Your savings number lives on a dashboard nobody watches. What pages someone when it drifts?

A documented metric bound, wired to a pager. The console lands on the verdict; press play and a lowest-net-cost routing metric drifts past its bound, fires a SEV, and opens its runbook. The dashboard never changes; the on-call does. One click plays it; the second panel steps a prior-auth subgroup gap to a SEV-1.

Verdict

Day 0 / 40

Panel 1 · Lowest-net-cost routing yield on an on-call

INSIDE BOUND, no page. Routing net-cost-per-fill nominal.

net-cost-per-fill, % above plan target bound = 3.0% → SEV-2

● SEV-2 · ROUTING DRIFTPAGES ON-CALL

Metric: Lowest-net-cost routing yield (net-cost-per-fill vs plan)
Bound: > 3.0% above target
Observed: —
Fired: —
Blast radius: Top-line financial: the savings claim erodes silently
Response: On-call paged. Per-partner routing share reviewed within 1h.

Runbook RB-ROUTING-001 — open

First 5 minutes

Acknowledge the page. Confirm the breach is real, not a price-feed artifact.
Pull per-partner routing share for 14 days. Find which partner's stale price moved the mix.
Check for a concurrent formulary change, contract update, or partner outage.

Decide

A savings number that drifts unpaged is a number you can no longer stand behind. Page, do not average it away.
Stale price feed correlates: freeze routing to last-known-good partner ranking pending refresh.

Generated post-mortem template

—

Panel 2 · Prior-auth subgroup parity ledger

Subgroup approval-rate gap within bound. No page.

Step each weekly PA review. The approval-rate gap between subgroups widens toward over-denial; crossing 4 pp pages a SEV-1.

● SEV-1 · PA SUBGROUP GAPPAGES ON-CALL

Metric: Prior-auth approval-rate gap, subgroup vs baseline
Bound: > 4.0 pp gap
Observed: —
Blast radius: Clinical / regulatory: a member denied a needed drug
Response: Licensed-clinician review before any automated denial stands.

Runbook RB-PA-001 — open

First 5 minutes

Page on-call clinical owner. Hold automated denials for the drifting subgroup.
Pull per-subgroup PA decisions for 14 days; confirm the gap is real, not a coding artifact.
Bias toward holding the automated decision; over-denial is harm-bearing.

Context

This is the surface under active litigation at the medical-claims insurers; presented as analogy, not a PBM precedent. A decision surface that cannot show when it drifts is a liability waiting for a docket number.

Sources & method

Both metric series are SYNTHETIC and generated deterministically (seeded) in this page; no real SmithRx savings, routing, or prior-auth number is shown or implied. The numbers are illustrative; the operating shape (a bound wired to a page, with a runbook) is the point.
Method: metric-SEV, documented metric bounds treated as paged on-call severities with runbooks and post-mortems. Published at jeffpinto.com/notes/metric-sev, grounded in a real Meta data-quality lockdown (roughly 14x detection-lift, a published figure).
Honest bound: this is the operating shape on a synthetic series, not your live metrics. Standing the ladder up on your real savings, routing, and PA metrics with named owners is the internal extension this scopes.