← ClaudeAtlas

kiyaslisted

Disciplined ideation and analogical inference for research that is stuck. Use whenever the user wants to generate ideas, break through a blocked problem, find analogies or cross-domain relationships, reframe a question, brainstorm hypotheses, or explore "what else could explain this / what am I not seeing." Also use to open new experimental directions from a dead end, connect two unrelated fields, or turn a vague intuition into a testable hypothesis. Triggers include "fikir üret", "tıkandım", "analoji kur", "başka nasıl bakabilirim", "yeni yön", "bağlantı kur", "brainstorm", "ideate", "reframe", "what am I missing". The generative upstream partner of Mizan (mizan audits/refutes; kiyas generates candidates shaped for that audit). Every idea it produces ships ready to enter a Mizan registry as a preregistered [S]/[H] hypothesis.
XINMurat/Kiyas · ★ 1 · AI & Automation · score 62
Install: claude install-skill XINMurat/Kiyas
# Kıyas — Disciplined Ideation and Analogical Inference > **[Revised 2026-07-22 — A1–A4]** A 4-scenario test → 3 seeds → Mizan audit → > prior-art lit-check cycle exposed and fixed three preregistration-hygiene > gaps: (A1) prior-art declaration is mandatory and gates `[H-aday]`; > (A2) capacity-confound control arm; (A3) instrument-specific calibration; > (A4) trace base-rate prior (AD6). The analogical core did not change; what > changed is the hygiene a candidate passes before entering Mizan. > **[Revised 2026-08-20 — G10/G11]** The last two findings of > `PROSE-SCHEMA-AUDIT.md`. G10 came from the sibling Mizan validator, which > had caught the arbiter self-contradiction since R8 shipped — the two repos > had each solved something the other had not, and comparing them was worth > more than either audit alone. > **[Revised 2026-08-19 — G8/G9 + warnings]** A systematic pass over this > file against the schema (`PROSE-SCHEMA-AUDIT.md`) found two places where the > data contract permitted what the prose forbids outright: an `[H-aday]` with > no cheapest refutation (constraint 3), and an `[H-aday]` whose prior art was > never searched (constraint 4 — G3 only ever covered the narrower > superiority-claim gate). Both now block. The validator also gained a > non-blocking warning channel, because a single blocking channel teaches > authors to write batches that avoid triggering rules — G6's own lesson, > applied to the tool. > **[Revised 2026-08-19 — G7]** The discard list beca