Notification Priority Review¶

Review — instantiates Salience-Significance Decoupling

Re-examines an alerting system so that what pages a human is governed by significance and escalation criteria, not by how loud or how often an alert happens to fire.

Notification Priority Review is the alerting-domain instance of the archetype: a periodic review of a live notification stream whose job is to pull escalation apart from salience. An alert's prominence — how red it is, how frequently it fires, how it was styled — is not the same as its significance, yet routing rules quietly conflate them, so the alert that fires most captures the on-call human's attention regardless of what it predicts. The review re-derives which signals earn escalation from significance criteria, and installs a gate so that loudness alone can no longer reach a person.

Example¶

An on-call SRE is drowning: a disk-usage warning fires hourly and resolves itself; a certificate-expiry alert scrolls past unread. A Notification Priority Review inventories every alert twice — once by how often and how loudly it fires, once by what it actually predicts (a customer-facing outage? irreversible data loss?). It then resets escalation thresholds against the second column: the chatty disk warning drops to a dashboard tile, while the quiet cert-expiry is promoted to a page. Finally it installs a decision-weight gate — no alert pages a human unless it clears a significance bar, however alarming its colour. The outcome is fewer pages and a higher hit rate: the signal that matters now gets through the noise that used to bury it.

How it works¶

What distinguishes it from a general prioritisation tool is that it operates on a live alert stream and its routing rules:

Score each alert on two columns — fires-a-lot / loud versus predicts-something-significant — for the actual firing history.
Reset escalation thresholds on the significance column, deliberately stripping frequency and styling out of what decides routing.
Gate the page on criteria, not on prominence — an alert must clear a significance bar to reach a human, no matter how salient it looks.

Tuning parameters¶

Escalation bar — how significant an event must be to page rather than log. Set too high it misses real incidents; too low it re-floods the pager.
Loudness decoupling — how aggressively colour and firing-frequency are stripped out of routing decisions.
Gate strictness — what significance threshold action (paging, auto-remediation) requires before it fires.
Review cadence — a scheduled periodic sweep versus a review triggered after every incident or false-alarm storm.

When it helps, and when it misleads¶

Its strength is that it attacks alarm fatigue directly — the state in which frequent, salient, non-actionable alerts desensitise responders until a real one is missed in the din.^[1] Separating escalation from loudness is precisely what restores trust in the pager.

Its failure mode is over-correction: tuning aggressively to cut noise can raise a bar high enough to mute a rare but genuine signal. The classic misuse is silencing an alert because it is inconvenient rather than insignificant — turning off the smoke detector because it keeps going off during cooking. The discipline that guards against this is to justify every threshold change against outcomes — events missed versus events caught — rather than against how annoying the alert felt, and to re-check the gate after each incident.

How it implements the components¶

Notification Priority Review fills the escalation-and-gating side of the archetype — the components that govern whether prominence reaches a person:

escalation_threshold_separation — its signature move: the threshold that escalates an alert is decoupled from the alert's loudness and firing frequency, set instead on what the alert predicts.
decision_weight_gate — installs the gate that blocks a prominent (red, frequent) alert from driving action before its significance criteria are met.

It does not score general items on a salience/significance grid (reweighting_and_revision_rule — that's Salience-Significance Matrix), define the underlying significance rubric (significance_criteria_register — that's Evidence Weighting Rubric), or reconstruct sampling frames (comparison_set_restoration — that's Sample Frame Reconstruction). It applies gating to a live alert stream.

Instantiates: Salience-Significance Decoupling — it is the alerting-specific gate that keeps loud alerts from being mistaken for urgent ones.
Consumes: Evidence Weighting Rubric or Salience-Significance Matrix — supplies the significance criteria the escalation bar is set against.
Sibling mechanisms: Salience-Significance Matrix · Dashboard Salience Calibration · Evidence Weighting Rubric · Ranking Semantics Legend · Shown-vs-Unshown Audit

Notes¶

Do not confuse it with Dashboard Salience Calibration: that mechanism adjusts what a dashboard visually emphasises, whereas this one governs what escalates and pages. A dashboard can safely make a low-significance metric prominent for context; a pager cannot, because prominence there is a claim on a human's night.

References¶

[1] Alarm fatigue — well documented in clinical and industrial safety: a high volume of frequent, non-actionable alarms desensitises operators, so that a genuinely significant alarm is ignored along with the rest. Decoupling escalation from firing-frequency is the standard corrective. ↩