Skip to content
DeepForgeby svx
15 categories · real Python execution · local-first

Learn by building.
Trust what you can test.

DeepForge is a local-first ML practice platform and an experiment in reliable, evidence-based programming education. Solve 5,730+ problems with real in-browser Python execution — no sklearn, no torch, no shortcuts.

Start practicingBrowse pathsNo account. Progress saves locally.

Practice is the product. Evidence is the research agenda.

Reliable and evidence-based programming education in the age of generative AI.

The work is grounded in three practical questions — not claims of a finished theory.

Read the publications

Test adequacy

Can a test suite distinguish correct programs from plausible-but-wrong ones?

Silent Bug Hunt pairs a reference function with a test-passing one-line divergence, then asks you to find the input the suite missed.

Try Silent Bug Hunt

Behavioral evidence

Can an edit reveal a meaningful behavioral change without becoming a grade?

Behavioral Delta Ledger reports only whether hidden checks changed after an edit. It never affects progress, review, or certificates.

Explore the ledger

Auditable feedback

Can learning guidance show the evidence behind it and respond when that evidence is refuted?

Warrant Lab makes each derived claim inspectable: see its checks, challenge it, and audit the resulting bookkeeping.

Open Warrant Lab
5,730+
problems
15
categories
33
learning paths
0
dependencies
∞
executions

Categories

Fifteen foundations. Pick one and start at the top.

Explore

Everything in DeepForge, in one place. Pick a card to open it.

Learn

Practice

Compete

You

Community

Engineering — how we build DeepForgeRead the blog →