{
  "id": 683748,
  "title": "Question about Leaderboard precision and handling of tied scores",
  "url": "/competitions/deep-past-initiative-machine-translation/discussion/683748",
  "author_name": "",
  "post_date": "2026-03-22T00:56:55.675782900Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n<p>I noticed that the Public Leaderboard displays scores truncated to a certain number of decimal places, which has resulted in a large number of participants sharing the exact same score</p>\n<p>I'm curious about how Kaggle handles this internally to ensure fair ranking.\nIs there evidence or official documentation stating that Kaggle calculates the evaluation metric to a much higher precision internally than what is displayed?</p>\n<p>Thanks in advance for any insights! Good luck to everyone</p>",
  "messages": [
    {
      "id": "3426155",
      "postDate": "03/22/2026 00:56:55",
      "content": "<p>Hi everyone,</p>\n<p>I noticed that the Public Leaderboard displays scores truncated to a certain number of decimal places, which has resulted in a large number of participants sharing the exact same score</p>\n<p>I'm curious about how Kaggle handles this internally to ensure fair ranking.\nIs there evidence or official documentation stating that Kaggle calculates the evaluation metric to a much higher precision internally than what is displayed?</p>\n<p>Thanks in advance for any insights! Good luck to everyone</p>",
      "rawMarkdown": "Hi everyone,\n\nI noticed that the Public Leaderboard displays scores truncated to a certain number of decimal places, which has resulted in a large number of participants sharing the exact same score\n\nI'm curious about how Kaggle handles this internally to ensure fair ranking.\nIs there evidence or official documentation stating that Kaggle calculates the evaluation metric to a much higher precision internally than what is displayed?\n\nThanks in advance for any insights! Good luck to everyone",
      "votes": null
    },
    {
      "id": "3426229",
      "postDate": "03/22/2026 06:13:13",
      "content": "<p>When the competition is over, the scores will be displayed with more precision. </p>",
      "rawMarkdown": "When the competition is over, the scores will be displayed with more precision.",
      "votes": null
    },
    {
      "id": "3426246",
      "postDate": "03/22/2026 07:56:22",
      "content": "<p>Full precision is used for calculating a LB score. If several teams share the same score (e.g. submitted the same public notebook), then a tiebraker is decided by the earliest submission time</p>",
      "rawMarkdown": "Full precision is used for calculating a LB score. If several teams share the same score (e.g. submitted the same public notebook), then a tiebraker is decided by the earliest submission time",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3426229,
      "author_name": "tennogh",
      "author_url": "",
      "post_date": "03/22/2026 06:13:13",
      "content": "<p>When the competition is over, the scores will be displayed with more precision. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3426246,
      "author_name": "samson8",
      "author_url": "",
      "post_date": "03/22/2026 07:56:22",
      "content": "<p>Full precision is used for calculating a LB score. If several teams share the same score (e.g. submitted the same public notebook), then a tiebraker is decided by the earliest submission time</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3426155": "Hi everyone,\n\nI noticed that the Public Leaderboard displays scores truncated to a certain number of decimal places, which has resulted in a large number of participants sharing the exact same score\n\nI'm curious about how Kaggle handles this internally to ensure fair ranking.\nIs there evidence or official documentation stating that Kaggle calculates the evaluation metric to a much higher precision internally than what is displayed?\n\nThanks in advance for any insights! Good luck to everyone",
    "3426229": "When the competition is over, the scores will be displayed with more precision.",
    "3426246": "Full precision is used for calculating a LB score. If several teams share the same score (e.g. submitted the same public notebook), then a tiebraker is decided by the earliest submission time"
  },
  "source": "meta"
}