A plan evolving over three months

One trader, one setup, one quarter, with the numbers at each step. The setup is the MES overnight-range breakout from Module 1, and every figure is invented to be plausible rather than impressive. What matters is the sequence: what was logged, what was noticed, what was allowed to change, and what was not.

Week 0: v1.0

The plan is the day trader's example from Module 1, with one difference: the filter starts at "overnight range ≥ 0.3% of price". Account $15,000, 1R = $112. Expected numbers on the journal's front page: win rate 40 to 50%, average winner 1.8R, average R around 0.2R after costs.

Month 1 (July): 38 trades

Top mistake tags: late-entry (8), early-exit (5), outside-setup (6, the C-grades), stop-widened (2).

A-grade trades: 21, average +0.31R. B and C: 17, average −0.29R.

Weekly reviews produced four execution actions: finish pre-market by 09:20 (week 1, after three late entries were traced to levels still being marked at the open); grade the setup before the order goes in (week 2); close the platform when the two-loss rule fires (week 3, after both stop-widenings happened after the breaker); no action week 4.

Monthly review. Average R is inside the noise of the expected 0.2R at this sample size, but plan-follow at 58% means the month has not tested the plan. The A-grade split is encouraging and is not a finding at 21 trades. Decision: no plan changes. Execution focus on the C-grade trades: "if any box is unticked, no trade" printed on the checklist card. Candidates list gets one entry: "filter 0.3% may be too loose; trades on 0.3 to 0.4% days: 9, average −0.4R. Wait for 30."

The trader wanted to change the time stop, the filter and the trail. All three went on the list.

Month 2 (August): 41 trades, v1.0 continues

C-grades fell from 6 to 2. Late entries fell from 8 to 2. early-exit is now the top tag at 5.

Filter shadow tag: trades on 0.3 to 0.4% overnight-range days now total 22 (9 + 13), average −0.35R. Trades on days above 0.4%: 57, average +0.30R. Still under 30 in the low band.

Monthly review. Plan-follow at 80% is close to the 85% bar. Average R for the month is inside expectations. The filter split is at 22 trades and a difference of 0.65R: over the effect threshold, under the sample threshold. Decision: no plan change. Execution action: all five early exits were before invalidation and before the stop, and the journal note on three says "looked like it was going to reverse". The invalidation level is added to the order-entry checklist so the exit is compared to it in the moment.

The trader is now impatient with the filter. The rule says wait. The trader waits.

Month 3 (September): 44 trades, v1.1 from 15 Sep

Two weeks in, the low-band shadow tag reaches 31 trades, average −0.33R, against +0.29R above the band. Sample met, effect 0.62R. The candidate goes to test: v1.1, filter 0.3% → 0.4%, valid from 15 September. Prediction: the excluded trades are roughly 28% of volume at −0.33R; removing them should lift overall average R by about 0.15R. Decision rule: evaluate at 30 v1.1 trades; keep if cumulative average R on v1.1 ≥ +0.25R, revert if under +0.15R.

Twenty-three trades on v1.1 is short of the thirty needed to decide. The number looks good and the rule says wait.

Quarterly review, first weekend of October. The six questions:

1. Was the plan followed? 74% at score 3 over the quarter, 86%+ in September. The quarter as a whole did not test the plan; September did.
2. Edge inside expected range? v1.0 cumulative +0.13R on 100 trades, against expected ~0.2R. Below, but at 100 trades the noise is about ±0.3R. Not a finding, and the July execution problems are part of that number.
3. A-grade carrying it? Cumulative A-grade average +0.29R on 66 trades; B/C −0.20R on 34. Yes. B-grades are still 20% of trades, which is high. Execution action: B-grade trades at half size for Q4, as a test.
4. Filter earning its keep? Under test as v1.1.
5. Edge gone? No flags. Winners stable at 1.65R, drawdown inside history, costs unchanged.
6. One major change? None. The v1.1 test continues to 30 trades; the B-grade half-size test runs in parallel because it is an execution rule, not a plan rule, and does not confound the filter test.

Candidates list at quarter end: time stop 30 → 60 minutes (time-stopped trades: 27, average −0.15R, 17 would have reached +1R in the next 30 min; wait for 30). Trail change (no data yet; shadow column added 1 Oct). Adding MNQ (100 MES trades reached; deferred until v1.1 resolves).

Key idea: Over three months, one plan rule changed, on 31 trades of shadow evidence, with a written prediction and a revert threshold. Everything else that improved was execution, and execution was most of the improvement.

What the quarter shows

Average R went from +0.04R to +0.36R across the three months, and almost none of that came from the plan. It came from finishing pre-market earlier, a line on a checklist card, writing the invalidation level down, and closing the platform when a breaker fired. The filter change may add another 0.15R; the trader will know in a few weeks.

The trader wanted to change five things in July. One was tested by October. The rest were resolved by execution, are waiting for sample, or sit on the list with a shadow column. That restraint is the skill this course teaches.

Try it: Draft your own version of this timeline for the next three months, blank: three monthly sheet skeletons, a quarterly review page with the six questions, and a candidates list with three columns: idea, date, trades in subset. Post the July sheet to the trading journals forum when you have it, in R, with the plan version and the one decision.

Recap

• One plan rule changed in three months, with 31 trades of shadow evidence, a prediction and a revert threshold. Everything else was execution.
• Most of the improvement in average R came from execution fixes found at weekly reviews: pre-market timing, grading before entry, closing the platform on breakers, writing the invalidation down.
• Candidates that reached the effect threshold but not the sample threshold waited, and the waiting was the rule working.
• The quarterly review found the quarter as a whole had not tested the plan, and said so, rather than reading a setup verdict into an execution problem.
• Execution tests (half size on B-grades) can run alongside a plan test because they do not change what is being measured.