Between April and September 2026 I closed 43 of my bot's 413 live trades by hand, using the exit button on my dashboard. That average of about one in ten hides the real story: 10 of the 43 came in the first four months, and 33 came in August and September.
I've mentioned the manual exit button before, in paper trading vs live trading and in the September scorecard. This post puts all the numbers in one place and asks the question I've been avoiding: did stepping in help?
The data
Every live trade my bot takes is logged with a fixed exit reason, from a short list that includes "manual exit" (why the journal uses a fixed list). For this post I counted every live trade record from 9 April, the first live trade, to 29 September, the last live trade in the copy of the log I used. That's 413 records over 79 trading days.
I split them into two groups:
- Manual exits: trades I closed from the dashboard (43).
- Rule-based exits: everything else, meaning stops, targets, time exits and end-of-day closes (370).
To keep money out of it, I measure results in average losses. One unit is the average loss of a rule-based losing trade. A manual winner of 0.56 means the average manual win was a little over half the size of a typical rule-based loss. This is the same idea as the payoff ratio in profit factor, payoff ratio and expectancy.
How often I stepped in, month by month
| Month | Live trades | Closed by hand | Manual wins | Manual losses |
|---|---|---|---|---|
| April | 72 | 2 | 1 | 1 |
| May | 75 | 5 | 4 | 1 |
| June | 115 | 2 | 2 | 0 |
| July | 45 | 1 | 1 | 0 |
| August | 63 | 16 | 7 | 9 |
| September | 43 | 17 | 10 | 7 |
From April to July I closed 10 of 307 live trades myself, 3.3%. In August and September it was 33 of 106, 31.1%. In all, the 43 manual exits were spread over 20 of the 79 days with live trades.
The month the rate jumped is also the month I added a daily cap on live trades (the guards post explains the cap). I can't tell from the log whether one caused the other. Fewer live trades a day may have made each one feel more important, or it may be a coincidence with something else that changed in August. All I can say is that the two happened together.
Were the manual exits better or worse?
Here are the two groups side by side for the whole period.
| Manual exits | Rule-based exits | |
|---|---|---|
| Trades | 43 | 370 |
| Wins | 25 | 212 |
| Losses | 18 | 157 |
| Win rate | 58.1% | 57.3% |
| Average win (in average losses) | 0.56 | 0.51 |
| Average loss (in average losses) | 0.84 | 1.00 |
| Payoff ratio | 0.66 | 0.51 |
| Average time in the trade | 8.0 minutes | 7.7 minutes |
(One rule-based trade closed flat, which is why 212 and 157 add to 369.)
On this table, my manual exits look slightly better than the rules: about the same win rate, smaller losses and a higher payoff ratio. That matches what I wrote in the paper vs live post, that most of the trades I closed were in profit when I closed them.
But the average hides a change over time. Here is the average result per trade, again in average losses, split at the start of August:
| Period | Manual exits | Rule-based exits |
|---|---|---|
| April to July | +0.157 (10 trades: 8 wins, 2 losses) | −0.116 (297 trades) |
| August to September | −0.084 (33 trades: 17 wins, 16 losses) | −0.206 (73 trades) |
| Whole period | −0.028 (43 trades) | −0.134 (370 trades) |
From April to July I stepped in rarely, and when I did, 8 of the 10 trades ended as wins. In August and September I stepped in on almost a third of trades and the wins and losses came out nearly even: 17 against 16.
The rule-based trades were weaker in August and September too, so I was stepping in during a harder period for the strategies themselves. That makes the comparison less clean than it looks. A manual exit still did better than the rule-based average in both periods. But "less bad than the rules in a bad patch" is a very different claim from "my judgement adds something".
A worked example of reading these numbers
The whole-period figure for manual exits, −0.028, can be checked roughly from the first table:
- 25 wins × 0.56 = 14.00 units won
- 18 losses × 0.84 = 15.12 units lost
- Net: 14.00 − 15.12 = −1.12 units over 43 trades, or about −0.026 per trade
That's close to the −0.028 calculated from the unrounded figures; the difference comes from rounding 0.56 and 0.84. The point of the check is that a 58% win rate still produced a small net loss, because the average manual loss (0.84) was bigger than the average manual win (0.56). A win rate alone would have told me I was doing fine.
When in the day I step in
| Trade opened | Live trades | Closed by hand |
|---|---|---|
| 9 AM hour | 73 | 14 |
| 10 AM hour | 101 | 10 |
| 11 AM hour | 59 | 8 |
| 12 PM hour | 49 | 7 |
| 1 PM hour | 54 | 1 |
| 2 PM hour | 66 | 2 |
| 3 PM hour | 11 | 1 |
32 of the 43 manual exits were on trades opened before noon: 13.7% of the 233 morning trades. After noon it was 11 of 180, or 6.1%. The first hour stands out most, with 14 of 73.
Part of this is simply that more trades open in the morning, but the rate is higher too, not just the count. My time-of-day study looks at when the Nifty itself makes its high and low; this log can't tell me whether the market's mornings or simply my being at the screen is what draws me in.
What the data can't tell me
The most important question is what each trade would have done if I'd left it alone. I meant to answer it from the price path recorded after entry, but only 7 of the 43 manual-exit trades have that path stored, against 173 of the 370 rule-based ones. Seven trades are too few to say anything, so I haven't measured this.
Other limits:
- 43 trades is a small sample. One or two large trades either way would move every average in the tables above.
- The two groups aren't random. I chose which trades to close, so the manual group isn't a fair sample of all trades. I probably stepped in more on trades that were already in profit.
- Different strategies were live in different months. Part of the change between the two periods comes from which strategies were running, not only from me.
What I take from it
The honest summary is that my manual exits haven't clearly hurt, and that's exactly why they're a problem. They're not bad enough to force a change, but each one removes a trade from the test of the rules. In August and September I removed almost a third of them.
Two things follow directly from the numbers:
- The bar from the September scorecard still applies: five or fewer manual exits a month. September's 17 was well above it, and each monthly scorecard reports the count.
- The missing price paths are the real gap. With 7 of 43 recorded, the most useful question here can't be answered yet. Storing the path for every live trade would let a future version of this post compare each manual exit with what the rule-based exit would have done.
The first hour is where 14 of the 43 happened, so it's the obvious place to watch.
If you're running your own bot, the cheapest version of this is a fixed exit-reason field in the journal and a monthly count. You don't need anything clever to see a jump from 3% to 31%.
This post describes my own experience building and running a trading system, for information only. It is not investment advice, and past results say nothing certain about the future. I am not registered with SEBI as an investment adviser or research analyst.