---
id: reactive/retry-semantics
name: Implement retry semantics — event log append with new requestId, status projection respects retries
status: pending
depends_on:
  - reactive/workflow-root
scope: narrow
risk: medium
impact: component
level: implementation
---

## Description

Implement the retry semantics where retries are natural append events. A retry creates a new `call.requested` with a new requestId, and the status projection derives the current state by scanning for the most recent event per node. No `retried` status or state mutation needed.

## Acceptance Criteria

- [ ] `nodeKeyToRequestId.set(nodeKey, newRequestId)` overwrites the previous requestId — the projection tracks the latest attempt
- [ ] `getResult()` derives from the most recent terminal event (responded/error/aborted) for the current requestId
- [ ] Retry sequence: `call.error(requestId=1)` → `call.requested(requestId=2)` → `call.responded(requestId=2)` → status is "completed" (from latest event)
- [ ] Previous attempt events preserved in `eventLog` — full history maintained
- [ ] Conditional.test and Map.over see the latest result including retry outcomes
- [ ] Unit tests: retry sequence produces correct final status, getResult reflects latest attempt, event log preserves full history

## References

- docs/architecture/reactive-execution.md — retry semantics, event-to-status mapping
- docs/architecture/schema.md — CallEventMapValue

## Notes

> To be filled by implementation agent

## Summary

> To be filled on completion