Senior craft research: corrected evidence snapshot

Reviewed 2026-09-22. This replaces the former “condensed from primary sources” summary: organization names and search excerpts were not claim-level citations. See the claim ledger for exact URLs, date basis, role/sample, source status and limits; withdrawn assertions remain explicit.

Retained mechanisms

Reliability claims C01–C03 support user-facing ratios, careful retry semantics and recovery feedback. The teaching arithmetic is constructed: a million eligible requests with ten thousand failures produces 99% success and consumes ten 99.9% request budgets, regardless of a short outage duration. Three total attempts at three layers gives 27 leaf attempts; three retries after the original gives 64. The runnable reliability lab verifies those countermodels and adds an unfamiliar incident with raw metrics.

Recent primary incident reports C04–C06 supply production case context. They document operator observations, not hiring frequency, universal safe settings, or independent causal proof. AWS service mappings, workloads, diagrams of proposed designs and acceptance thresholds are curriculum constructions.

For data/observability, C11–C13 support narrow version-specific semantics. Conditional object writes do not grant a multi-object transaction. PostgreSQL heap layout does not follow every primary-key choice. SDK/specification stability must be checked per component. Exact market shares, overhead guarantees and “default” platforms have been removed rather than promoted from untraceable notes.

Curriculum decisions, not research findings

We teach a design by naming the operation, invariant, workload and failure boundary; drawing a simple baseline; breaking it with a counterexample; then calculating the improved design's cost and recovery behavior. Caches, replicas and shards are possible choices after measuring the bottleneck, not an unconditional sequence every system must follow.

A senior practice answer should connect evidence to a reversible mitigation, an explicit budget and the next observation that could change the decision. Cross-team ownership and recovery authority form the lead extension. These are declared teaching standards, not statistical claims about a hiring market.