Feature flag retirement: remove the old behavior safely
A proposed retirement memo checks five deployed-consumer cohorts, missing-flag behavior and rollback builds before temporary configuration is removed.
Blog
Find a check for the system you are building
Explore SaaS engineering through runnable examples and decision worksheets. Browse by topic or search all articles.
Data migration: equal row counts can hide different data
Six target fixtures all match the source row count, but only two reconcile by key and value. Run the comparison and inspect every discrepancy list.
CSV export: separate cell quoting from formula safety
Run 12 harmless CSV cases: five accepted values round-trip unchanged and seven are rejected. Inspect the bounded text and numeric policies and their limits.
API deprecation: make customer migration verifiable
A proposed 5-cohort matrix separates API migration from missing traffic, delivered notices and unknown callers. Define the evidence each consumer needs.
Why a redriven message returns to the dead-letter queue
Four executed event fixtures show why a schema fix improves redrive without repairing malformed messages. Inspect 16 attempts and cohort review decisions.
Saga compensation: define the state you can reach
A proposed booking recovery matrix separates released capacity, payment uncertainty and irreversible effects. Includes five failure drills awaiting evidence.
Why averaging p95 gives the wrong service latency
An executed 110-request fixture gives p95 of 1000 ms, while averaging instance p95 gives 505 ms. Reproduce the ranks and five traffic sensitivity cases.
Metric cardinality: count series before adding labels
Reproduce 144 versus 576 label identities, compare an unrestricted upper bound and show how replica turnover doubles the union across two windows.
SaaS usage metering: make corrections explainable
A six-delivery ledger fixture separates duplicate events, late usage and linked corrections. Inspect period totals before defining the billing integration.
Verify webhook signatures before parsing the body
Run eight harmless signature cases to separate body integrity, timestamp policy and duplicate effects. Includes Python fixtures and a verification worksheet.
Why a saved change disappears on the next read
An executed 16-read fixture shows why a 2-second primary window still returns stale data. Compare routing decisions and inspect the diagnostic worksheet.
Build a file upload workflow with a quarantine boundary
Run six upload-state checks that reject obsolete scan results, reset approval after replacement and tie retrieval to the owning tenant and exact bytes.
Why a deleted record returns to your search index
Replay revision 10 after deletion at 11. An executed Python model compares physical deletion, retained markers and pruning, with a complete 15-row trace.
Secret rotation: prove consumers switched credentials
An executed 24-probe model compares early revocation with consumer adoption. Use the trace and handoff worksheet to distinguish refresh from authentication.
Audit log design: record the decision without the payload
Review a proposed 13-field role-change audit record, its trusted sources and commit boundary. Includes a field-purpose map and access-review worksheet.
Error budget burn rate: choose the right denominator
Four synthetic intervals produce burn rates of 1.099 or 50.5 depending on weighting. Inspect the counts and low-traffic sensitivity before tuning alerts.
SaaS entitlements: define what a subscription change does
Turn upgrades, scheduled downgrades and renewal grace into a seven-state policy matrix with effective times, project limits and rules for accepted work.
Prevent lost updates with ETags and If-Match
Run a two-client SQLite example that exposes an unsafe write, rejects a stale revision and checks weak tags, missing preconditions and wildcard behavior.