Grid Bot Architecture

This note describes the intended grid-bot design.

Core idea

A survivable grid bot has three layers:

Micro layer , places trades and captures oscillation.
Meso layer , adapts the grid structure.
Macro layer , protects capital in bad regimes.

If one layer is missing:

micro only, the bot can blow up
macro only, it barely trades
meso only, the behavior is unstable

Operational model

The bot should not be static. It should:

scale spacing with volatility
slide with the market
recenter only occasionally
stop or reduce activity in strong trends

Pseudocode outline

INIT:
  center_price = current_price()
  grid_levels = 12
  recenter_threshold = 0.05
  max_inventory_pct = 0.7
  trend_filter_enabled = true

  build_grid(center_price)

LOOP:
  price = current_price()
  volatility = ATR(lookback=50)

  grid_step = clamp(k * volatility, min=0.008, max=0.025)

  handle fills
  slide grid when price leaves the active band
  recenter only when deviation is large
  pause new orders in strong trends

Key mechanisms

1. Sliding grid

This is the default adjustment mode.

Effect:

keeps the bot active
avoids full reset shock
behaves more like a market maker

Use this as the main structural adjustment.

2. Re-centering

Use only when price drifts too far.

Rule of thumb:

keep it rare
use only for larger deviations, e.g. 5 to 8 percent

Too much recentering kills the edge.

3. Volatility-based spacing

Fixed spacing is fragile.

Better:

high volatility, wider grid
low volatility, tighter grid

This reduces overtrading and avoids getting wrecked in spikes.

4. Trend filter

This is the survival layer.

A simple version can use:

price vs MA200
slope of MA50

When a strong trend is detected:

pause new orders
optionally allow only one-sided behavior

Tunable parameters

Suggested starting range:

grid spacing: ~1 to 1.5 percent for XRP
grid width: ~±10 to 20 percent
recenter threshold: 4 to 8 percent
inventory cap: never above ~70 percent of capital
volatility multiplier k: ~0.8 to 1.5

Failure modes

Common ways grid bots die:

always-on grid with no trend filter
over-adjustment and too many recentering events
ignoring inventory drift
spacing too tight, so fees eat the edge
static behavior while market regime changes

Mental model

Think of the layers like this:

Grid , engine
Sliding , steering
Volatility scaling , suspension
Trend filter , brakes

Summary

A good grid bot sells local volatility, respects regime shifts, and stays bounded in inventory. The goal is not constant activity, it is controlled adaptation.

Grid_Bot_Architecture.md 2.7 KB Istoric Crud