Market Mentor watches the Indian market all day and tells you which shares are worth buying. The part that finds good shares works well. The part that decides when to sell them is not yet consistent, and that is the piece we are fixing next. This note explains both, in plain language.
Working well
Finding shares breaking out of a range, with volume and market strength behind the move, and checking the underlying business before showing them to you.
Being improved
Deciding the exact price to cut a loss and the price to take a profit. One part of the app does this properly; the others do not do it the same way, and that inconsistency is what we are addressing.
There are three jobs the app does: it finds opportunities, it checks whether they are any good, and it keeps score of what happened. Here is each part in turn.
Every share spends most of its life trading inside a range. It bounces around between a floor and a ceiling for weeks, then eventually breaks out and starts a new trend. The scanner draws a box around that range — the floor and ceiling of where the share has been trading — and watches for the moment price pushes out of it.
A breakout on its own means very little. Prices poke above a ceiling all the time and fall straight back. What matters is whether the breakout has conviction behind it, so the app also measures how much trading volume came with it, and whether the share is outperforming the wider market.
For example
A share has been trading between ₹240 and ₹260 for a month. Today it closes at ₹264 — above the ceiling. On its own, that is unremarkable.
But today's volume was 2.1 times its normal daily volume, and over the past month it has risen 8% while the wider market rose 2%. Now it is interesting: a lot of people bought it today, and it has been stronger than the market for a while. That combination is what the scanner looks for.
Finding a breakout is easy. Deciding whether it is worth your money is the hard part. Twice a day, the Deep Analysis engine takes 250 smaller Indian companies and puts every one through a five-point test. A share has to pass on trend, on momentum, on strength relative to the market, and on the maths of the trade itself before it appears on your screen.
Anything that survives is then measured on fourteen further things — how volatile it has been, how easily it can be bought and sold, whether its recent volume is rising or falling, how fresh the trend is, and so on. All fourteen are shown on the card, so you can see why a share was chosen rather than being asked to trust a ranking.
A good chart on a failing company is a trap. Every night the app downloads the full company page for each candidate — quarterly results, who owns the shares and whether that is changing, whether promoters have pledged their holdings, and the transcripts of management calls.
That information is put in front of an AI model along with the chart data, and it is asked a specific question: is this a good business, or just a good-looking chart? It can rank the shortlist, and it can refuse a name outright.
How the refusal works in practice
A share can appear as a candidate to bet against because its chart is weak. The model reads the accounts, and if it finds the business is actually strengthening — record quarterly sales, rising profit — it refuses the trade with its reason stated.
A falling chart on a company whose profits are at an all-time high is usually a temporary dip, not the start of a decline. The app now drops any pick the model explicitly refuses.
The same idea applied to the shares that have options available. The Buy desk finds shares expected to rise; Put picks finds shares expected to fall. Both refuse to fight the wider market — if the whole market is rising, a bet on a share falling has to clear a higher bar before it is shown.
Shares still inside their box, where the daily range is getting tighter and volume is drying up. That squeeze often comes immediately before a large move. These are for watching, not chasing — the breakout has not happened yet.
The app can place real orders. Position size is worked out from risk rather than from how much cash is available, so every trade risks the same rupee amount whatever the share price. Trades can also be placed automatically when a scan finds something, though that is switched off unless deliberately armed.
On every holding you can record what you expect to happen and why, across eight timeframes from five years down to one day, and check later whether you were right. The app's opinion and your opinion are tracked separately.
Nothing is judged on memory. Every pick is followed daily until it either reaches its target, hits its stop, or runs out of time, and the outcome is stored permanently. This is what will let us measure the change described below with real numbers rather than opinions.
Every trade needs two prices decided in advance: the point where you accept you were wrong and get out, and the point where you take your profit. One part of the app works these out properly. The others do not do it the same way, and that is what we are fixing.
A stop-loss is the price at which you admit the idea was wrong. It has one job that is harder than it sounds: it must be far enough away that ordinary daily wobble does not reach it, but close enough that a real reversal does not cost you a fortune.
Set it too close and you get thrown out of good trades by noise, over and over, losing a little each time. Set it too far and one bad trade wipes out several good ones. Almost everything about whether a strategy makes money lives in this one decision.
