bcbeidel/reversibility

<objective> Not all decisions carry equal risk. Two-way doors (easily reversible) should be made quickly with less analysis. One-way doors (hard to reverse) deserve careful deliberation. Classify the decision to calibrate the right amount of analysis. </objective> <process> 1. State the decision being considered 2. Assess: if this goes wrong, can we undo it? At what cost? 3. Classify as one-way door (irreversible) or two-way door (reversible) 4. If two-way: identify the rollback path and its cost 5. If one-way: identify what makes it irreversible and the blast radius 6. Calibrate analysis effort to the door type 7. If one-way, identify partial reversibility or ways to reduce commitment </process>

<output_format>

Reversibility Analysis: [Topic]

Decision

[What's being decided]

Classification: [One-Way Door / Two-Way Door]

Reversibility Assessment

• Can it be undone? [Yes/Partially/No]
• Cost to reverse: [Low/Medium/High/Impossible]
• Time to reverse: [Minutes/Days/Months/Never]
• Blast radius: [Self/Team/Organization/Public]

If Two-Way Door

• Rollback path: [How to undo]
• Recommendation: Decide quickly, course-correct later

If One-Way Door

• What makes it irreversible: [Specific factors]
• Ways to reduce commitment: [Pilot, phase, or partial approach]
• Recommendation: Analyze thoroughly before committing </output_format>

<example> ## Reversibility Analysis: Choosing a Database for the Events Service

Decision

Use PostgreSQL vs DynamoDB for a new event-sourcing service.

Classification: One-Way Door

Reversibility Assessment

• Can it be undone? Partially — data can be migrated, but schema design and query patterns are deeply coupled
• Cost to reverse: High — rewriting data access layer, migrating data, revalidating correctness
• Time to reverse: Months (data migration + regression testing + gradual cutover)
• Blast radius: Team — affects the events team and all downstream consumers of the event stream

If One-Way Door

• What makes it irreversible: Query patterns, schema design, and operational tooling all become database-specific within weeks. After 6 months of production data, migration becomes a project in itself.
• Ways to reduce commitment: Start with a repository abstraction layer so business logic doesn't call database APIs directly. Build for PostgreSQL first (team knows it), but keep the option to swap the storage backend if DynamoDB's scaling becomes necessary.
• Recommendation: Analyze thoroughly. Default to PostgreSQL (known quantity, team expertise, adequate for projected scale). Revisit DynamoDB only if event volume exceeds 50K/sec — a threshold we're unlikely to hit in year one. </example>

Key Instructions

• If a decision appears one-way but has a viable partial path (pilot, phased rollout, abstraction layer), always surface it — full irreversibility is rarer than it appears.

Anti-Pattern Guards

1. Treating all one-way doors as equally risky — blast radius matters; a one-way door affecting only you deserves less caution than one affecting the whole organization.
2. Using irreversibility as a reason to avoid deciding — one-way doors still require decisions; the output is appropriate deliberation level, not paralysis.

Handoff

Chainable to: opportunity-cost (to weigh the cost of commitment), second-order (to trace consequences of irreversible choices)