Copies of the experiment records from the Barg Labs alfred repository at commit c1c7bf7f591d2af13ee565efa9c303b06e4dfd1b, mirrored 2026-10-01T15:43:09Z. The preregistration was signed at alfred 4c3dee91; the erratum at 9f70185e.
Read the result's first paragraph before any number: the primary run kept the package's default leading-token truncation for inputs longer than Laya's 512-token window, which the preregistration said would be refused. The erratum records it; the page at cejel.dev/experiments/laya-2026-09-29/ leads with it.
Layout: the preregistration, erratum and run approvals at the root; result/ holds the unedited result (markdown and aggregate JSON); harness/ holds the Laya adapter, runner, secondary arms and test. The harness runs inside the Jev harness published at https://github.com/BargLabs/jev-judge-calibration, whose corpus this run used unedited (corpus digest in the result).
Not published under the disclosure boundary: call logs, response payloads, corpus membership beyond the Jev record, labels, reviewer material.