Hello, I would like to ask for clarification about the use of proprietary models in this competition.\nAfter reading the competition rules and Kaggle’s foundational rules, my understanding is that proprietary tools or models are not automatically prohibited unless they violate a specific rule, but their use may still be limited by a reasonableness standard related to cost, accessibility, licensing, and compliance.
\nCould you please explain in detail where the boundary is between acceptable and prohibited use of proprietary models for this competition?\nIn particular, I would like to confirm whether it is allowed to use a closed-source or commercial model for feature extraction, pseudo-label generation, distillation, or as part of the training or inference pipeline, as long as I do not violate any data, code-sharing, or intellectual-property rules.
\nI would also appreciate clarification on whether there are any restrictions regarding paid API-based models, privately licensed pretrained weights, externally hosted inference services, or proprietary components that cannot be fully transferred or reproduced by the organizers.
\nIf a proprietary model is used only as an auxiliary tool during development, but the final submitted solution is reproducible and compliant, would that still be acceptable?\nFinally, could you provide concrete examples of what would be considered clearly allowed, borderline, and clearly disallowed use of proprietary models in this competition, especially under the “reasonable” use standard described in the rules?
", "messages": [ { "id": "3434717", "postDate": "04/03/2026 06:40:50", "content": "Hello, I would like to ask for clarification about the use of proprietary models in this competition.\nAfter reading the competition rules and Kaggle’s foundational rules, my understanding is that proprietary tools or models are not automatically prohibited unless they violate a specific rule, but their use may still be limited by a reasonableness standard related to cost, accessibility, licensing, and compliance.
\nCould you please explain in detail where the boundary is between acceptable and prohibited use of proprietary models for this competition?\nIn particular, I would like to confirm whether it is allowed to use a closed-source or commercial model for feature extraction, pseudo-label generation, distillation, or as part of the training or inference pipeline, as long as I do not violate any data, code-sharing, or intellectual-property rules.
\nI would also appreciate clarification on whether there are any restrictions regarding paid API-based models, privately licensed pretrained weights, externally hosted inference services, or proprietary components that cannot be fully transferred or reproduced by the organizers.
\nIf a proprietary model is used only as an auxiliary tool during development, but the final submitted solution is reproducible and compliant, would that still be acceptable?\nFinally, could you provide concrete examples of what would be considered clearly allowed, borderline, and clearly disallowed use of proprietary models in this competition, especially under the “reasonable” use standard described in the rules?
", "rawMarkdown": "Hello, I would like to ask for clarification about the use of proprietary models in this competition.\nAfter reading the competition rules and Kaggle’s foundational rules, my understanding is that proprietary tools or models are not automatically prohibited unless they violate a specific rule, but their use may still be limited by a reasonableness standard related to cost, accessibility, licensing, and compliance.\n\nCould you please explain in detail where the boundary is between acceptable and prohibited use of proprietary models for this competition?\nIn particular, I would like to confirm whether it is allowed to use a closed-source or commercial model for feature extraction, pseudo-label generation, distillation, or as part of the training or inference pipeline, as long as I do not violate any data, code-sharing, or intellectual-property rules.\n\nI would also appreciate clarification on whether there are any restrictions regarding paid API-based models, privately licensed pretrained weights, externally hosted inference services, or proprietary components that cannot be fully transferred or reproduced by the organizers.\n\nIf a proprietary model is used only as an auxiliary tool during development, but the final submitted solution is reproducible and compliant, would that still be acceptable?\nFinally, could you provide concrete examples of what would be considered clearly allowed, borderline, and clearly disallowed use of proprietary models in this competition, especially under the “reasonable” use standard described in the rules?", "votes": null }, { "id": "3435242", "postDate": "04/03/2026 21:33:49", "content": "Dear @mrpc2003 ,
\nUse of proprietary or closed-source models is as you said not automatically prohibited. IMHO, The issue is not whether a component is proprietary by itself, but whether its use remains reasonable in practice, and compatible with the competition’s reproducibility and code-delivery requirements.
\nSo, if a proprietary model is used only as an auxiliary development tool, and the final submitted solution is itself compliant, reproducible, and properly described, we would not view that alone as a problem. However, if the proprietary component is essential to the submitted method in a way that makes the final system non-transferable or non-reproducible, that may not be acceptable.
\nIn other words. if a proprietary model is used in a way that creates a submission that cannot be meaningfully explained, transferred, reproduced, or validated by the organizers if needed, that would be problematic. By contrast, using proprietary components as part of the research path or development process is not necessarily an issue, provided the final submission remains compliant and the overall methodology is presented clearly and honestly.
\nSo in practice, we would focus less on whether a component is “open” or “closed,” and more on whether there is a clear and reasonable path from the method described in the paper to a competition-compliant final solution. We care about the path and the paper, with a clear story about what was used for development, what was used in the final system, and what can be provided if requested under the competition rules.
\nHope it helps.
