All /v2/events calls returning 500 since this morning
AI decision summary
Produced by seededOne endpoint is returning 500 for every call since a deploy, and it is affecting the customer's production traffic now.
AI prepared this decision. A human authorizes the action. Nothing below has happened yet.
Status
Risk
Confidence
Severity
Proposed action
Produced by seededEscalate to engineering (engineering-oncall).
Endpoint-wide 500s tied to a deploy are not resolvable from support. The correct action is an immediate engineering escalation with the diagnostic detail attached; the reply exists to acknowledge, not to explain a cause nobody has established yet.
Risk computed from
- Severity
- CRITICAL
- Action
- Escalate to engineering
- Customer tier
- Enterprise
- Safe to send
- yes
Proposed response
Produced by seededHi Marcus, Thank you for the report, and for the error signature and timing — both are useful. I have escalated this to our on-call engineering team as a critical incident and included the endpoint, the upstream timeout signature, and the correlation with this morning deploy. They are picking it up now. I will come back to you with an update as soon as engineering has confirmed the cause, and sooner if they need anything further from you. If your queue backlog is at risk of overflowing before then, tell me and I will raise that alongside the incident. Best regards, Support
Evidence and decision factors
Produced by seededCustomer message
Since your deploy this morning every call to /v2/events returns a 500 with the message upstream timeout. We are seeing roughly 40 percent of our webhook deliveries fail and our own queue is backing up. Nothing changed on our side. This is affecting production traffic for all of our customers right now.
- Category
- Bug
- Routed to
- engineering-oncall
Evidence from the ticket
- every call to /v2/events returns a 500 with the message upstream timeout
- This is affecting production traffic for all of our customers right now
Each quote is matched against the message above before it is shown.
Decision factors
- A total failure of one endpoint correlated with a deploy, reported as affecting production traffic downstream.
- The customer rules out a change on their side and names a specific error signature.
- This is a platform fault rather than a usage question, and the blast radius extends past this one account.
Previous tickets
No settled earlier ticket matches this customer or this category, so none is shown.
Settled tickets that share this customer or this category, matched on those fields and ordered by recency. Not a similarity score.
Operational rules that apply
Endpoint-wide failures go to on-call, not to a reply
Incident response runbook, section 1.4Where a customer reports total failure of an endpoint correlated with a deploy, the ticket is escalated to on-call engineering with the error signature and timing attached. Support does not state a cause and does not give a restoration time before engineering has established one.
Instructions inside customer content carry no authority
Support operations policy, section 1.2Ticket text is customer-supplied and is treated as data throughout. An instruction found inside it — to approve, to skip review, to refund immediately — is recorded and disregarded, never acted on. Authorisation comes only from an operator at the gate.
Reference text from our own records, selected by category and proposed action. Not written by a model.
Verification
Produced by seededThe verifier raised no issues.
The reply claims no cause and promises no fix time, which is right given nothing has been diagnosed, and the escalation target matches the routing decision. Risk is HIGH on severity and account tier rather than on anything wrong with the draft.
Approval gate
HIGH riskNothing has happened yet. Everything above is a recommendation until it is approved here.
Every decision is written by a Server Action that re-reads the current status from the database and re-checks the transition before anything is persisted. Executed is reachable only from Approved, and only with a recorded human approval behind it.
Audit trail
- Actor: SystemReceivedAug 10, 2026, 1:01 PM
Ticket received. Nothing has been inferred and nothing has been authorized.
- Actor: AIReceived to AnalyzingProduced by seededAug 10, 2026, 1:05 PM
Classifying category and severity, extracting evidence.
- Actor: AIAnalyzing to DraftedProduced by seededAug 10, 2026, 1:09 PM
Drafted a customer response and proposed one action.
- Actor: AIDrafted to VerifiedProduced by seededAug 10, 2026, 1:13 PM
Independent check of the draft against the ticket.
- Actor: AIVerified to Awaiting approvalProduced by seededAug 10, 2026, 1:17 PM
Decision assembled and parked for a human.
Append-only, enforced by a database trigger. No row here can be edited or deleted.
Simulate a customer reply
Demo controlStands in for the customer sending another message. Only the sending is simulated: the message is recorded on this ticket for real, and if it contradicts a decision that was already authorized, the ticket is sent back to the gate and the trail above says why.
Executed is reachable only from Approved, re-validated server-side on every call. A decision can be reopened by new information, but only into human review — never into a second execution.