Celestial MegaOdds

This project builds an experimental pipeline for Mega-Sena draws:

1. Parse draw history
2. Compute astrology features (local, no API)
3. Run a backtest predictor with adjustable weights
4. Tune weights via walk-forward optimization

Note: Mega-Sena is designed to be random. The predictor is for exploration only.

Prerequisites

• Node.js 16+ (Node 18+ recommended)
• npm 8+
• Windows PowerShell 5.1+ or PowerShell 7+
• The historical input file Mega-Sena.xlsx in the project root

Check versions:

node -v
npm -v
powershell -Version

Quick start

1. Extract draw history (from the Excel file):

powershell -ExecutionPolicy Bypass -File .\scripts\extract_megasena.ps1

2. Compute astrology features:

node .\scripts\compute_astro.js

3. Run backtest:

node .\scripts\backtest_predict.js

Outputs:

• data/mega_sena.csv
• data/mega_sena.json
• data/mega_sena_astro.json
• data/backtest_results.json

Predictor config (detailed)

The predictor reads defaults from config/predictor.json. Environment variables override any field, so you can test changes without editing the file.

Config file

Path: config/predictor.json

{
  "windowSize": 200,
  "halfLife": 20,
  "explore": 0.1,
  "hotBoost": 0.05,
  "coldBoost": 0.1,
  "coldWindow": 25,
  "ticketsPerDraw": 10,
  "minHistory": 50,
  "seed": 42
}

Fields

• windowSize: How many previous draws are used to compute weights. Larger values smooth noise but can lag shifts.
• halfLife: Recency decay in draws. Smaller values prioritize recent draws.
• explore: Mixes in a uniform distribution (0.0 = no exploration).
• hotBoost: Multiplier for numbers recently drawn (see coldWindow).
• coldBoost: Multiplier for numbers NOT drawn recently (see coldWindow).
• coldWindow: Number of latest draws used to define hot/cold numbers.
• ticketsPerDraw: Number of predicted tickets per draw (backtest only).
• minHistory: Minimum history before prediction starts.
• seed: RNG seed for reproducible backtest results.

Environment overrides

Any of these can be set as environment variables:

• WINDOW
• HALF_LIFE
• EXPLORE
• HOT_BOOST
• COLD_BOOST
• COLD_WINDOW
• TICKETS
• MIN_HISTORY
• SEED

Example (PowerShell):

$env:WINDOW='120'
$env:HALF_LIFE='12'
$env:EXPLORE='0.18'
node .\scripts\backtest_predict.js

Tuning

The tuner runs walk-forward evaluation without peeking at future draws.

node .\scripts\tune_predictor.js

For full coverage, the tuning can be split into chunks and merged:

$env:STRIDE='1'; $env:RANDOM_TRIALS='20'; $env:START_IDX='50'; $env:END_IDX='800'
node .\scripts\tune_predictor.js .\data\mega_sena_astro.json .\data\tuning_chunk_1.json .\data\tuning_chunk_1.csv

$env:STRIDE='1'; $env:RANDOM_TRIALS='20'; $env:START_IDX='800'; $env:END_IDX='1550'
node .\scripts\tune_predictor.js .\data\mega_sena_astro.json .\data\tuning_chunk_2.json .\data\tuning_chunk_2.csv

$env:STRIDE='1'; $env:RANDOM_TRIALS='20'; $env:START_IDX='1550'; $env:END_IDX='2300'
node .\scripts\tune_predictor.js .\data\mega_sena_astro.json .\data\tuning_chunk_3.json .\data\tuning_chunk_3.csv

$env:STRIDE='1'; $env:RANDOM_TRIALS='20'; $env:START_IDX='2300'; $env:END_IDX='2954'
node .\scripts\tune_predictor.js .\data\mega_sena_astro.json .\data\tuning_chunk_4.json .\data\tuning_chunk_4.csv

Merge and report:

node .\scripts\merge_tuning_chunks.js .\data .\data\tuning_results_full.json .\data\tuning_results_full.csv
node .\scripts\report_tuning.js .\data\tuning_results_full.json .\data\tuning_report_full.html

Predict next draw (CLI)

Generate tickets for any target draw date/time using the current tuned config:

node .\scripts\predict_next.js --date YYYY-MM-DD --hour 21 --minute 0 --tickets 20

Example:

node .\scripts\predict_next.js --date 2025-12-31 --hour 22 --minute 0 --tickets 20

Or via npm script:

npm run predict:next -- --date 2025-12-31 --hour 22 --minute 0 --tickets 20

Optional flags:

• --input path to historical dataset (default: data/mega_sena_astro.json)
• --out output JSON path
• --csv output CSV path
• --lat latitude (default: -23.564)
• --lon longitude (default: -46.651)

Notes

• The astrology calculations use circular-natal-horoscope-js.
• Lunar phases include phase name, angle, and illumination.
• The predictor is experimental and not intended for real-world betting.