test-evidencelisted
Install: claude install-skill lloydbrian/lloydbriantech-ascent
# test-evidence
The deep load-test and mutation check behind [quality-review](../quality-review/SKILL.md). It is a sibling that shares that skill's severity table, contract and evidence rules. It answers two questions: would a load test fail if the service missed its latency or throughput target, and is there evidence that the tests fail when the code is broken?
Source: [test-strategy.md](../../references/test-strategy.md#load-tests-and-mutation-evidence) (the tools, how thresholds and rates are read, report formats and dates).
## Rules
- All `quality-review` rules apply: read-only; repository-relative `file:line` evidence; fresh evidence; severities only from the `quality-review` table; no personal data.
- **Never run a load test or a mutation tool.** A load test reaches a live environment, and a mutation run rewrites the code under test and can take hours. Both are the operator's; write the command for them when a run would settle a question, and read the redacted output they paste.
- **Targets are read, never copied.** `p95_latency_ms` and `peak_rps` come from `architecture.quality_targets` (set by `ascent-architecture`), `min_score` and `max_age_days` from `quality.mutation`. Never write a target into the `quality:` section. With no latency or throughput target, the load lens is unmeasured.
- **A NOTE is never a pass.** A load set in virtual users has no request rate the script can read: open the script and say what rate it reaches, or that it cannot be told.
## Step