AAprompt - GC_Probability

STEP 1
For Backend : I need a backend FastAPI server running, ask agent to Install.
Run server from terminal:
cd C:\Users\bogda\Dropbox\VSCode\GC_Probability\backend
C:\Users\bogda\Dropbox\VSCode\GC_Probability\.venv\Scripts\python.exe -m uvicorn main:app --host 127.0.0.1 --port 8000

STEP 2:
We added backend API in main.py: FastAPI endpoint: /intervals/{coin}
Now when opening: http://127.0.0.1:8000/intervals/BTC-USDT
The backend: reads CSV, builds intervals, and the endpoint returns aggregated JSON
From this backend endpoint a new frontend server will read it

Logic

For each coin:

1. no CSV → manual download button active

2. CSV exists → check automatically if update is needed

3. if needed → update CSV

4. rebuild/read JSON from CSV
5. compute GC probability

Backend structure

1. download_full_history.py

bootstrap for one coin
downloads and saves all available daily candles locally CSV

2. update_history.py

reads existing file
fetches only missing new daily candles
appends them

3. build_intervals.py (aggregated: 3D is now combining 3 daily candles into one candle)

reads daily candles CSV
builds: 1 day, 3 days, 1 week, 1 month, 3 months, 6 months, 1 year, 2 years, full
returns JSON (in memory)
optionally saves debug JSON file

4. calculate_gc_probability.py

reads interval data
computes Golden Cross probability

5. config.py

coins list
number of coins = 25
file paths
constants

EMA_short, EMA_long → computed from price series
clamp = "Restricts (or "clamps") a value to fall within a specified range [min, max]"
AverageTrueRange = “On average, how much does the price move per candle?”
closes_above_shortEMA = 7, LookbackCandleCount = 10, means: 7 out of last 10 candles closed above EMA_short
Close(t) = closing price of the candle at time t

GC_probability =

0.26 * (1 - clamp(|EMA_short(t) - EMA_long(t)| / (1.5 * AverageTrueRange(14), 0, 1))

# measures how close the two EMAs are (distance to cross)

# normalized by 1.5 * AverageTrueRange → defines “relevant distance”

# → very close EMAs → value approaches 1

# → far apart EMAs → value approaches 0

+ 0.24 * clamp(((EMA_short(t) - EMA_long(t)) - (EMA_short(t-1) - EMA_long(t-1))) / (0.2 * AverageTrueRange(14)), 0, 1)

# measures how fast the EMA gap is closing (velocity)

# normalized by 20% of a typical candle move (AverageTrueRange)

# → if gap is closing fast → value approaches 1

# → if slow or negative → value approaches 0

+ 0.16 * clamp((((EMA_short(t) - EMA_long(t)) - (EMA_short(t-1) - EMA_long(t-1))) - ((EMA_short(t-1) - EMA_long(t-1)) - (EMA_short(t-2) - EMA_long(t-2)))) / (0.1 * AverageTrueRange(14)), 0, 1)

# measures how the speed of the EMA gap change is evolving (acceleration)

# normalized by 10% of a typical candle move (AverageTrueRange)

# → accelerating toward the cross → value approaches 1

# → slowing down or reversing → value approaches 0

+ 0.14 * (closes_above_shortEMA / LookbackCandleCount)

# closes_above_shortEMA = number of recent candles that closed above EMA_short → computed input (from your candle data)

# LookbackCandleCount = total number of recent candles checked (e.g. 10) → measures how consistently price stays above the short EMA

# → more closes above → value approaches 1 (strong support)

# → fewer closes above → value approaches 0 (weak support)

+ 0.12 * clamp(((EMA_long(t) - EMA_long(t-1)) + (0.05 * AverageTrueRange(14))) / (2 * (0.05 * AverageTrueRange(14))), 0, 1)

# measures the slope of the long EMA (trend resistance)

# remapped using ±5% of AverageTrueRange as a “flat zone” threshold

# → falling long EMA → value approaches 0 (resistance still strong)

# → flat long EMA → value ≈ 0.5 (neutral)

# → rising long EMA → value approaches 1 (trend turning favorable)

+ 0.08 * ((abs(Close(t) - Close(t - LookbackCandleCount)) / sum(abs(Close(i) - Close(i-1)) for i in last LookbackCandleCount))* clamp((AverageTrueRange(14) / Close(t)) / 0.02, 0, 1))

# LookbackCandleCount = total number of recent candles checked (e.g. 10)

# trend_efficiency:

# straight movement / total movement → smooth trend vs noisy

# volatility_support:

# ATR relative to price, scaled so normal volatility → ~1

# Close(t) = closing price of the candle at time t

# Close(t-1) = closing price of previous candle

# Close(t-2) = closing price two candles ago

# comes directly from your candle data (KuCoin API)

# example:

# candle = { open, high, low, close }

# Close(t) = candle.close

GC_probability = clamp(GC_Prob * 100, 0, 100)

# =========================

# COMPUTED INPUTS (from data)

# =========================

# EMA_short(t), EMA_short(t-1), EMA_short(t-2) → computed from price series

# EMA_long(t), EMA_long(t-1), EMA_long(t-2) → computed from price series

# AverageTrueRange(14) → computed from candle data (volatility)

# Close(t), Close(t-1), Close(t-2), ... → directly from candle data

# closes_above_shortEMA → computed by counting closes above EMA_short over recent candles

# =========================

# FIXED PARAMETERS (tunable)

# =========================

# 0.26 → weight of distance term

# |EMA_short - EMA_long| / (1.5 * AverageTrueRange)

# 1.5 → defines how far EMAs can be and still be “close”

# 0.24 → weight of velocity term

# velocity / (0.2 * AverageTrueRange)

# 0.2 → defines what counts as strong gap closing speed

# 0.16 → weight of acceleration term

# acceleration / (0.1 * AverageTrueRange)

# 0.1 → defines what counts as strong increase in speed

# 0.14 → weight of structure term

# closes_above_shortEMA / LookbackCandleCount

# LookbackCandleCount = 10 → number of candles checked

# 0.12 → weight of long EMA slope term

# ((EMA_long(t) - EMA_long(t-1)) + (0.05 * ATR)) / (2 * 0.05 * ATR)

# 0.05 → defines “flat zone” for long EMA

# 0.08 → weight of volatility/efficiency term

# (AverageTrueRange / Close) / 0.02

# 0.02 → defines “normal volatility” level

# =========================

# DERIVED (internal calculations)

# =========================

# EMA gaps, velocity, acceleration

# trend_efficiency

# volatility_support

# → all computed inside the formula from the above inputs