{
  "id": 11674440,
  "title": "I Run a Lottery Model That Publishes Its Own Failures — 51 Draws of Mark Six Data",
  "url": "https://urgent.news/2026/10/03/i-run-a-lottery-model-that-publishes-its-own-failures-51-draws-of",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-03T12:00:13.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/mystiqueracing/i-run-a-lottery-model-that-publishes-its-own-failures-51-draws-of-mark-six-data-23ia"
  },
  "original_language": "en",
  "account": "This reporter from the wire details a unique lottery model that openly publishes its failures. The model generates three sets of numbers: one based on statistical frequency, recency, and omission-gap weighting, another using physics-based priors like ball-machine simulation and positional bias, and finally, a control group with a deliberately non-statistical baseline based on numerology rules. If the control group ever matches the statistical models' hit rate, it indicates the statistical models lack any signal. So far, the control group has not matched the statistical models, which is keeping the other two honest.\n\nThe last draw, number 26104, had winning numbers [4, 28, 31, 44, 47, 48] plus an extra number 19. The statistical model achieved 0/6 hits, the physics model got 1/6 hit, and the control group had 0/6 hits, resulting in a total of 1 hit versus the expected 2.2 for random picks of 18 numbers out of 49. This performance is below random, and the reporter still publishes this information. The model's self-calibration process is also discussed, where every signal the model emits is scored against actual results, and the weights automatically update. Over the last 24 parameter updates, the 'late steam' weight oscillated between 0.947 and 0.992, showing the model's ability to correct itself without human intervention.\n\nThe next draw, number 26105, with picks published in advance, is scheduled to be released tomorrow. The consensus across models predicted number 48 and 7 to appear in the winning numbers. The reporter emphasizes that honesty is the actual product, highlighting the importance of providing full prediction history, post-mortems, public health checks of the pipeline, and calibration stats that include losing streaks. This approach distinguishes the project from a market for lemons, where everyone claims high accuracy rates without any auditing. The model is built on Cloudflare Workers + Pages + KV/D1, with a cron snapshot every 2 minutes on race days, and an EMA drift loop in Python. The reporter is available to answer questions about the model's architecture.",
  "summary": "Most lottery prediction sites show you their wins. Mine shows you the misses — every single one, on a public status page. Here's why I built it that way, and what 51 draws of Hong Kong Mark Six data actually look like under an honest model. The setup I maintain three independent number generators for Mark Six (6 numbers out of 49, plus an extra number): Science — frequency, recency, and…",
  "key_points": [
    "Model publishes failures openly, including 0/6 hits from statistical model",
    "Physics-based model achieved 1/6 hit, control group 0/6 hits",
    "Model self-calibrates, weights oscillated between 0.947 and 0.992"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}