\nBest regards,\nLukas
", "rawMarkdown": "Dear @mrpc2003 ,\n\nUse of proprietary or closed-source models is as you said not automatically prohibited. IMHO, The issue is not whether a component is proprietary by itself, but whether its use remains reasonable in practice, and compatible with the competition’s reproducibility and code-delivery requirements. \n\nSo, if a proprietary model is used only as an auxiliary development tool, and the final submitted solution is itself compliant, reproducible, and properly described, we would not view that alone as a problem. However, if the proprietary component is essential to the submitted method in a way that makes the final system non-transferable or non-reproducible, that may not be acceptable.\n\nIn other words. if a proprietary model is used in a way that creates a submission that cannot be meaningfully explained, transferred, reproduced, or validated by the organizers if needed, that would be problematic. By contrast, using proprietary components as part of the research path or development process is not necessarily an issue, provided the final submission remains compliant and the overall methodology is presented clearly and honestly.\n\nSo in practice, we would focus less on whether a component is “open” or “closed,” and more on whether there is a clear and reasonable path from the method described in the paper to a competition-compliant final solution. We care about the path and the paper, with a clear story about what was used for development, what was used in the final system, and what can be provided if requested under the competition rules.\n\nHope it helps.\n\nBest regards,\nLukas", "votes": null } ], "comments": [ { "id": 3435242, "author_name": "picekl", "author_url": "", "post_date": "04/03/2026 21:33:49", "content": "Dear @mrpc2003 ,
\nUse of proprietary or closed-source models is as you said not automatically prohibited. IMHO, The issue is not whether a component is proprietary by itself, but whether its use remains reasonable in practice, and compatible with the competition’s reproducibility and code-delivery requirements.
\nSo, if a proprietary model is used only as an auxiliary development tool, and the final submitted solution is itself compliant, reproducible, and properly described, we would not view that alone as a problem. However, if the proprietary component is essential to the submitted method in a way that makes the final system non-transferable or non-reproducible, that may not be acceptable.
\nIn other words. if a proprietary model is used in a way that creates a submission that cannot be meaningfully explained, transferred, reproduced, or validated by the organizers if needed, that would be problematic. By contrast, using proprietary components as part of the research path or development process is not necessarily an issue, provided the final submission remains compliant and the overall methodology is presented clearly and honestly.
\nSo in practice, we would focus less on whether a component is “open” or “closed,” and more on whether there is a clear and reasonable path from the method described in the paper to a competition-compliant final solution. We care about the path and the paper, with a clear story about what was used for development, what was used in the final system, and what can be provided if requested under the competition rules.
\nHope it helps.
\nBest regards,\nLukas
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3434717": "Hello, I would like to ask for clarification about the use of proprietary models in this competition.\nAfter reading the competition rules and Kaggle’s foundational rules, my understanding is that proprietary tools or models are not automatically prohibited unless they violate a specific rule, but their use may still be limited by a reasonableness standard related to cost, accessibility, licensing, and compliance.\n\nCould you please explain in detail where the boundary is between acceptable and prohibited use of proprietary models for this competition?\nIn particular, I would like to confirm whether it is allowed to use a closed-source or commercial model for feature extraction, pseudo-label generation, distillation, or as part of the training or inference pipeline, as long as I do not violate any data, code-sharing, or intellectual-property rules.\n\nI would also appreciate clarification on whether there are any restrictions regarding paid API-based models, privately licensed pretrained weights, externally hosted inference services, or proprietary components that cannot be fully transferred or reproduced by the organizers.\n\nIf a proprietary model is used only as an auxiliary tool during development, but the final submitted solution is reproducible and compliant, would that still be acceptable?\nFinally, could you provide concrete examples of what would be considered clearly allowed, borderline, and clearly disallowed use of proprietary models in this competition, especially under the “reasonable” use standard described in the rules?", "3435242": "Dear @mrpc2003 ,\n\nUse of proprietary or closed-source models is as you said not automatically prohibited. IMHO, The issue is not whether a component is proprietary by itself, but whether its use remains reasonable in practice, and compatible with the competition’s reproducibility and code-delivery requirements. \n\nSo, if a proprietary model is used only as an auxiliary development tool, and the final submitted solution is itself compliant, reproducible, and properly described, we would not view that alone as a problem. However, if the proprietary component is essential to the submitted method in a way that makes the final system non-transferable or non-reproducible, that may not be acceptable.\n\nIn other words. if a proprietary model is used in a way that creates a submission that cannot be meaningfully explained, transferred, reproduced, or validated by the organizers if needed, that would be problematic. By contrast, using proprietary components as part of the research path or development process is not necessarily an issue, provided the final submission remains compliant and the overall methodology is presented clearly and honestly.\n\nSo in practice, we would focus less on whether a component is “open” or “closed,” and more on whether there is a clear and reasonable path from the method described in the paper to a competition-compliant final solution. We care about the path and the paper, with a clear story about what was used for development, what was used in the final system, and what can be provided if requested under the competition rules.\n\nHope it helps.\n\nBest regards,\nLukas" }, "source": "meta" }