DEDUPED Hot AI repos, turned into use cases Get 5 a day

How to check if a harness score is comparable before citing it with Eval

Confirm the egress policy and comparability flag before you put a new score next to published ones.

Who it is forA researcher or engineer about to publish or compare a new harness score against the Kimi K3 baselines.

The problemPublished baselines did not record their applied network allowlist, so a new score can look comparable when it is not.

The outcomeYou know whether your run can be ranked, and what control run you still need to establish comparability.

What you need first

  • A Runta account, with the runta CLI installed and authenticated via runta login
  • jq, Node.js 18 or newer, and Python 3.9 or newer
  • A provider API key for Kimi K3 (FIREWORKS_API_KEY, MOONSHOT_API_KEY, OPENROUTER_API_KEY, or TOGETHER_API_KEY)
  • A harness repository you can build on a clean Linux box, plus an install script

Steps

  1. Open results/eval-data.json and metadata/harness-versions.json to see the published aggregate results and the harness versions used for the run.
  2. In your own run folder, open run.json and read egress_policy for its mode, scope, and exact allowed_hosts.
  3. Check that methodology_comparable is set to false for any new run that has not been compared against a control run.
  4. Run a control harness under the same policy and environment, then compare the two runs before claiming comparability.

Prompt for your agent

In the frontier-harness-eval/eval repo, read metadata/harness-versions.json and results/eval-data.json, then open the run.json in <run folder>. Report the egress_policy mode and whether methodology_comparable is true or false. Verify that the flag is false unless a matched control run is recorded.

Caveat. The README says the published baselines did not record their applied allowlist, so no new score can be ranked against them without a matched control run.

Share: X LinkedIn Hacker News Email Copy link

Disclaimer

Deduped lists open-source repositories that belong to other people. We do not own, maintain, audit, or vouch for any of them, and a repo appearing here is not an endorsement or a safety check. The use cases, steps, and prompts are written by AI and may be incomplete, outdated, or just wrong. Where a use case is marked "Seen in the wild", we link to public posts where someone described something similar, but we have not verified those posts and the use case itself is still our AI's own write-up. Use all of this at your own risk. Only run code, commands, or agents on systems and accounts you are authorized to use, and read the code and every command before you run it, especially anything that touches credentials, files, or the network. Nothing here is professional, security, legal, or financial advice. All trademarks and project names belong to their respective owners, and their use here does not imply affiliation. Content is provided as is, without warranties of any kind. Found a mistake, a safety problem, or something you want removed? Email [email protected] and we will look at it promptly. Full disclaimer page