**Prioritization Framework:**

Impact = Speedup × Frequency

Example:

- Query A: 10x speedup, 1000 calls/day = 10,000 impact

- Query B: 100x speedup, 10 calls/day = 1,000 impact

- Fix Query A first (higher total impact)

**Effort Estimation:**

Low effort (minutes):

- Add simple index

- Update statistics

- Adjust work_mem

Medium effort (hours):

- Add covering index

- Rewrite query

- Add materialized view

High effort (days):

- Partition table

- Redesign schema

- Add caching layer

**Risk Assessment:**

Low risk:

- Add index (can drop if not helpful)

- Update statistics (reversible)

- Increase work_mem for session

Medium risk:

- Rewrite query (must validate results)

- Change configuration (affects all queries)

High risk:

- Schema changes (requires migration)

- Partitioning (complex, hard to reverse)

**Measurement Best Practices:**

1. Baseline: Measure before optimization

2. Controlled: Change one thing at a time

3. Validated: Verify results are correct

4. Sustained: Monitor for regressions

5. Documented: Record what, why, results

pg_stat_statements shows total time (execution × calls) - prioritize high total time

EXPLAIN ANALYZE red flags: Seq Scan on large table, estimated ≪ actual, Disk usage

Missing index: 10-1000x speedup, low effort, high impact

Stale statistics: 3-50x speedup, ANALYZE takes seconds