Hi CM4AI Challenge Organizers, I have a question regarding the requirement that submitted container images must be publicly accessible. I noticed that several teams' submission images are publicly available on GitHub Container Registry (GHCR). These images can be discovered through a simple GitHub Packages search and downloaded by anyone, potentially exposing participants' model architectures, inference implementations, and other proprietary code. I'm concerned that this requirement may unintentionally expose unpublished research and methods, particularly for teams planning to publish their work after the competition. Could you please clarify: 1. Is public accessibility strictly required? Would it be possible to submit a private container image while granting the organizers permission to pull it for evaluation? 2. Are there alternative submission methods? For example, could participants provide temporary access credentials or share their images privately with the evaluation infrastructure? 3. If public images are mandatory, how long must they remain publicly accessible? Can participants make their images private immediately after evaluation is completed? I fully understand the need for reproducibility and reliable evaluation, but I hope there might be a way to satisfy these requirements without requiring participants to publicly disclose their implementations before publication. Thank you for your clarification and for organizing the challenge!

Created by CubeEscape
While all methods submitted to the challenge will need to be open sourced, we fully understand the need for preserving the security of a method leading up to the final evaluation. To deal with the complexity around dealing with a private image registry credentials we added a 'weights' file parameter as an alternate route for passing private elements for a prediction model. This artifact is created by you, added to the submission queue and presented to the runner code at evaluation time. On the [command line reference](https://www.synapse.org/Synapse:syn77349038/wiki/643952) see the instructions for the `--weights` argument. This file is presented to your WDL runner script at the `weights` parameter. In our [user example](https://github.com/cm4ai-anvil-dream-challenge/blast-proticelli) this is a zip file that contains Blast indices and the trained weights file. You can use this mechanism to pass your runner code any number of files, including other runner code. One way to do this is build a runner container with all required dependencies and script the takes the weights file, unzips it and then passes the arguments to the prediction script that was stored in the zip.

Concern About Public Container Images Exposing Participants' Model Code page is loading…