each failure should map to a narrow next action. Confidence-based sizing can become a hard size field. Missing skips can become a required blocked-condition field. Drawdown amnesia can become a pause rule that must be checked before any entry is eligible. Avoid broad rewrites unless the evidence shows the entire system is unclear.

Paper trading agent risk control workflow FAQ

What is a paper trading agent risk control workflow?

It is a repeatable paper-mode process for checking whether an AI trading agent stayed inside simulated size, exposure, drawdown, skip, and pause limits before the next review action.

Should an AI paper agent skip trades?

Yes. A healthy paper agent should document skipped trades when the setup, data, risk, or invalidation rule is unclear. Skips are part of the evidence.

Can this workflow approve live trading?

No. The workflow reviews simulated evidence only. Trading Boy does not execute live trades, hold funds, provide financial advice, or approve live capital decisions.