A useful paper-trading system does not stop at entries and exits. Use a repeatable feedback loop to convert journal evidence into better agent rules and cleaner review habits.
The feedback loop is the part of the system that stops paper trading from becoming random practice. Each loop should connect a written pre-trade note, a simulated decision, a post-trade review, and one specific next action. The goal is not to make every trade look smart. The goal is to make the process easier to inspect and improve.
| Loop step | Output | Common mistake |
|---|---|---|
| Pre-trade note | Setup, invalidation, paper size, and review question | Entering without a testable hypothesis. |
| Decision record | Agent rationale, context, and alert history | Treating the alert as a signal instead of workflow evidence. |
| Post-trade review | Result, rule fit, behavior, and process quality | Judging the trade only by profit or loss. |
| Lesson classification | Rule issue, execution issue, market issue, behavior issue, or sample-size issue | Changing too much after one paper trade. |
| Rule update | One clear adjustment or a decision to keep collecting samples | Adding complexity without evidence. |
Observation: The agent repeatedly enters late after strong one-hour moves.
Classification: Timing issue, not a thesis issue. The setup may be valid, but the entry trigger is too loose.
One rule update: Require the pre-trade note to identify whether the entry is early, on-time, or late relative to the planned trigger.
The loop should produce fewer vague entries, clearer invalidation notes, better paper sizing, and agent rules that are easier to inspect. A strong feedback loop also knows when not to change anything because the sample is too small.
Avoid rewriting the whole system after one result, using paper profit as proof of edge, or adding more indicators because the review feels uncomfortable. A narrow process update is usually more useful.
A weekly review should group paper trades by setup type and behavior tag. Look for repeated timing issues, repeated sizing issues, repeated invalidation changes, and repeated cases where the agent acted outside its persona. One isolated mistake can be noise; a repeated mistake is a candidate for a rule update.
When the evidence is mixed, the correct action may be to keep collecting samples under the current rule rather than changing the agent too quickly.
Not every review note deserves the same weight. A repeated behavior tag across similar setups is stronger evidence than one dramatic outcome. A rule violation that appears before the result is known is stronger evidence than a story written after the trade. A risk issue that appears across several paper entries is stronger than one entry that simply lost money.
Use the hierarchy to keep the feedback loop calm: journal completeness first, rule fit second, risk consistency third, outcome last. Outcomes still matter, but they should not be the only reason a process changes.
This hierarchy is especially useful when paper results look good. A profitable run with weak thesis notes, late entries, or drifting risk can still be a warning that the workflow is not ready to expand.
Ownership keeps the loop from becoming anonymous churn. Every rule change should have a reason, an expected effect, and a later review date.
It is a repeatable review process that connects pre-trade notes, simulated decisions, post-trade outcomes, behavior tags, and one specific process update.
Rules should change only when the journal evidence supports a specific adjustment. One paper trade is usually too small a sample unless it reveals a clear rule violation.
Use repeated journal patterns, rule-fit tags, risk review, and comparable paper samples. Avoid rule updates based only on a single profit or loss result.
This feedback loop is for simulated trading practice and process review. It does not predict future returns or recommend live trades.