A Claude skill that turns an experimental-science discipline into a portable tool for evaluating any claim set and maintaining living hypothesis registries. Bilingual TR/EN.
Evidence-tiered auditing and preregistration methodology, packaged as a Claude skill. Kanıt-katmanlı denetim ve önkayıt metodolojisi — bir Claude skill’i olarak paketlenmiş.
Repository · Latest release · Kıyas · İskele · ux-mizan · the family · aile
Mizan (Turkish/Arabic: the scale) turns a rigorous experimental-science discipline into a portable tool for evaluating any claim set — an AI-generated summary, a project report, a year-in-review, a codebase, a PRD — and for maintaining living hypothesis registries.
Its core commitments: every claim gets an evidence tier, with no untagged assertions. Thresholds are locked before results, or the HARKing risk is declared rather than silently absorbed. Every hypothesis carries a refutation condition — a claim that cannot fail is not audited, it is decorated. Refuted entries are never deleted. A surprising positive waits for its symmetric control before it gets a headline. And three confirming anecdotes are selection bias; a scored prediction record is evidence.
Worked examples (in the repository): a registry that passes every rule, the portability runs that test whether the discipline survives a hostile setup, and auditing a document whose source you cannot reach — where the failure mode is not refusal but tier laundering.
The registry schema has an LLM-free validator (rules R1–R16, plus a non-blocking warning channel). It enforces the mechanical invariants and nothing else: semantic judgement stays with a human or a frontier model, which is rule R7 itself.
Mizan (terazi/ölçü), titiz bir deneysel-bilim disiplinini herhangi bir iddia setini değerlendirmek ve canlı hipotez registry’leri sürdürmek için taşınabilir bir araca dönüştürür.
Temel taahhütler: her iddia bir kanıt katmanı alır, etiketsiz iddia olmaz. Eşikler sonuç görülmeden önce kilitlenir; mümkün değilse HARKing riski sessizce soğurulmaz, açıkça beyan edilir. Her hipotez bir çürütme koşulu taşır — başarısız olamayan iddia denetlenmiş değil, süslenmiştir. Çürütülen girdiler silinmez. Sürpriz pozitif, manşetten önce simetrik kontrolünü bekler. Ve üç doğrulayıcı anekdot seçilim yanlılığıdır; puanlanmış bir tahmin kaydı kanıttır.
İskele kurar · Mizan tartar · Kıyas üretir. İskele turns a vague project intent into an executable delivery kit. Mizan audits claims and maintains preregistered registries. Kıyas generates the candidates Mizan weighs, and a Mizan registry’s refuted entries flow back to Kıyas as negative constraints — the loop closes.