Use this workflow to review AI paper-agent decisions one row at a time. Entries, skips, exits, blocked conditions, risk notes, and next actions all need review before a prompt change can be trusted.
Trading Boy does not execute live trades, hold funds, or provide financial advice. This workflow reviews paper decisions so a trader can improve a simulated process. It is not a signal service, live order system, investment recommendation, or approval path for real capital.
Each row should make the agent's decision auditable before the outcome is known. If a required field is missing, the review should mark the row incomplete rather than inferring the missing context after the result.
| Field | Entry review | Skip review | Why it matters |
|---|---|---|---|
| Version | Which prompt and rule set created the entry? | Which prompt and rule set created the skip? | Prevents hidden prompt drift from being mistaken for strategy quality. |
| Setup name | Which written setup fired? | Which setup almost fired or failed? | Separates repeatable patterns from vague market commentary. |
| Evidence | What facts supported the simulated entry? | What facts were missing or contradictory? | Keeps the review grounded in pre-outcome conditions. |
| Invalidation | What would make the thesis wrong? | Was invalidation too unclear to act? | Blocks entries that cannot be reviewed later. |
| Risk note | What paper size, exposure, and drawdown state applied? | Which risk control blocked the trade? | Makes risk discipline visible on winners, losers, and no-trades. |
| Human note | Did the reviewer agree with the rule fit? | Did the reviewer agree with the blocked condition? | Separates agent output from human evaluation. |
| Next action | Keep collecting, tighten one field, reduce size, or retire? | Keep collecting, clarify the rule, or add a blocked condition? | Prevents broad prompt rewrites from small samples. |
Entry row: The AI paper agent logs a simulated long setup after a breakout. It names the setup, describes the market evidence, records invalidation, checks paper size, and notes that related exposure is below the workflow limit.
Skip row: Two hours later, the agent sees another similar setup. The output says the setup is attractive, but it records a skip because the existing simulated position already covers the same market driver. The skip row names the blocked exposure rule.
Human review: The reviewer accepts the entry row and the skip row because both follow the written rule. The outcome of either market move is not used to reverse the review. A missed paper winner that was correctly skipped still counts as discipline.
Next action: No prompt change is made. The sample needs more decisions before the team can tell whether the exposure rule is too strict or working as intended.
This workflow assumes the agent output is structured enough to inspect. If entries and skips do not share consistent fields, start with the AI trading agent output format template. A decision review cannot fix a messy record after the sample ends.
For broader setup work, use the AI paper trading agent workflow to connect rules, prompt, output, risk, review, and versioning.
The decision review should produce the evidence that justifies a prompt change. If the same field fails repeatedly, log that in AI trading agent prompt versioning. If the issue happens once, the conservative decision is often to keep collecting paper evidence.
Use the AI trading agent human review checklist to make sure a reviewer signs off on the next action.
Score the row on completeness before scoring the idea. A complete row names the version, setup, evidence, invalidation, paper risk, and next action. An incomplete row should not be rescued by a good simulated outcome. If the paper trade won but the invalidation was missing, the review should still mark the row weak because the system could not be audited before the result.
For skipped decisions, the review asks whether the agent named a real blocked condition. Good skip reasons include unclear invalidation, stale data, correlated exposure, breached drawdown state, or setup mismatch. Weak skip reasons include vague caution, uncertain market, or low confidence without a rule reference. The skip row should teach the next reviewer why the agent did nothing.
The workflow should produce one of a few conservative outputs. Keep collecting paper evidence when the sample is still small. Tighten one output field when repeated rows are incomplete. Reduce simulated size when confidence keeps overriding the paper risk cap. Add a blocked condition when skips are vague. Retire the workflow when the agent cannot follow the rule even after a narrow fix.
Do not use paper decision review to declare an agent ready for live execution. Even a clean decision record cannot prove liquidity, real fills, slippage, uptime, emotional discipline, or future returns. The useful output is a cleaner simulated process and a more honest next review.
It is a structured paper-mode review of each AI agent entry, skip, exit, risk note, blocked condition, and next action before the prompt or rule set changes.
Skipped decisions show whether the agent respects filters, unclear data, exposure limits, and invalidation rules. Without skips, the sample overstates activity and hides discipline.
No. Decision review organizes simulated practice evidence. Trading Boy does not execute live trades, hold funds, or provide financial advice.