---
title: "Our first technical report"
url: "https://ankush.one/timeline/2026.first-tech-report/"
description: "Co-authored Construct Technical Report 2026-01 with Nischal Naik, a study of swapping TypeSafe’s Jev for Cloudflare’s Clef decision models in a production agent."
author: "Ankush Singh"
date: 2026-10-05
lastmod: 2026-10-12
tags: ["construct","ai-agents","machine-learning"]
links:
  - {"archived":false,"label":"Read the PDF","url":"https://construct.computer/blog/clef-vs-jev-benchmark/compatible-apis-incompatible-thresholds.pdf"}
---

# Our first technical report

Links:

- [Read the PDF](https://construct.computer/blog/clef-vs-jev-benchmark/compatible-apis-incompatible-thresholds.pdf)

"Compatible APIs, Incompatible Thresholds" is Construct Technical Report 2026-01, with my name on the cover next to Nischal's.

When Cloudflare released Clef, we pointed the decision specs we had written for Jev at it without changing a line and ran 1,053 calls against 117 labelled cases. Clef came out a little ahead, though not by a statistically significant margin, and gave the identical answer on every repeated call. The bigger lesson was that swapping decision models is a thresholds and reproducibility problem, not just an API problem. The [blog post](https://ankush.one/blogs/clef-vs-jev-benchmark/) has the short version.