Why do I only find out I tagged a metric wrong when the invoice arrives?
機会
Tagging a metric with a high-cardinality attribute like a user ID, session ID, or request path multiplies a single time-series into hundreds of millions, and the bill follows a week later rather than a CI failure follows a minute later. By 2026, observability spend has reached 15 to 25 percent of total cloud spend at affected SaaS companies, with documented single-service incidents generating $90,000 in charges over a weekend. Every major backend, including Datadog, Grafana Cloud, and Prometheus, detects cardinality explosions only after ingestion, meaning the only signal is a budget alert or an invoice line item, not a pull request annotation. The OpenTelemetry SDK added optional cardinality caps in early 2026, but these are runtime circuit-breakers that silently drop data rather than upstream gates that catch the problem before it ships. There is no widely adopted CI primitive that ana
重要な理由
Every engineer who has tagged a metric with a user ID has had a bad Tuesday; catching it at the pull request instead of on the invoice permanently changes the economics of instrumentation.
機会をどう評価するか
Opportunity Scoreは測定値ではなく、私自身の見解です。どれほど痛みを伴うか、どれほど頻繁に影響を与えるか、そして今日時点で解決策がいかに少ないか。スコアが高いほど、構築する価値が高いと私は考えています。
それが現れたときにどれほどの痛みをもたらすか。
実際にどれほど頻繁に人々がそれに直面するか。
今日時点で、それに対する優れたツールがいかに少ないか。