{
  "slug": "bathtub-curve",
  "term": "Bathtub curve",
  "aliases": [
    "Bathtub failure-rate curve"
  ],
  "definition": "A bathtub curve is a failure-rate profile with decreasing early-life hazard, an approximately constant middle period and increasing wear-out hazard.",
  "note": "The three regions describe a population as it ages, not a countdown for an individual device. For first failures, the vertical axis is hazard conditional on survival, with inverse-time units, not cumulative failure probability. State when the age clock starts. For repairable systems, the analogous curve describes the rate of occurrence of failures instead. [nist-bathtub]\n\nThe shape is a model to test, not a universal disk lifecycle. A FAST 2007 study examined about five years of HPC1 replacements and three years for the first HPC4 drive type. Their age profiles did not show the assumed extended flat middle period. Only customer-visible replacements were observed; early failures caught during manufacturing, integration or installation testing could be absent. These results do not establish one aging pattern for all HDDs or SSDs. [schroeder-age]\n\nBefore comparing a fleet with this curve, specify the event, population, age origin and exposure. An annual replacement count and the interval between successive replacements across a cluster answer different questions from a device's first-failure hazard. [schroeder-age]\n\nSurvivor selection can produce a falling-then-rising cohort hazard while all subgroup hazards rise. [vaupel-zhang-selection] A constructed equal-slope example has its minimum at 19.78 model years and 0.042765/year. These are not estimated drive lifetimes; the example has no extended flat middle phase. [hesela-age-clocks]",
  "status": "shipping",
  "advantages": [],
  "limitations": [],
  "related": [
    "hazard-rate",
    "survival-function",
    "weibull-distribution",
    "annual-replacement-rate",
    "rate-of-occurrence-of-failures",
    "left-truncation"
  ],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "citations": [
    {
      "id": "nist-bathtub",
      "title": "NIST e-Handbook 8.1.2.4: Bathtub curve, lifetime phases and repairable-system rate",
      "url": "https://www.itl.nist.gov/div898/handbook/apr/section1/apr124.htm",
      "accessed": "2026-11-18"
    },
    {
      "id": "schroeder-age",
      "title": "Schroeder and Gibson: Disk failures in the real world, FAST 2007, sections 2.1, 2.4, 4.2 and 5.3",
      "url": "https://www.usenix.org/legacy/events/fast07/tech/schroeder/schroeder_html/index.html",
      "accessed": "2026-11-18"
    },
    {
      "id": "hesela-age-clocks",
      "title": "Hesela: Disk replacement rates are not disk lifetimes",
      "url": "https://hesela.com/analyses/mttf-vs-field-replacement-rates/",
      "accessed": "2026-11-18"
    },
    {
      "id": "vaupel-zhang-selection",
      "title": "Vaupel and Zhang: Attrition in heterogeneous cohorts (2010), equations 1a-1b, proof and section 4.1",
      "url": "https://www.demographic-research.org/articles/volume/23/26",
      "accessed": "2026-11-21"
    }
  ],
  "provenance": [
    "Source synthesis from the NIST handbook and the original FAST 2007 HTML paper; no new fleet measurements or reproduction of the original logs.",
    "Added an analytic turning-point interpretation of the existing Hesela survivor-hazard example; checked by 60-digit decimal arithmetic and derivative bisection, not new fleet data."
  ],
  "updated": "2026-11-21"
}
