No-Code SaaS Automation Platform · View 17 of 21 · Operations
The finding
- The Silence row is the one that matters most and is the easiest to omit. Every other row fires when something goes wrong; this one fires when nothing happens at all, which is what a dead subscription, a stuck cursor and a paused automation all look like.
- Correctness is a monitored signal family, not a test-suite property: duplicate effects, double admissions and dedup gaps are measured continuously in production because they cannot be reproduced in staging.
Targets on this view
- Push acknowledgement p99 ≤ 250 ms; start latency p95 ≤ 3 s.
- Outbound 429 rate ≤ 0.1% — the egress governor's own SLO, and the number that tells us whether we are a good citizen.
- Unclassified share of author-facing failures ≤ 2%; above that, the failure taxonomy itself is the defect.
- ≤ $0.55 per 100,000 step attempts; poll infrastructure ≤ 25% of platform compute.
Deliberately omitted
- Per-workspace dashboards are not an operator surface; the author-facing column is the product's run history and repair inbox, which is a different consumer with a different vocabulary.
