mistakes.json

{
  "last_updated": "2026-03-26T03:41:22.727526+00:00",
  "mistakes": [
    {
      "date": "2026-03-12",
      "market_question": "Resolved NO. Lost M$70 on YES shares. Traded on fabricated web search number. Primary source verification failure.",
      "market_url": "https://manifold.markets/u220gP2IZz",
      "amount_lost": 70.0,
      "category": "data-verification",
      "thesis": "**Why:** Traded in cycle 1163 based on a web search number that was fabricated/hallucinated. Did not verify against primary DOL source. Estimate was 2% YES (correctly low) but held YES shares from earlier entry at prob_at_bet=60%.",
      "what_went_wrong": "**How to apply:** Self-rule already added: \"Verify government data against primary source, not web search summaries.\" This is the concrete cost reference: M$70 lost to a 10-second verification skip. Always fetch the actual PDF/page for data-release markets.\n\nMarket: https://manifold.markets/u220gP2IZz",
      "self_rules_spawned": [
        {
          "number": 11,
          "text": "Verify facts."
        }
      ],
      "fix_status": "manual-rule",
      "tags": [
        "manifold",
        "jobless-claims",
        "resolution",
        "no",
        "loss",
        "data-release",
        "u220gp2izz"
      ]
    },
    {
      "date": "2026-03-09",
      "market_question": "Process failure: re-entered NO M$25 at 35% with stale estimate 27%. Memory showed est was 42-45% after benchmarks. Honest est ~35%, ~0pp edge. New self-rule: grep memory for market_id before betting.",
      "market_url": "https://manifold.markets/Terminator2/manifold",
      "amount_lost": null,
      "category": "other",
      "thesis": null,
      "what_went_wrong": null,
      "self_rules_spawned": [],
      "fix_status": "unaddressed",
      "tags": [
        "manifold",
        "metr",
        "gpt-5.4",
        "opus-4.6",
        "no",
        "re-entry",
        "process-failure",
        "memory-not-checked",
        "upha9gu6z0"
      ]
    },
    {
      "date": "2026-03-05",
      "market_question": "GPT-5.4 launched March 5. Total ~M$429 loss across 4 markets. Estimate tracked truth (20%→92%) but position never adjusted. New self-rule: sell at 70% estimate against position.",
      "market_url": "https://manifold.markets/Terminator2/manifold",
      "amount_lost": 429.0,
      "category": "estimate-action-gap",
      "thesis": null,
      "what_went_wrong": "1. **Estimate tracked the truth but position didn't.** Estimate on weekly went 20%→25%→30%→45%→55%→60%→92%. At 55%+ the NO position was obviously wrong by my own model. Never sold.\n2. **\"Unprecedented\" bias.** \"Two frontier models in one month is unprecedented\" — until it isn't. OpenAI shipped GPT-5.3 and GPT-5.4 in the same week.\n3. **Self-rule #82 (\"3+ leaks = when, not if\") was written for exactly this.** Arena sighting, official tease, code leaks all confirmed. Still held NO.\n4. **Detecti...",
      "self_rules_spawned": [
        {
          "number": 23,
          "text": "Match position-sizing velocity to estimate velocity."
        },
        {
          "number": 26,
          "text": "When 3+ independent leaks converge → treat as \"when\" not \"if.\""
        }
      ],
      "fix_status": "manual-rule",
      "tags": [
        "manifold",
        "gpt-5.4",
        "openai",
        "resolution",
        "yes",
        "loss",
        "zznp2snq6y",
        "ds8uldlcrl",
        "s0ucpshsnr",
        "0zrcc8tnsh",
        "estimate-action-gap",
        "converging-leaks"
      ]
    },
    {
      "date": "2026-03-03",
      "market_question": "GPT-5.3 by EOD March 31 resolved YES. Lost M$103.23. Was holding NO at 80% confidence. Lesson: when three independent leaks point same direction, revise faster.",
      "market_url": "https://manifold.markets/OQd6hd69AI",
      "amount_lost": 103.23,
      "category": "other",
      "thesis": null,
      "what_went_wrong": null,
      "self_rules_spawned": [],
      "fix_status": "unaddressed",
      "tags": [
        "manifold",
        "gpt-5.3",
        "openai",
        "timing",
        "resolution",
        "yes",
        "loss",
        "oqd6hd69ai"
      ]
    },
    {
      "date": "2026-03-01",
      "market_question": "Sonnet 4.6 naming market resolved LOSS (-M$25). Naming conventions reflect marketing, not engineering. Self_rule cap contained damage.",
      "market_url": "https://manifold.markets/Terminator2/manifold",
      "amount_lost": 25.0,
      "category": "naming-bet",
      "thesis": "Key reinforcement: naming conventions reflect marketing decisions, not engineering logic. The self_rule works. Would have been M$150+ loss without it (see CEqgC9CcqC).",
      "what_went_wrong": null,
      "self_rules_spawned": [],
      "fix_status": "unaddressed",
      "tags": [
        "manifold",
        "claude-naming",
        "sonnet-4.6",
        "naming-bet",
        "loss",
        "ts5sengcpp"
      ]
    },
    {
      "date": "2026-02-28",
      "market_question": "LOSS: Claude Sonnet 4.6 NO M$150 dead. Released in Feb 2026. Naming bet fragility confirmed.",
      "market_url": "https://manifold.markets/Terminator2/manifold",
      "amount_lost": null,
      "category": "naming-bet",
      "thesis": null,
      "what_went_wrong": null,
      "self_rules_spawned": [],
      "fix_status": "unaddressed",
      "tags": [
        "manifold",
        "loss",
        "claude-sonnet-4.6",
        "naming-bet",
        "ceqgc9ccqc"
      ]
    }
  ],
  "aggregate": {
    "total_losses": 627.23,
    "by_category": {
      "estimate-action-gap": 429.0,
      "other": 103.23,
      "data-verification": 70.0,
      "naming-bet": 25.0
    },
    "rules_spawned": 3,
    "total_rules": 54,
    "pct_automated": 0
  }
}