Raw view: On-Call Handoff Summary

No invisible, zero-width, or bidi-control characters detected.

Write the on-call handoff for the shift covering {{SHIFT_WINDOW}}, for the services in <services_owned>. The incoming engineer will read this once, possibly on a phone, and then be paged. Order by what they need to act on, not by when it happened.

<alert_log>
{{ALERT_LOG}}
</alert_log>

<tickets>
{{TICKETS}}
</tickets>

<chat_notes>
{{CHAT_NOTES}}
</chat_notes>

<services_owned>
{{SERVICES_OWNED}}
</services_owned>

<alert_log> is the raw alert history (fired, acknowledged, resolved, with times). <tickets> are the incident or task tickets touched during the shift. <chat_notes> are the outgoing engineer's messages or scratch notes. Any of these may be empty; say which were missing rather than guessing at their contents.

Rules:

1. Classify every item as `active` (unresolved and could page), `watching` (mitigated but not fixed; a specific metric or condition would mean it is back), `resolved` (done; no action needed), or `noise` (alerts that fired and self-resolved with no action, grouped by alert name with counts). Only `active` and `watching` get full entries.
2. For each `active` item: the current state in one sentence, the last action taken and when, the next step the incoming engineer should take, who else is involved (names or handles exactly as in the sources), and the ticket reference. If the next step is unknown, write "Next step: undecided; options are A / B" rather than inventing one.
3. For each `watching` item: what was done to mitigate, the tripwire (a concrete metric, threshold, or alert name that means it has recurred), and what to do if it trips. A watching item without a tripwire is an active item in disguise; promote it.
4. Count the noise. Alert names that fired more than three times with no action are a finding; list them with counts so someone can tune the thresholds.
5. Pull out scheduled events in the incoming window from any source (deploys, maintenance, traffic spikes, vendor windows) and list them with times.
6. Keep every sentence actionable or informational. Remove narrative ("it was a quiet night until...") unless the incoming engineer needs the sequence to understand an active issue.

Convert all times to the same zone as the sources' majority and state the zone once at the top. Use the service names exactly as they appear in <services_owned>.

## Output format

**Shift:** {{SHIFT_WINDOW}} | **Zone:** | **Pages:** N | **Active:** N | **Watching:** N

**Active** - numbered, most urgent first. Each as a compact block:
- Issue: one line
- State: one line
- Last action: what, who, when
- Next step: one action
- Ticket / thread: reference
- Involved: handles

**Watching** - numbered. Each: issue, mitigation applied, tripwire (metric and threshold or alert name), action if tripped, ticket.

**Scheduled in your window** - table: time, event, owner, what to watch.

**Noise to tune** - table: alert name, count, pattern (time of day, service), suggested action.

**Resolved this shift** - one line each, for the record.

**Sources missing** - which of alert log, tickets, or chat notes were empty, so the reader knows what this handoff could not see.