The best method in the app measures how much each share moves on a normal day — its Average True Range, or ATR. A share that typically swings ₹5 in a day has an ATR of ₹5. It then places the stop a set number of normal days' movement below the entry price, rather than at some fixed percentage.
The same rule on two very different shares
A calm share at ₹250 that usually moves about ₹4 a day. Its stop goes roughly ₹10 away, at ₹240 — about four per cent.
A wild share, also at ₹250, that routinely swings ₹15 a day. Its stop goes far wider — around ₹212, some fifteen per cent away.
Both stops are the same distance in the language that share speaks. A flat "always use 5%" rule would be far too loose for the first share and would throw you out of the second one almost immediately.
The profit targets follow from the same number. If the stop is ₹10 below the entry, you are risking ₹10, so the first target is set ₹20 above — twice what you risked — and the second ₹30 above. The reward is always defined as a multiple of the risk, by design rather than by luck.
Most people assume you need to be right more often than wrong. You do not. What matters far more is how much you make when right compared with how much you lose when wrong.
Why being right less than half the time can still work
Suppose you make ₹2,000 every time you are right and lose ₹1,000 every time you are wrong. Out of 10 trades you are right only 4 times.
4 wins × ₹2,000 = ₹8,000. 6 losses × ₹1,000 = ₹6,000. You are wrong more often than right and you still finish ₹2,000 ahead. That ratio — average win divided by average loss — is the single most important number in the whole system, and sizing the stop properly is what protects it.
Not every engine calculates the stop this way. Some take whatever stop happens to be sitting on the record already — the floor of the trading box, a level chosen because of where the share has been, with no relationship whatever to how much that share moves in a day.
Why a box floor makes a poor stop
A share breaks out at ₹264. The floor of its old box is at ₹240 — but this share routinely swings ₹15 in a day.
Two ordinary down days and it is at ₹234, through the stop, and you are out — not because anything went wrong, but because the share breathed. The box floor was never a measure of risk. It was just the bottom of a rectangle.
There is a second complication. When the AI model runs, it can supply its own stop and target. When the model is unavailable, a different set of levels is used instead. That means the rule governing every exit could switch depending on whether one external service was reachable.
Five steps. The first two are straightforward engineering. The last three are measurement, and most of the machinery needed to measure them is already running.
Use one method for every pick
Move every engine onto the same volatility-based stop, with targets defined as a multiple of the risk. One formula, calculated in one place, used everywhere — so the stop is always sized to how the share actually moves.
About a day's work
Let the AI adjust the level, never replace it
The model should be able to widen or tighten a stop within set limits and give its reason — it often knows something the maths does not, such as results being announced next week. What it must not do is substitute a number of its own invention, and the levels must stay the same when the model is unavailable. An outage should cost us the commentary, not change the strategy.
Half a day, on top of step 1
Prove it, rather than argue about it
Every options scan already runs twice — once with the current rules, once with the new ones — and stores both sets of results separately, scored by the same scorecard. Only the first set can ever place a real order; the second is purely for comparison.
Point that same machinery at the equity engines and let it run. After a few weeks the ledger tells us which method is better, with real numbers, and nobody has to win an argument.
Runs alongside normal trading, a few weeks
Set profit targets from evidence, not round numbers
For every position the app already records the best price it reached before it closed. That tells us how far these trades actually run — which is the only honest basis for deciding where to take profit. A target placed where trades have historically reached is a measured decision; "take 10%" is a guess that merely looks precise.
A few days, once step 3 has data
Retune the live trailing stop
On real money the app takes part of the position off at a small gain, moves the stop to break-even, and lets the rest run behind a trailing stop. The exact settings were chosen from a rough observation and were always meant to be revisited once we had real data. That data now exists and is ready to be examined.
A day, and overdue
Everything above will be measured on real, recorded outcomes before it is trusted — not on expectation. The scorecard runs the current method and the new method side by side, and only the current, unchanged method ever places a real order while the comparison is under way.
Anything we conclude will carry its own honest caveats: tracked outcomes are not the same as real broker fills, early samples are small, and no approach has yet been tested through a sustained falling market — which is exactly when stop-losses matter most.