Skip to content

Vitals Methodology (ADR-0133)

How the vitals instrument fleet is read as evidence: the widened-limit multiplier that holds the fleet's family-wise false-alarm budget, the minimum detectable effect per instrument, and which metrics belong on a time-between-events chart. Deterministic — a function of the registered series set and the ADR-0133 arithmetic alone. No live threshold is changed by this surface; it is the readiness signal for the eventual widened-limit migration.

Widened control limit — the registered second-order fleet

  • Registered series (charts evaluated together): 8
  • Family-wise budget: 5% / month (Bonferroni, two-sided)
  • Widened multiplier L: 3.000σ (classic Shewhart is 3.0σ)
  • Status: at the 3σ floor — 8 charts at a 5% budget computes a limit looser than 3σ, so the floor holds it at 3.0σ. The fleet does not need widening yet; the verdict's live 3σ limit is already correct.

Widening curve (readiness projection)

Where the 3σ floor lifts as the instrument fleet grows:

Registered charts Widened L Above 3σ?
8 ← current 3.000σ floored
15 3.000σ floored
19 3.008σ yes
30 3.144σ yes
50 3.291σ yes
70 3.384σ yes

Minimum detectable effect (per instrument, per window)

Baseline events a rate metric needs to detect a given rate ratio at α=0.05, 80% power. A metric whose window volume is below its row cannot evidence that effect — chart it as time-between-events instead of a rate.

Rate ratio to detect Baseline events needed
11
1.5× 39
1.25× 141
1.1× 824
0.9× 745
0.75× 109
0.5× 23

Scarce-event metrics (time-between-events candidates)

A per-window count that is usually zero (escapes on a low-merge repo) has no power as a rate chart — a monthly 'flat' reading on ~1.5 events/month is not evidence. These move to a Benneyan g/t-chart on the interval between events:

  • escapes per window (escapes) — escapes are rare by design; a g-chart on merges-between-escapes (+ the consecutive-zeros run rule) reads deterioration a rate chart's 0-pinned lower limit cannot.