⚙️ Backend Engineering · Distributed Systems

Choose consistency models per data class deliberately

Strong vs bounded-staleness vs eventual mapped to business harm scenarios, not fashion.

advanced~45 minBackend EngineersAPI DevelopersPlatform Engineers

Steps

1List data classes with concrete stale-read harm examples in money terms
2Default everything eventual; upgrade ONLY classes failing the harm test
3For upgrades prefer session/bounded-staleness before paying strong-consistency latency
4Document chosen model per entity in architecture decision records
5Design compensations for eventual paths (idempotent corrections, reconciliation)
6Test partition behavior deliberately; verify degraded modes match documentation

Common Pitfalls

▲Strong consistency everywhere 'for safety', latency budgets destroyed
▲Eventual everywhere 'for scale', double-charges discovered by finance

Commands

Install with skills CLI

$ npx skills add aniruddhaadak80/skills --skill distributed-systems-consistency-model-selection

Install globally

$ npx skills add aniruddhaadak80/skills --skill distributed-systems-consistency-model-selection -g

Tags

#consistency#architecture#distributed#backend-engineering#distributed-systems

Install this skill

$ npx skills add aniruddhaadak80/skills --skill distributed-systems-consistency-model-selection -a opencode

SKILL.md

# Choose consistency models per data class deliberately

> Strong vs bounded-staleness vs eventual mapped to business harm scenarios, not fashion.

## When to Use This Skill

Strong vs bounded-staleness vs eventual mapped to business harm scenarios, not fashion.

## Steps

1. List data classes with concrete stale-read harm examples in money terms
2. Default everything eventual; upgrade ONLY classes failing the harm test
3. For upgrades prefer session/bounded-staleness before paying strong-consistency latency
4. Document chosen model per entity in architecture decision records
5. Design compensations for eventual paths (idempotent corrections, reconciliation)
6. Test partition behavior deliberately; verify degraded modes match documentation

## Common Pitfalls

- Strong consistency everywhere 'for safety', latency budgets destroyed
- Eventual everywhere 'for scale', double-charges discovered by finance

View source on GitHub ↗

Residency, replication lag, failover semantics, and write-conflict strategy before the second region exists.

⚙️ Backend Engineering·~50m

Open →

Adopt event sourcing without the graveyard mistakesadvanced

Command/event separation, upcasting discipline, and projection rebuild strategies that survive year three.

⚙️ Backend Engineering·~45m