Methodology

generated 2026-10-13T15:16:05Z - feed

Selection rules

Question batteries

Fresh CVEs

{
  "version": 1,
  "questions": {
    "kev_within_90d": {
      "type": "noul",
      "instructions": "Will this CVE be added to the CISA Known Exploited Vulnerabilities catalog within 90 days?",
      "criteria": {
        "true": "The CVE ID is added to the CISA Known Exploited Vulnerabilities catalog within 90 days of this forecast; CISA lists it because there is reliable evidence of exploitation in the wild.",
        "false": "Ninety days pass after this forecast without the CVE ID appearing in the CISA Known Exploited Vulnerabilities catalog."
      }
    }
  }
}

Show HN

{
  "version": 2,
  "questions": {
    "points_30_within_48h": {
      "type": "noul",
      "instructions": "Will this Show HN post have 30 or more points when checked at or after 48 hours?",
      "criteria": {
        "true": "The post's score on Hacker News is 30 points or higher at the first check on or after 48 hours from this forecast.",
        "false": "The post's score stays below 30 points at the first check on or after 48 hours from this forecast."
      }
    }
  }
}

HackerOne disclosures

{
  "version": 1,
  "questions": {
    "same_cwe_disclosed_30d": {
      "type": "noul",
      "instructions": "Will a publicly disclosed HackerOne report with the same CWE be disclosed for a different program within 30 days of this report's disclosure?",
      "criteria": {
        "true": "Within 30 days after this report's disclosure date, at least one other publicly disclosed HackerOne report shares its CWE and belongs to a different program.",
        "false": "No publicly disclosed HackerOne report with the same CWE appears under a different program within 30 days of this report's disclosure."
      }
    }
  }
}

Baselines

Every prediction freezes a baseline record {value, method, window} measuring the same event over the same window the prediction resolves on; the method string on the entry states the sampling rule. Samples below the minimum fall back to a stated constant.

slotwindowmin samplefallback
Fresh CVEs90d after NVD publish1000.05
Show HN48h after post creation200.3
HackerOne disclosures30d after anchor disclosure50.3

Resolution rules

Scoring

Rows report Brier score for snap and for the frozen baseline over resolved predictions (lower is better); unresolved entries are counted and excluded from the score. Verdicts render once a slot reaches 5 resolved predictions; below that the rows show counts only.

Known limitations

H1 CWE coverage measured at 78.0% of recent disclosed reports (baselines.measure_cwe_coverage).