Reliability engineering · 2026 · Solo — research, systems design, build, verification

Retry Yard

An interactive virtual-time switchyard for reviewing webhook retry policy before shipping a production worker.

The problem

Webhook retry behavior is usually spread across queue settings, worker code, HTTP response handling, and idempotency storage. That makes policy interactions difficult to review before implementation. Retry Yard compresses those concerns into one deterministic local simulation without sending payloads, calling endpoints, or claiming parity with a specific provider.

Architecture

Key decisions

01

Use virtual time instead of timers

Every run resolves synchronously into an auditable trace. The same seed and policy produce the same schedule, which makes retry behavior reproducible in tests and reviewable in the interface.

02

Model reliability concerns together

Backoff, jitter, Retry-After, idempotency, permanent failure, and maximum-attempt exhaustion share one state machine so their interactions are visible rather than explained in isolation.

03

Send no webhook traffic

The simulator requests neither endpoints nor payloads. It is a bounded planning instrument for policy review, not a vendor emulator or production delivery worker.

04

Make causality the interface

The safety-orange control tower, envelope switchyard, and selectable docket expose why each attempt moved, waited, succeeded, deduplicated, or entered the dead-letter bay instead of reducing the result to summary cards.

Metrics

16
behavior and UI tests passing
5
explicit attempt dispositions
2
backoff policies modeled
0
network requests sent by a simulation