Statistical inference from operational experience can bound a system's failure rate without a structured assurance argument
3 events · 1 assessment · 1 decision
Structured and assessed
First pass (structure_and_assess). Decomposed into two named arguments: a supporting line resting on one new subclaim (statistical confidence bounds from failure-free exposure; Matcher confirmed novel, seeded at 0.97, importance 0.15 so left a deferred stub) and a counter line linking two existing claims found via match/search (19d2b0f5 on the infeasibility of demonstrating ultra-low rates by testing; de5b07d6 on assumption-dependence of test evidence), both attached as contradicts under the against-argument. Evidence pass: Littlewood & Strigini CACM 1993 and IEC 61508 proven-in-use/Route 2H practitioner sources confirm both the statistical practice and the qualifications; two affirming instances recorded (Intertek, I&E Systems), none denying. Assessed SUPPORTED (confidence 0.8, credence 0.85, marginal yield 0.15): the claim is true in its plain reading, with the live dispute confined to whether the inference's assumptions make it implicitly argument-dependent; kept below verified because the contradicting subclaims are unassessed and that objection has not been run to ground. Importance raised slightly to 0.3 (contestation 0.3): notable premise in the live assurance-argument debate, niche beyond it. Canonical form left unchanged; it is already a neutral, contrastive statement both sides would accept. No curator escalations needed.
Assessed Supported
verdict confidence 0.80 · credence 0.85
Statistical reliability theory and industrial practice both bear this claim out in its plain reading. Standard statistical methods yield confidence bounds on a failure rate from failure-free operational exposure: zero-failure operation over a known number of demands or hours supports an upper confidence bound by direct calculation, and functional-safety standards codify the practice, with IEC 61508's proven-in-use and Route 2H provisions deriving dangerous-failure rates at stated confidence levels from operating history alone. No structured assurance argument enters these derivations. Two credible qualifications limit what the claim licenses without unseating it. First, statistical testing alone cannot demonstrate the ultra-low failure rates required of safety-critical systems: the exposure needed to bound a failure rate at, say, one in a billion hours is infeasible, a point established by Littlewood and Strigini and by Butler and Finelli, so the bounds obtainable in practice are far weaker than the targets safety certification typically demands. Second, the evidential weight of operational data depends on assumptions about representativeness: the bound holds only insofar as future use resembles the observed operating profile and failures were reliably detected, and justifying those assumptions is itself a form of argument. The claim is therefore best read as true about what statistics can do, and silent about whether what it does suffices for assurance at safety-critical levels.
Claim entered the graph