We ship the things that shouldn't be possible — in public, with receipts.
The AI revolution already happened. Code is free now; attention and trust are what's scarce. Anyone can generate a repo. Almost nobody can prove their software actually works. That proof is our product.
- Go-Hybrid — a hybrid
language and toolchain that compiles a Rust-checked, Go-shaped language
(
hy) to readable Go. The loan checker rejects data races, use-after-free-shaped bugs, and unbuildable emissions beforego buildever runs. 127-fixture adversarial corpus, differential harness, arm64 bridge, Docker images atghcr.io/limitlessprojects/go-hybrid, published verdict windows. - Autoland — contributor fixes are verified end-to-end (repro on main → patched repro with the exact expected error → full battery → corpus → differential loop) and landed by the machine, with evidence, while the humans sleep. The queue drains itself; every close gets a written reply; nothing merges without proof.
- ClockBench — the living, neutral, leakage-aware benchmark for epigenetic aging clocks: same clock, different packages, different ages — we measure it, with signed artifacts and a public leakage ledger.
- Explainer films as code — deterministic, rendered-from-source motion-graphics films about our own systems (followbot-explainer).
Every PR gets a written reply — merged or not, no silent closes. Real verification, not vibes: your reported bug is reproduced on main, your fix is probed against the repro and over-reach controls, and the full battery must stay green. Same-fix duplicates are credit-closed with the evidence spelled out and your work named.
Bounties pay real money (BOUNTY.md) — including the Wildcard Track: contrarian, high-risk, "impossible" ideas. Beautiful failures are published as learnings. And paid.
Projects, plural. Limits, none. We ship weekly, publish failures, and let receipts do the marketing.
If that sounds like you: star Go-Hybrid, take a good first issue, or fire a wildcard.