Escalation, reminders and quiet time

Three different answers to "nobody has dealt with this yet", and they live in three different places because they are three different questions.

Control	Question	Lives on
Escalation	if this stays unhandled, who else should hear?	the severity section
Reminder (repeat)	how often should this delivery chase?	the notify action
Quiet time	when should nobody be told — and told what, afterwards?	its own step (Quiet time), or Automations → Settings for every automation at once

---

Escalation

A chain of further actions taken when the alert is still live after N minutes.

Escalation is per severity, not per action. The base actions section hosts the rule-level chain — which is exactly the chain the engine resolves for an alert sitting at the base severity — and each per-severity section hosts its band's own.

On an action row, "if this stays unhandled, do more" would read as "if this email goes unanswered", while the one chain actually fires for the whole tier.

A stored per-action chain is hoisted and merged into its severity's chain when you open the automation: tiers concatenate and sort by afterMin, and since each tier carries its own actions the deliveries and their timings are unchanged. The action-level chains are then stripped so the sweep cannot fire both. Per-action chains stay in the schema, so a rule this wizard has not re-saved keeps working.

Each tier (+ Escalation Action; up to five per chain) carries:

Field	Means
Escalate after N minutes unhandled	wall time from the fire, up to a week
Actions	a full action list — usually a wider recipient set
Repeat every … min, max …	the tier's own re-send cadence (5 minutes or more, off by default) and how many times (1–20, default 5)

The chain as a whole carries Stop escalating when: Acknowledged (or cleared), the default, or Cleared only — acknowledging does not stop it.

The pattern this exists for

The trigger notifies Asset's L1 Region Users — the site's own people. Tier 1, after 15 minutes, notifies Asset's L2 Region Users — the division over them.

One automation, two honest audiences, no duplication.

---

Reminders

"Repeat this action" at the foot of a notify row: re-send every N minutes (default 15), until Acknowledged or Cleared only, …and give up after N hours (optional; blank means never). Quiet time is the automation's own step — below.

A reminder re-sends notify actions and nothing else, which is why it belongs to the action (rule 56). "Page the on-call every five minutes and leave the nightly digest alone" is one automation with two honest answers, and no per-automation or per-severity cadence can say it.

It is offered on the firing lists only — never on reset actions, and never on an escalation tier (which has its own repeatEveryMin). A recovery has nothing to chase.

Presence, not truthiness. An action declaring repeat: null is an answer ("this one does not chase"); an action carrying no key at all inherits the rule-level clock. Saving the step writes an explicit answer onto each action — but only when you have actually opened the step.

stopAfterHours is wall time from the fire, quiet time included. The wizard warns about that pairing rather than extending the deadline.

---

Quiet time

Hours during which an automation stays quiet (rule 92). A quiet period never drops an alert. The alert is still raised, shows on the Active Alerts page with a QUIET pill, and writes its audit event; scripts and API calls still run. What changes is who hears about it, and when.

There are two places to set one:

Where	Applies to	Set from
An automation's own quiet time	that automation only	the wizard's Quiet time step (step 6)
A global quiet time	every automation that has no quiet time of its own	Automations → Settings → Global Quiet Times

The automation's Quiet time step is a three-way choice:

Setting	Means
Off	No quiet time of its own. The global quiet times apply to it.
Ignore Global Quiet Time	No quiet time at all. It sends whatever the hour, even inside a global quiet time — what a critical automation usually wants.
Override Global Quiet Time	Its own quiet time, set below the choice. The global ones do not apply to it.

Either of the last two leaves the automation alone when a global quiet time opens — that is how a critical automation keeps paging while everything else waits.

What goes quiet

Each quiet time says, per severity, which sends it holds. Under every severity you tick are four boxes:

	Means
Alerts	the first alert (and a grouped alert's growth update) — the one send that owes a summary email afterwards
↳ Reminders	the alert's own reminders ("Repeat this action")
Escalation alerts	each escalation tier's first run
↳ Reminders	a tier's own repeats

Ticking a severity ticks all four; a ticked box is held during the quiet period, an unticked one goes out live whatever the hour. Untick a severity to let it through entirely — the usual shape is everything but critical, and a new global quiet time starts that way. Untick Alerts but leave the rest to hold only the chasing: the first alert and the all-clear still send, the chasing waits for the period to end. With no severity holding its first alert there is nothing to report, and the summary section disappears.

