Wordle Solver: Heuristic + Trained Strategy

Completed

Summary

Developed a Python Wordle solver with both a heuristic baseline and a lightweight trained strategy model. The project includes an interactive Streamlit UI and a benchmark pipeline for evaluating solve performance at scale.

• Tag: Hobby/Other
• Project Type: Personal Project
• Repository: GitHub
• Tech Stack: Python, Streamlit

Example feedback pattern from the interactive solver interface.

Approach

1. Heuristic Solver

The baseline solver uses letter-frequency heuristics and candidate elimination:

• Candidate words are filtered by feedback
• Early turns prioritize information gain with letter-frequency-biased guesses
• Later turns exploit narrowed candidate sets

2. Trained Strategy Solver

A lightweight linear model scores candidate guesses using handcrafted features:

• Unique letter frequency score
• Positional letter frequency score

A random-search training routine optimizes the feature weights to improve solve rate and average number of turns.

3. Benchmarking

A benchmark runner simulates full games automatically and reports solve rate, average turns, and hard/unsolved words with the current strategy.

Results

Latest benchmark summary (2026-03-27):

• Words tested: 2315
• Solved within 6 turns: 2301
• Failed within 6 turns: 14
• Solve rate: 99.40%
• Average turns (solved words): 3.609

Tools and Libraries

• Python for solver logic, simulation, and model training
• Streamlit for interactive web UI
• JSON-based model artifact for trained strategy weights and metrics