cot-quilt graph viewer — run 20261001T195410Z-503b10b9
prompt: When should a cell in a quilted system ask the JEV oracle (an external judge) versus compute the answer locally from its neighbors? Design the decision rule, name its failure mode, and say how the ledger catches it.
seeds: 36807 / 17141 / 55843mothquantum 16-bit × 33 CoT samples5 clusters1 divergent routes5 critique cells6 cluster edges (p=mean typesafe dependency)
6/10v4-pro self-judged thoroughness The decomposition captures the high-level workflow but omits specific prompt requirements like naming the failure mode and detailing the ledger's enforcement mechanism.
Explicit naming of the failure mode
Specific ledger enforcement mechanism (e.g., fraud proofs)
Cost-benefit threshold calculation logic
Incentive alignment for honest reporting
Integration of Silent Cartel risk into main flow
● cluster (size=occurrence, hue=load) — edge (width=count, opacity=p) □ divergent route ◆ critique cell — hover for full text
cell signature — in: {prompt, n_seeds, lenses} · out: {graph_v1, graph_v2, judge, receipts} ·
clusters → value cells · edges → links with p as dial weight · judge.gaps → question-cells · critique nodes → next generation
repo: github.com/SuperInstance/cot-quilt · erised playground: erised-mirror.pages.dev