{
  "aliases": [],
  "status": "research",
  "advantages": [],
  "limitations": [],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-21",
  "slug": "feedback-age",
  "term": "Feedback age",
  "definition": "Feedback age is the elapsed time between observing a system state and using that observation to make a later decision.",
  "note": "Specify both endpoints: sample-to-decision age includes transport plus any time the receiver holds the sample before using it. Receipt-to-use age omits the first interval. An exact cross-host age requires comparable clocks or an explicitly bounded clock error. [age-example]\n\nIn Hesela's synthetic example, two queue reports each take 2 ms to arrive but are 102 ms and 2 ms old at selection. Hidden arrivals can reverse which replica is faster without changing either report. A policy that discards the older report helps in one constructed world and harms in the other; freshness alone is not an accuracy guarantee. [age-example]",
  "related": [
    "tars",
    "adaptive-replica-selection",
    "tail-latency"
  ],
  "citations": [
    {
      "id": "age-example",
      "title": "Hesela. Feedback age: identical observations, opposite routing outcomes. Exact synthetic construction (2026).",
      "url": "https://hesela.dev/research/feedback-age/README.md",
      "accessed": "2026-11-21"
    }
  ],
  "provenance": [
    "Definition uses an explicit sample-to-decision convention; the companion finite model also exposes transport and receipt-to-use intervals."
  ]
}
