LotteryApp — Lottery Analyzer Pro

A Python/PyQt6 desktop app to fetch, analyze, and study Kentucky Powerball data.

Version 2.5.3. These tools are for analysis and entertainment. Nothing here predicts lottery numbers or improves your odds — every 5+1 combination has the same probability, 1 in 292,201,338. Where the app produces numbers, they are labeled as best historical fit, not predictions.

---

Features

• Data — fetch recent draws from the Kentucky Lottery API, or load the full modern-era history (~1,600 draws since the Oct 2015 matrix change) directly from NY State Open Data (the same national Powerball numbers). A readout shows the current draw count and date range at a glance.
• Analysis — frequency, gap/overdue, positional, and pair analysis.
• Pick Numbers — weighted number generation and a "Draw & Rank" tool that scores random tickets against historical patterns.
• Predictability — a forensic/educational tab that cracks a known pseudo-random generator (LCG) to prove the method works, then runs the same cracker on real draws and shows plainly that it fails. The null result is the point: real draws carry no recoverable generator.
• Best Candidate Search — a time-boxed, cancellable research search over many model families, encodings, and ensembles. It uses a chronological train/validation/holdout split (the holdout is scored exactly once), ranks by held-out log-loss versus the random baseline, and always returns the strongest candidate found — with an honest verdict and an underpowered-data warning. Reports save to Documents/Lottery Analyzer/reports.

Requirements

• Python 3.10+
• PyQt6, selenium (see requirements.txt)
• Microsoft Edge WebDriver (only needed for the live KY scrape fallback; the NY Open Data history fetch and all analysis need no browser)

pip install -r requirements.txt
python LotteryApp.py

Self-test

The core algorithms run headlessly:

python LotteryApp.py --selftest      # 8 checks in LotteryApp.py
python selftest_search.py            # 10 checks for the search backend

Building the Windows EXE

Locally: run build.bat (installs deps, runs the self-test, then PyInstaller). Output: dist/LotteryAnalyzerPro.exe.

On GitHub (automatic): push a version tag and the included workflow (.github/workflows/build.yml) builds the exe on windows-latest and attaches it to the matching release:

git tag v2.5.3
git push origin v2.5.3

Module layout

File	Purpose
LotteryApp.py	GUI, data fetching, DB, all tabs
scoring.py	log-loss scoring rules + random baselines
validation.py	chronological split + held-out evaluation
historical_models.py	non-PRNG candidate models + ensemble
prng_tests.py	LCG / LFSR solvers + PRNG search candidates
candidate_search.py	the time-boxed evolutionary search
report_builder.py	plain-English report + honesty guards
selftest_search.py	backend self-tests

Note on results

The honest, expected result of the Best Candidate Search on real draws is "best candidate found, but no reliable predictive signal." That is not a bug — it is the correct finding for a fair lottery. Read the verdict, not the hope.