Real hardware. Real runs.
Published results.
Everything here is measured on a self-hosted home lab and published the honest way: genuine test data, raw outputs linked from every result, and a publish pipeline that scrubs internal addresses before anything reaches the internet.
Sections
Local LLM Benchmarks
Small language models run on consumer GPUs — the same nine tests for every model, code executed and verified, games rendered headlessly, raw outputs published verbatim.
Retail Price Watch
Daily monitoring of Australian grocery and hardware prices across Woolworths, Bunnings and BigW — every observation kept, per-100g normalised, alerting only on genuine new lows.
How publishing works
Genuine runs only. Fixed prompts, budgets and checks in one registry; every test writes its own atomic result files. Nothing on this site is hand-typed.
Private stays private. A scrubber replaces internal hosts with public hardware labels, then a leak gate fails the build if any internal IP, hostname or port survives.
Automated end to end. CI rebuilds every page from the raw data and ships it to GitHub Pages — the numbers you see were built by the pipeline, not edited by hand.