Callback Lab.
Two ways to inspect delivery: compare a two-event scenario, or vary the original single-event parameters. Follow each attempt and its evidence.
Callback Scenario Workbench
Receipt, effect and acknowledgement are different facts. Follow two numbered events through a receiver with an explicit ordering policy.
Six local experiments. The clock is virtual; the seed changes bounded retry jitter, not a real service's reliability. The single-event parameter lab below remains available for independent experiments.
No scenario executed. The worked ordering example below is readable without JavaScript.
State meanings, limits and the ordering policy
- delivered / failed
- An envelope arrived / a chosen failure occurred. A lost acknowledgement can follow a successful effect.
- retry_scheduled / exhausted
- A bounded virtual delay is scheduled / no attempt remains. Neither state proves whether an effect exists.
- duplicate / identity_conflict
- A matching committed identity reuses its outcome. The same identity with a different payload is quarantined.
- out_of_order
- A later business sequence waits in the receiver's buffer. It has no acknowledgement yet.
- committed / acknowledged
- An effect exists / its stored result was acknowledged. Lost acknowledgements are not retroactively invented.
Only two business events and at most three envelopes are used. Delay = min(ceiling, base × 2attempt−1 + floor(seed-generated value × base / 4)). Events wait at most five attempts; blocked future events remain listed. This is a single local stream and an assumed atomic identity/effect boundary, not a database or transport implementation.
Parameters and traces stay in memory. There is no JSON import, persistent storage, authentication, API request or personal data. Export contains the model version, exact configuration, transitions and limitations; it is bounded to 64 KiB.
Read the ordering trade-offs and the six-scenario acceptance matrix before transferring any assumption to a real system.
Single-event parameter experiment
Execution trace
Choose parameters and run the simulation. No events have been generated.
[]Read the model before the result
The single-event parameter experiment is a deterministic, sequential simulation of one synthetic event. Duplicates are additional envelopes with the same event ID. Once mode ignores duplicate and retry settings. In at-least-once mode each envelope has at most one plus your retry count attempts.
The injected failure occurs before processing on the selected attempt of each envelope; it may be unreachable if an earlier attempt succeeds. Processing a new effect takes 400ms. A smaller timeout loses its acknowledgement even though the effect commits. With idempotency on, later attempts acknowledge the stored result in 50ms. Backoff is 250 × 2attempt−1 ms, without jitter.
Dead-letter means the acknowledgement budget was exhausted. It does not prove that the effect is absent. The trace clock advances virtually; no real waiting, network, storage or concurrent workers are involved. Parameters and results stay in memory. Copying uses your clipboard only when you press Copy JSON.
A worked example, even without JavaScript
Default parameters deliver two envelopes with idempotency enabled, no injected failure and a 1000ms timeout. The original commits at 400ms. The duplicate arrives at 500ms and is ignored at 550ms. Result: two attempts, one effect, one duplicate ignored.
| Virtual time | Event | Meaning |
|---|---|---|
| 0ms | received | Original envelope arrives. |
| 400ms | processed | One effect commits. |
| 500ms | received | A second envelope carries the same event. |
| 550ms | duplicate_ignored | Stored outcome acknowledged. |
Compare this model with a real transaction boundary and the experiment limitations.