Epic: consolidate Sentry + Alertmanager alerting into one real-time notification path

[JSONbored]

Problem: Two disconnected observability pipelines exist for this self-host instance (edge-nl-01). Prometheus/Alertmanager has 34 alert rules and genuinely routes to a live, watched Discord channel. Sentry independently captures rich, well-contextualized error data — but its only notification path is the shipped default alert rule (email to issue owners, falling through to active members), and every one of its 28 currently-open issues is unassigned. Neither pipeline reaches PagerDuty. The result: real, ongoing, high-volume production bugs (one at 4,030 events over 12 days) went completely unnoticed until a manual audit found them.

Area: Self-host observability / alerting

Proposal: Consolidate both pipelines into one real notification path an operator actually sees in real time, tracked as three independently-shippable sub-issues.

Deliverables:

• Sub-issues filed and linked below, each independently mergeable:
  • Wire Sentry to real-time notification (Discord/PagerDuty), not the email-only default #5008 — wire Sentry to real-time notification (Discord/PagerDuty)
  • Add a native PagerDuty receiver to Alertmanager for severity:critical alerts #5009 — add a native PagerDuty receiver to Alertmanager for severity: critical
  • Fix Sentry fingerprinting for 3 fragmented issue groups (analyzer_timeout, AI-review-notes, publish-failed) #5010 — fix Sentry fingerprinting for the 3 fragmented issue groups (merged via fix(observability): fix Sentry fingerprinting for 3 fragmented issue groups #5099)
• After all three: a real problem in either Sentry or Alertmanager reaches a human within minutes, not via an unread inbox.

Acceptance criteria:

• A Sentry issue reaching high priority notifies the same real-time channel Alertmanager already uses (Discord and/or PagerDuty). (Wire Sentry to real-time notification (Discord/PagerDuty), not the email-only default #5008)
• A severity: critical Prometheus alert reaches PagerDuty, not just Discord. (Add a native PagerDuty receiver to Alertmanager for severity:critical alerts #5009)
• The three known Sentry-issue-fragmentation groups (analyzer_timeout, "AI review did not produce public notes", "PR public-surface publish failed") are consolidated so their true combined volume is visible at a glance. (Fix Sentry fingerprinting for 3 fragmented issue groups (analyzer_timeout, AI-review-notes, publish-failed) #5010, merged via fix(observability): fix Sentry fingerprinting for 3 fragmented issue groups #5099)

Boundaries:

• Scoped to this self-host instance's own alerting, not the separate "centrally-hosted fleet for other operators" initiative (Stand up centrally-owned observability for a hosted fleet #4934/Wire on-call/paging for the central hosted service #4937/Internal ops observability & alerting #4808) — those are a different, unrelated product effort.
• Does not include fixing the 13 individual bugs Sentry surfaced (tracked separately: fix(review): claude-code subscription-CLI hangs/times out repeatedly (4,030 Sentry events, ongoing) #4994–fix(rees): investigate a REES /v1/ping 503 (7 Sentry events) #5006) — this epic is about the notification pipeline itself, not the bugs it should have surfaced sooner.

Cross-reference (added 2026-07-12)

See also the new AMS observability batch in milestone #25 (Miner Wave 4 — AMS Hardening & Packaging): #5184 (Grafana datasource for AMS's local ledgers), #5185 (coding-agent usage/cost dashboard), and #5186-#5188 (Prometheus alert rules for AMS). A self-hoster running both ORB and AMS on one box should get ONE consolidated alerting pipeline, not two independently-built ones — keep this epic's design in mind when AMS's alert rules land.

[linear-code]

JSO-3624

[JSONbored]

Status: #5010 done (merged). #5009 verified — Discord side already live, PagerDuty side needs a real account created (not something I can do). #5008 attempted directly via Sentry's API — blocked on the available token's scope (403 on org/integration endpoints), documented on that issue with exactly what's needed to finish it.