# Nonlinear Cost of Incremental Reliability Each additional nine of availability may cost 100x more than the previous one. This is the fundamental economic argument for "reliable enough, but no more." ## The Math For a $1M revenue service: - 99.9% → 99.99% improvement = 0.09% more availability = $900 incremental revenue - If engineering cost to achieve that nine > $900, it's not worth it ## Two Cost Dimensions 1. **Redundant resources** — spare machines, parity codes, capacity buffers 2. **Opportunity cost** — engineers building resilience instead of features users want ## The ISP Noise Floor Typical ISP background error rate: 0.01% to 1%. If you drive service errors below this floor, the improvement disappears into the noise of the user's own infrastructure (phone, WiFi, ISP). You're paying for reliability no one can perceive. ## Strategic Implication This cost curve means every reliability decision is a resource allocation question, not a perfectionism question. The right question is never "how reliable can we make this?" — it's "what reliability level maximizes total business value?" ## Cross-Domain Connections - [[Risk as a Reliability Continuum]] — the framing that makes this cost trade-off explicit - [[Error Budget Model]] — the mechanism for managing the trade-off quarter by quarter --- *Source: Site Reliability Engineering, Chapter 3 (Alvidrez, 2016)* *Extracted: 2026-03-26*