{
  "id": 671805,
  "title": "Official Update: Online Round Scoring Rubric Revised",
  "url": "/competitions/dl-sprint-4-0-bengali-long-form-speech-recognition/discussion/671805",
  "author_name": "Shadman Tabib",
  "post_date": "2026-02-04T07:26:41.979000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>To ensure a more balanced and fair evaluation, we've updated the Online Evaluation scoring weights:</p>\n<p><strong>Public Leaderboard Score will now contribute 15% (previously 20%)</strong></p>\n<p><strong>Private Leaderboard Score will now contribute 35% (previously 30%)</strong></p>\n<p><strong>The Hidden Test Set Score (50%) remains unchanged.</strong></p>\n<p>This change is designed to discourage leaderboard overfitting and reward solutions that generalize better across diverse datasets.</p>\n<p>We strongly encourage all participants to submit their most robust and consistent models. Focus on solutions that perform well not just on the leaderboard but across the full evaluation spectrum.</p>\n<p>Happy Deep Learning, and best of luck to all competitors.</p>\n<p>Find the rulebook in this link (same as before):\n<a href=\"https://tinyurl.com/DLSprint4Rulebook\" target=\"_blank\">https://tinyurl.com/DLSprint4Rulebook</a></p>",
  "messages": [
    {
      "id": 3401632,
      "postDate": "2026-02-04T07:26:41.980Z",
      "content": "<p>To ensure a more balanced and fair evaluation, we've updated the Online Evaluation scoring weights:</p>\n<p><strong>Public Leaderboard Score will now contribute 15% (previously 20%)</strong></p>\n<p><strong>Private Leaderboard Score will now contribute 35% (previously 30%)</strong></p>\n<p><strong>The Hidden Test Set Score (50%) remains unchanged.</strong></p>\n<p>This change is designed to discourage leaderboard overfitting and reward solutions that generalize better across diverse datasets.</p>\n<p>We strongly encourage all participants to submit their most robust and consistent models. Focus on solutions that perform well not just on the leaderboard but across the full evaluation spectrum.</p>\n<p>Happy Deep Learning, and best of luck to all competitors.</p>\n<p>Find the rulebook in this link (same as before):\n<a href=\"https://tinyurl.com/DLSprint4Rulebook\" target=\"_blank\">https://tinyurl.com/DLSprint4Rulebook</a></p>",
      "rawMarkdown": "To ensure a more balanced and fair evaluation, we've updated the Online Evaluation scoring weights:\n\n**Public Leaderboard Score will now contribute 15% (previously 20%)**\n\n**Private Leaderboard Score will now contribute 35% (previously 30%)**\n\n**The Hidden Test Set Score (50%) remains unchanged.**\n\nThis change is designed to discourage leaderboard overfitting and reward solutions that generalize better across diverse datasets.\n\nWe strongly encourage all participants to submit their most robust and consistent models. Focus on solutions that perform well not just on the leaderboard but across the full evaluation spectrum.\n\nHappy Deep Learning, and best of luck to all competitors.\n\nFind the rulebook in this link (same as before):\nhttps://tinyurl.com/DLSprint4Rulebook",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3401632": "To ensure a more balanced and fair evaluation, we've updated the Online Evaluation scoring weights:\n\n**Public Leaderboard Score will now contribute 15% (previously 20%)**\n\n**Private Leaderboard Score will now contribute 35% (previously 30%)**\n\n**The Hidden Test Set Score (50%) remains unchanged.**\n\nThis change is designed to discourage leaderboard overfitting and reward solutions that generalize better across diverse datasets.\n\nWe strongly encourage all participants to submit their most robust and consistent models. Focus on solutions that perform well not just on the leaderboard but across the full evaluation spectrum.\n\nHappy Deep Learning, and best of luck to all competitors.\n\nFind the rulebook in this link (same as before):\nhttps://tinyurl.com/DLSprint4Rulebook"
  }
}