SPECIMEN 01 / duplicate-write

The double-tap write

One intent. Two commits.

A retry of the same key and body commits again, even though the first result is still cached.

PUT THE TWO POLICIES SIDE BY SIDE

Same requests. Different memory.

LOCAL MODEL REPLAY

Walk through 4 authored operations. The first visible difference appears at operation 3. Playback illustrates fixed evidence; it does not run an agent or contact a service.

The complete evidence

Always available below, with or without playback.

The double-tap write: fixed reference and faulty observations. Write counts are cumulative; a dash means no receipt reached the caller.
Operation	Request	Delivery	Correct reference	Faulty policy
01	Tick 0key B · value B	Received	201receipt #1 · writes 1	201receipt #1 · writes 1
02	Tick 1key A · value A	Received	201receipt #2 · writes 2	201receipt #2 · writes 2
03	Tick 2key A · value A	Received	200receipt #2 · writes 2	201DIFFERSreceipt #3 · writes 3
04	Tick 3key B · value A	Received	409receipt — · writes 2	409DIFFERSreceipt — · writes 3

WHAT MAKES THIS ONE TICK

A small failure, unpacked.

Operation 2 creates A. Operation 3 repeats the same intent before expiry. The reference returns receipt 2 without a write; the duplicate-write model creates receipt 3. The independent B operations give you removable context to investigate.

AWS describes why retryable APIs need to avoid extra side effects and why a caller-provided request identifier makes intent explicit. This exhibit turns that general concern into a deliberately small model; its status codes and six-tick lifetime belong to our contract.

TAKEAWAYCount durable writes as well as responses: a plausible reply can hide a duplicate side effect.

Read the foundations

These sources explain the underlying retry and idempotency concepts. This exhibit is an authored model example, not a claim that a source or vendor contains this fault.

AWS Builders’ Library: Making retries safe with idempotent APIs ↗

SPECIMEN LABEL / 01

Take it apart.
Take it home.

Evidence: Bundled model comparison
Operations: 4 finite synthetic steps
Fault: duplicate-write
Boundaries: Keys A/B · values A/B · 6-tick expiry
Provenance: Authored learning fixture

Standalone Node test ↓Candidate JSON ↓Markdown explanation ↓

node --test duplicate-write.test.mjs

Runs the bundled models locally by default, with no account or network request. To check your implementation, use BUG_ZOO_ADAPTER and the local adapter instructions. Your selected adapter runs with your permissions. The seven-case test kit covers more of the contract. Download asset license: MIT.

THE NEXT MOVE IS YOURS

Can you make it smaller?

This fixture includes removable work. Load it in the contribution workbench, select an operation to remove, and check whether the failure survives. Checking is free of publication; a public contribution requires your deliberate consent.

Remix this specimen ↗Build on an accepted result ↗

An exhibit has no parent receipt. Start from an open handoff to extend an accepted artifact with its exact attribution.

A different failure or interpretation?Bring your own synthetic case ↗Discuss the policy ↗Continue saved work ↗

MEET THE NEXT SPECIMENThe identity mix-up