Quality signals and review

Use standalone Evolver asset and recall reports to decide whether to continue a pilot, gather more evidence, or stop adoption. Tie each conclusion to the asset, environment, and observation time.

Asset quality must come from observable execution and reuse outcomes. Model-reported confidence, download counts, or one success provide investigation clues, but none should decide promotion or team adoption alone.

Inspect current signals

evolver asset-health
evolver reuse-report
evolver recall-verify-report
evolver memory-graph status

Evidence hierarchy

recall-verify-report provides a GREEN, YELLOW, or RED gate together with success, missing and mismatch rates, errors, latency, and age. The gate is a review entry point, not an auto-publish button.

Establish what each report answers

Asset health, recall verification, and memory-graph status measure different things. Asset reports locate issues in experience records; recall verification describes retained selection and validation outcomes; graph status describes operational health of local relationship data. A healthy graph does not establish applicability, and one validated asset does not establish that every recall path works.

Identify the environment, retained-data window, and coverage of the intended assets and uses. Reports with different scopes cannot establish improvement or regression through color or success rate alone. Compare outcomes after aligning environment, version, and sample scope.

Scenario: GREEN does not approve a new environment

Experience receives GREEN in an existing project, but the team wants to use it with upgraded dependencies. The report supports its original samples; validate behavior in the new environment. Preserve the original report and add new inputs, validation commands, and outcomes rather than turning prior GREEN into a cross-environment guarantee.

For RED, inspect the cause. No retained samples indicate insufficient evidence; mismatches and failures require investigation of specific problems. Neither supports promotion, but the responses differ: collect validation evidence for the former, and repair content, matching, or environment before retesting the latter. Gate color directs attention; the review record explains the decision.

Administrative action

• GREEN: confirm target environment, impact boundary, and review requirements before promotion.
• YELLOW: collect more reproduction or real reuse outcomes and keep the asset in a pilot scope.
• RED: stop automatic recall or promotion and inspect the asset, index, version, and environment match.

Review cadence

Revalidate high-impact assets after dependency upgrades, platform changes, repeated failures, or long periods without reuse. Reports must retain failures; success-only data produces a misleading quality picture.

Turn a report into an adoption decision

Save the report time, asset scope, version, and environment, then check that samples cover the intended use. The report covers bounded outcomes retained in proxy state, not all-time history. With no recall-verification records, recall-verify-report returns RED; no data is not evidence of no failures. GREEN exits with code 0; YELLOW and RED exit with code 2. Collect the report body so automation can distinguish an unmet quality gate from failure to run the command.

Record reasons for failures or skipped checks, the owner of additional validation, and the next review time. Act on RED through runtime policy, trust decisions, and task handling; the report does not stop all execution for you. Close the review only when evidence covers the target scope, issues are addressed, and the adoption decision is recorded.

Related pages

• Reproducibility
• Memory graph and asset health

EvoX Docs · Developers · Experience asset management