Skip to content

Releases

0.5.0: one-command runner and two public benchmarks

0.5.0: one-command runner and two public benchmarks Latest
Latest

Choose a tag to compare

View all tags
@McSlaine McSlaine released this 11 Nov 03:05
v0.5.0
b10eaae

A low-cost model does the coding. A frontier model checks the exact result. You get frontier-level work at a fraction of the price, with a receipt.

Results

66 real feature tasks, graded by hidden tests Tasks solved Cost per task
Heavy model alone (gpt-6-astra) 56 $0.721
Frontier model alone (gpt-6.1-sol) 56 $0.123
KLP Standard (deepseek-flash builds, gpt-6.1-sol reviews) 59 $0.046
30 pairs of conflicting features Pairs delivered Cost per pair
Frontier model alone builds both 22 $0.196
Two KLP workers, joined by Semio 28 $0.103

Check every figure offline, no API keys: python3 -m bench.verify

New in 0.5.0

  • One-command runner: klp run "<task>" --check "<command>" builds in a sandbox, runs your checks, has an independent critic review the exact diff and returns a patch and a receipt.
  • Findings settled by execution: every blocking finding carries a probe test. A failing probe confirms it; a passing probe discards it. A confirmed finding stays open until its probe passes.
  • Critic on a subscription: the critic can be an agent CLI you already pay for.
  • Two public benchmarks with receipts, call ledgers and a harness to rerun them: BENCHMARKS.md.
  • Pages for AI agents: AGENTS.md, evidence.json and llms.txt.

Install

pipx install git+https://github.com/KrystalUnity/krystal-loop-protocol

Licence

Free for individuals and small businesses (PolyForm Small Business 1.0.0). Anyone can evaluate it and rerun the benchmarks. Larger organisations need a commercial licence: sales@krystalunity.com. Documentation is CC-BY-4.0. Revisions up to d225053 remain under Apache-2.0.

Assets 2
Loading