On an automation's own quiet time (Override Global Quiet Time) the tree shows only what that automation can actually send: its own severity and its severity bands' — not all five — and under each, Reminders only when a notify action at that severity repeats, Escalation alerts only when an escalation reaches it, and their Reminders only when a tier repeats. Add a reminder or an escalation on the Actions step and its box appears here, held.

Held means held, never skipped: nothing in the alert's escalation clock moves while a send is quiet. The reminder that ends a hold still says "Reminders resumed after a quiet period — this alert has been active for 9h 12m" and carries · ACTIVE 9h 12m in its subject (rule 44).

Which alerts

• Severities. The tree above. Untick a severity to let it through whatever the hour.
• Kinds of alert (global quiet times only). Any alert, or only the metrics and device states you pick — CPU, interface status, PoE, and so on. Audit-event and change automations carry no kind and match only "Any alert".
• Devices (global quiet times only). The same device filter the automation wizard uses.

When

The schedule editor is the Maintenance modal's. For specific days of the week you write periods — "Mon–Fri 22:00–06:00", "Sat, Sun all day" — each a row of day chips with either all day or one or more hour ranges; add as many periods as the week needs, or start from a preset (Nights and weekends, Every night, Weekends only, Outside business hours). Quiet outside these hours turns the periods inside out, so working hours are what you type and the quiet time is everything else. A week strip and a per-day breakdown under the periods show exactly what will be quiet. An overnight range carries into the next morning, so Nights and weekends puts Sunday on the 22:00–06:00 period as well as on the all-day one — that is what keeps Monday 00:00–06:00 quiet. A day that is quiet all day may share an overnight range with another period (only its next-morning tail counts); any other range on it is refused as a double listing. Monthly and yearly schedules take a day of the period and an hour list, and every schedule can carry a first and last date. Times are server-local wall clock — the zone is printed beside the hours, and a browser prefilling 22:00 from its own clock is the whole trap. A 22:00–06:00 window survives DST and midnight. A window set through the API that the editor cannot express (a one-shot) is listed read-only and kept as it is.

The summary email

When a quiet period that held everything ends, Polaris emails everyone the held alerts would have reached — the people on the notify actions, address-book contacts, typed addresses, and anyone who would have been pushed (reached by their account's email; a push preference is ignored, since there is no summary push). One email per person, in that person's own time zone, about that person's own alerts. A global quiet time can cover automations that page different teams, and each person's email lists only the alerts that would have reached them. Someone on two teams' notify lists sees both teams' alerts. Everyone else the quiet time covers gets one too: someone whose alerts all recovered, or whose automations held nothing that night, gets the all-quiet email (below), or nothing if the all-quiet email is turned off. It lists:

• Still outstanding — the held alerts that have not recovered or been cleared by the time it is sent: severity, device, what, since when, how long. An alert that came and went inside the period is not listed.
• Recurring — an alert that fired more than X times during the period for the same automation, device and component (the recurrence threshold on the quiet time), listed with the count and every time it fired, whether or not it has recovered. It is not repeated under Still outstanding.
• Sent during quiet time — alerts that fired during the quiet period at a severity the quiet time does not hold (a critical, say, on a quiet time that holds only serious and below). Those went out as they happened; the summary lists the ones that reached you, with when each fired and whether it has recovered, so the morning email tells the whole night. Someone paged overnight gets this list instead of an all-quiet email.

Each device name opens the device in Polaris.

The all-quiet email. When the quiet period ends with nothing to list for you — every held alert recovered, or nothing was held at all, and nothing was sent to you live — the summary still goes out, saying so: nothing is outstanding, how many alerts fired and recovered (or that none were held), and that its arriving confirms the quiet time and your email delivery are working. It goes to everyone the covered automations notify, whether or not another team's alerts were held that night: the people, addresses, roles, tags and regions their notify actions name. An action that notifies the device's region (or a region level above it) reaches the users of every region the automation's monitored devices are in — everyone it could have paged from any of them. A device's own address-book contacts are left out. It is on by default; untick All-Quiet Summary emails on the quiet time to get a summary only when there is something to list — something held, or an alert sent to you live.

When it is sent. When the quiet period ends, or at a send time you choose — a time of day. On a day it falls inside the quiet period the summary waits: nights and weekends with a 07:30 send time summarises each weekday night at 07:30 and rolls the weekend into Monday's. A send time that is inside the quiet period on every day it occurs could never go out on time, so the editor and the server refuse it. Alerts held until the send time roll into that summary; if another quiet period opens first, they roll into its summary. Through which channel: the one you pick on the quiet time, else the alert's own email channel, else the first enabled email channel. With no email channel at all the summary cannot be sent, an Event says so, and the alerts are still on the Active Alerts page.

