Insights · AI Methodology

How our trading bot works.

We split trading into two parts from the start. One half decides where to enter — strategy, filters, the entry point. The other decides how to open and exit — margin, averaging, stops, and risk limits. AI is only the executor; the strategy is ours.

One bot, two halves

Analysis

5 simulations, 24/7

Three run in the background for reference points, two for the entry point. AI doesn't decide or adjust anything on its own — it's only an executor.

Execution

30 / 70 + averaging

30% of the deposit is always in reserve. Entry 0.35%, averaging ×4 then ×2, max 5 positions, with a hard 30% drawdown cap.

Targets

Reactions, not movements

Short targets — 0.3–5% by timeframe. The trailing stop quickly moves into profit, so a position almost cannot close in the red.

We don't guess the market — we catch reactions

AI

Only an executor

AI doesn't decide or update anything on its own. It simplifies work with the algorithms, runs checks, and delivers data. The strategy is set by us, not the model.
Method

References → confirmations → filters

No forecasts. First we find the reference point, then we wait for confirmation, then we check the filters. Only then — entry.
Targets

Short and guaranteed

We catch the reaction, not the big move. Often after our take-profit the price runs another +10% — we take our small, guaranteed piece and exit.

We don't try to guess what will happen. We wait until the market itself confirms the reaction — and take the short, guaranteed part. The strategy is conservative on risk and active in the number of trades.

Three simulations run 24/7 in the background

Simulation 01
Once · at server startup

Cold start

On a restart or new machine it runs our entire strategy across all coins from the past year up to now. The output is an up-to-date reference point for every coin — then it has done its job.

Simulation 02
Continuous · in parallel

Reference updates

Continuously runs the parallel strategy on every coin and updates the reference points — built from Fibonacci retracements, candles, Fibonacci pivots, divergences, and local maxima/minima.

Simulation 03
Continuous · for opening positions

Clean references for entry

The entry itself uses many filters, but references must update without them — only volume, RSI, Ichimoku, and Bollinger. This keeps the reference point accurate and undistorted.

We trade on market reactions, so extreme highs and lows are critical benchmarks — it's from them that we look for the next entry point.

Entry decision: two more simulations + a macro filter

A

Opening zone

The first simulation answers: are we currently in a zone where we can open at all?

  • Fibonacci levels
  • Hidden divergences
  • Bollinger overbought zones
  • Filter check: CMD, BB, volume
B

Specific entry point

The second simulation finds the exact level to enter on. Our entries are always at levels, never at market.

  • Logarithmic support / resistance
  • Fibonacci pivot points
  • Fibonacci retracement levels
  • Trend channels

Complete entry scenario — step by step

01
Fresh reference
A background simulation already holds an up-to-date reference point for the coin.
02
Entered the zone
Simulation A confirms divergence, a Fibo zone, and Bollinger — we're in a possible entry area.
03
Filters OK
CMD, Bollinger, and volume all confirm the reaction.
04
Macro filter
Total cap + BTC dominance + USDT dominance give the overall market a green light.
05
Level touch
Simulation B holds the specific level. Price touches it — the position opens.

If any of the five steps lacks confirmation, no entry happens. The bot doesn't guess — it waits until the entire chain aligns.

Our own margin model — 30% always in reserve

Client deposit — how we split it

70% — working margin
30% — reserve
0%70%100%

For each entry we use 0.5% of working margin. Since working margin is 70% of the deposit, the actual entry volume is 0.35% of the client's total deposit.

Averaging — twice

If the market gives a better point, we add to the position — averaging by level, not from panic.

Entry
0.35%
of total deposit
1st averaging · ×4
→ 1.75%
adds +1.4% (0.35 + 1.4)
2nd averaging · ×2
→ ≈ 3%
doubles the current volume

A position that has passed both averagings occupies at most about 3% of the deposit — the ceiling for a single trade's size.

Five positions at once. A hard 30% stop-circuit.

Limit

Max 5 concurrent positions

The bot won't take more than five open trades at once — even if the strategy keeps firing signals.

Open
Open
Open
Open
Open
Stop-circuit

Total drawdown ≤ 30%

If all open positions combined reach minus 30% of the deposit, they all close automatically — the client can't physically go beyond it.

  • Calculated cumulatively across all trades
  • Triggers identically for everyone — black-day protection
  • With the 30% reserve, a double safety margin

Smart limit return — positions "closed in profit"

Once a position's stop-loss is moved into profit, the bot knows it will close in the positive zone either way, so it counts as "already completed" for the limit.

Example: 5 of 5 open, with 3 stop-losses already in profit. The bot allows 3 more — temporarily up to 8 — because the green ones won't stay long.

SL +
SL +
SL +
Open
Open
free
free
free

The stop-loss steps into profit

example

How the step works

  • The trigger fires on +0.4% movement
  • Stop-loss instantly moves to +0.2%
  • Then a +0.1% step for each position increase
  • At most it closes near zero or small profit
  • In the red — practically never

Values shown are illustrative. Actual parameters are set in configuration and may differ.

Ladder for an example position

Price moved
+0.4%
+0.5%
+0.6%
+0.7%
+0.8%
+0.9%
SL moves to
+0.2%
+0.3%
+0.4%
+0.5%
+0.6%
+0.7%
SL follows the price with a ~0.2% gap.

In a heavily manipulated market — aggressive wicks, a large gap between trigger and market price — such a position can close at a small loss. These cases are rare and don't break the overall strategy profile.

We trade on 12 timeframes, analyze on more

Timeframes for entry

3m10m13m15m21m30m45m89m4h1d3d1w

The analysis runs on even more timeframes at the same time, giving multi-level confirmation.

0.3%
Minimum target

On the shortest timeframes.

~1–2%
Typical target

Hourly and around it.

5%
Maximum target

On daily and weekly positions.

Why: we catch the reaction, not the entire move. On minutes the reaction is small, so we take 0.3%; on weekly the reversal is bigger, so up to 5%. Sometimes the price runs another +10% after our take — that's fine. Our job is the guaranteed piece, not the maximum.

The bot in three columns

Analysis5 simulations

Background sim.
3 (start + references + entry)
Sim. for entry
2 (zone + level)
Reference by
Fibo, candles, divs, min/max
Filters
CMD, BB, RSI, Ichimoku, volume
Macro
totalCap + BTC.D + USDT.D
AI role
executor only
Entry
always by level

Executionmargin

Deposit reserve
30% (not used)
Working margin
70%
Entry
0.5% of 70% = 0.35%
1st averaging
×4 → 1.75%
2nd averaging
×2 → ≈ 3%
Max positions
5 simultaneously
Stop-circuit
−30% total

Targets & stopsshort

SL trigger
+0.3% to +1%
Initial SL
+0.2%
Trail step
+0.1%
Closing in red
almost impossible
Minimum target
0.3%
Maximum target
5%
Entry TFs
3m … 1w (12)

Analysis

Five simulations that check each other. Until reference → zone → filters → macro → level all converge, the bot doesn't enter.

Execution

A strict risk model: 30% always in reserve, max 3% per trade, max 5 positions, total drawdown cut at 30%. Targets short and guaranteed; the stop-loss moves into profit fast.

Conservative on risk, active in trades

See the method turn into live signals.

Risk acknowledgement

This describes the general architecture and operating logic of the trading bot, not every strategy or parameter. AI-generated analysis and signals are informational tools, not investment advice. Models can be wrong and market conditions change. Trading involves substantial risk and the potential loss of capital.