Avoiding false precision
Linear extrapolation cannot predict step changes, storage cleanup, or workload migrations. Flat, noisy, stale, or short data produces a limitation instead of an exhaustion date. Resource pressure is observational and does not silently create another Cloud alert evaluator. A 100% observed endpoint success rate with 2% coverage is not a verified month of 100% availability. Missing minutes never enter the passing count.
Recovery layers
Encrypted artifact decryption and database integrity establish that a consistent database snapshot can be read. They do not establish that all dependencies, credentials, networking, or application behavior can be recovered. Scope and last tested level are therefore explicit, and absent restoration evidence reduces verified health. Receiving new exporter samples with an old backup timestamp does not refresh backup age. Where a schedule policy is not established in inventory, an age measurement alone earns no freshness credit.
Coverage accountability
Every canonical system has explicit expected service, metric, log, objective, and backup checks or an explained applicability exclusion. Unsupported sources stay unknown. Rules without verified system labels remain in a fleet-level assignment gap rather than inflating an arbitrary server score. The dashboard catalog continues to explain restricted panels and Cloud-only originals.
Workload evidence failure behavior
Collection is bounded to one hour, four MiB and five seconds per journal command. Invalid/future timestamps, malformed relevant timings, insufficient activity, changed PID or inaccessible logs emit no passing metric. Any observed non-2xx response or duration at least 220 seconds fails the check. Health GETs, another service, old PIDs and recovery processes are excluded. Collector age is capped at twenty minutes and source activity at fifteen minutes; the five-minute timer normally refreshes earlier. Raw journals, prompts and model responses are never published. Cloudflare log-query failure does not prevent independent local workload collection. Validate these negative paths and compare application PIDs after deployment; rollback the portal/collector source revisions together.