Once an alert has been named in a summary it behaves like any other: its reminders and escalation count from the summary, and its all-clear is sent. The all-clear of a held alert that was never summarised sends nothing — there is no inbox to resolve it in.

The Settings modal lists recent summaries: when, which quiet time, how many alerts it covered, how many it listed, and who it reached. Covered, Listed and Recipients are clickable: Covered opens every alert the summary stamped, as it is now, marking the ones the email named; Listed opens everything the quiet period held that was still outstanding or recurring, and every alert sent live during it, across every recipient; Recipients opens who it went to, with what their email listed, each address's sent / failed state and the error if there was one. View email on a recipient shows the email that person was sent: subject, the time zone it was written in, and the body as it looked in their inbox. A send that fails is retried for about ten minutes; a summary that still did not reach everyone shows Resend, which sends it again to the recipients it missed (and nobody else) once the channel is fixed.

Recipient exceptions

Some people should hear about an alert whatever the hour: an on-call phone, a NOC mailbox. Recipient exceptions names them (rule 120). The Recipient exceptions… button sits under What goes quiet on an automation's own quiet time, and on the When, and what happens step of a global quiet time. It opens a list of email addresses. Each address has its own copy of the severity tree:

• Nothing ticked (how a new row starts): that address ignores the quiet time and gets every alert, reminder and escalation live.
• Some boxes ticked: those sends are held for that address, during the same hours as everyone else. An exception can hold more than the quiet time itself, too — for example a manager who wants critical reminders held overnight while the on-call engineer still gets them.

Everyone not on the list follows the quiet time as usual. A push recipient is matched by their account's email address. Slack, Teams and Pushbullet channels have no address, so they follow the quiet time.

What each person gets:

• The first alert. One email goes out live, To the excepted addresses only, and the alert is held for everyone else. The alert gets a QUIET pill and a summary only when somebody was actually held.
• Reminders and escalation tiers. The send goes out live to the excepted addresses, so the alert's reminder clock keeps running. Everyone it was held for gets the newest held copy when the quiet period ends. They do not get one email per missed reminder. If the alert recovers or is acknowledged first, the held copies are dropped. The copy they get was written when it was held, so its active for time is the age at that moment.
• The summary email lists, for each person, only their own held alerts. An address held for nothing gets no summary, no all-quiet email and no list of what it was sent live — it was never quiet.
• The all-clear goes to everyone who heard about the alert, live or through a released reminder. It skips the people it was held for who have heard nothing since.

An alert email's also sent to footer names only the people that email actually reached, never the ones the quiet period held.

What quiet time is not

A maintenance window is not a quiet time. It stops polling the device, so no new alerts are raised about it at all, and it freezes an open alert for the whole window — except a device-down alert, which it clears (rule 16). Quiet time keeps watching and keeps raising; it only decides who is told, and when.

A wizard test delivery is never held — it exists to show you the email. Nor is an automation with no notify action — one that only writes the audit event or only runs a script has nothing for quiet time to withhold, so its alerts never appear in a summary.

Validation

A half-typed day contributes no window, so the step names the day — and the overlapping pair of hours — in the same words the server would, and a day listed both all day and with hours is refused by name. A send time inside the quiet period on every day it occurs is refused. The reminder note on the Actions step still warns that stopAfterHours counts quiet time too.

---

What stops all of it

Event	Effect
Acknowledge	stops escalation and reminders where stopOn says so; the alert stays live
Clear	ends the alert, runs the reset actions, stops everything
The reset condition becomes true	same as Clear, automatically
The device leaves the scope	alert cleared as out_of_scope
The device stops being monitored	same (rule 37)
A maintenance window opens	a device-down alert (monitorStatus == down) is cleared — down is what planned downtime looks like, and a device still down afterwards fires a new alert. Every other alert stays open with escalation and reminders paused until the window ends, then resolves or carries on as usual (rule 16)
The device itself goes down	an alert about something the device reports (CPU, an interface, a sensor…) clears as superseded and the asset-down alert speaks for the outage — see Triggers (rule 29)
The parent goes dark	dependency suppression retires it — a parent genuinely down, not one in maintenance (rule 38)
A more specific automation carves the device out	cleared as superseded (rule 18)
The pin is removed	a dimensioned alert clears (rule 57)

A quiet period is not in this table on purpose: it ends nothing and pauses nothing about the alert itself. It holds the sends, and the summary email reports what is still outstanding when it ends (rule 92).