Local-first observability for finding recurring failures across agent traces, verifying the evidence, and creating reviewable regression evals.https://perseval.dev/https://perseval.dev/blog/from-first-failure-to-first-regression-suite/https://perseval.dev/blog/from-first-failure-to-first-regression-suite/Follow a sanitized mutating timeout from trace evidence through human review to an accepted eval definition—the point where Perseval stops today.Thu, 16 Jul 2026 09:15:00 GMThttps://perseval.dev/blog/why-one-failed-trace-is-not-an-eval/https://perseval.dev/blog/why-one-failed-trace-is-not-an-eval/A failed trace is evidence from one execution. An eval requires a representative case, expected behavior, and a repeatable way to judge new runs.Thu, 16 Jul 2026 09:14:00 GMThttps://perseval.dev/blog/production-agent-failures-to-regression-evals/https://perseval.dev/blog/production-agent-failures-to-regression-evals/A practical workflow for moving from recurring trace evidence to a reviewed eval definition, with the execution boundary kept explicit.Thu, 16 Jul 2026 09:13:00 GMThttps://perseval.dev/blog/how-to-compare-two-agent-runs/https://perseval.dev/blog/how-to-compare-two-agent-runs/A useful comparison preserves both run identities, aligns equivalent work, and highlights the first meaningful divergence.Thu, 16 Jul 2026 09:12:00 GMThttps://perseval.dev/blog/ground-truth-and-calibration-for-agent-evals/https://perseval.dev/blog/ground-truth-and-calibration-for-agent-evals/Ground truth records the best supported label for a case, while calibration measures whether graders and confidence estimates deserve trust.Thu, 16 Jul 2026 09:11:00 GMThttps://perseval.dev/blog/what-is-white-box-replay-for-ai-agents/https://perseval.dev/blog/what-is-white-box-replay-for-ai-agents/White-box replay reruns agent logic while controlling recorded model, tool, time, or state dependencies.Thu, 16 Jul 2026 09:11:00 GMThttps://perseval.dev/blog/black-box-gray-box-white-box-agent-testing/https://perseval.dev/blog/black-box-gray-box-white-box-agent-testing/The three modes differ by how much of the agent execution and its dependencies the evaluation system can observe or control.Thu, 16 Jul 2026 09:10:00 GMThttps://perseval.dev/blog/replay-reexecution-and-regrading/https://perseval.dev/blog/replay-reexecution-and-regrading/Regrading reuses evidence, re-execution invokes the agent again, and controlled replay fixes selected dependencies.Thu, 16 Jul 2026 09:09:00 GMThttps://perseval.dev/blog/what-is-llm-as-a-judge/https://perseval.dev/blog/what-is-llm-as-a-judge/LLM-as-a-judge uses a language model to evaluate an agent result against a rubric, with evidence, uncertainty, and provenance kept explicit.Thu, 16 Jul 2026 09:09:00 GMThttps://perseval.dev/blog/deterministic-graders-vs-llm-judges/https://perseval.dev/blog/deterministic-graders-vs-llm-judges/Use deterministic checks for facts you can state precisely, and LLM judges for bounded semantic questions that preserve evidence and uncertainty.Thu, 16 Jul 2026 09:08:00 GMThttps://perseval.dev/blog/what-is-an-eval-runner/https://perseval.dev/blog/what-is-an-eval-runner/An eval runner invokes the agent for each case, captures the result and trace, and hands that evidence to graders.Thu, 16 Jul 2026 09:07:00 GMThttps://perseval.dev/blog/eval-definitions-cases-suites-and-runs/https://perseval.dev/blog/eval-definitions-cases-suites-and-runs/A precise vocabulary makes agent evaluation reproducible and prevents a test definition from being confused with its execution.Thu, 16 Jul 2026 09:06:00 GMThttps://perseval.dev/blog/offline-online-and-shadow-agent-evals/https://perseval.dev/blog/offline-online-and-shadow-agent-evals/Offline evals test controlled cases, online evals judge production behavior, and shadow evals observe a candidate without affecting users.Thu, 16 Jul 2026 09:06:00 GMThttps://perseval.dev/blog/what-is-an-agent-eval/https://perseval.dev/blog/what-is-an-agent-eval/An agent eval is a repeatable test that combines an input, expected behavior, execution evidence, and a grader.Thu, 16 Jul 2026 09:05:00 GMThttps://perseval.dev/blog/logs-traces-and-evals-difference/https://perseval.dev/blog/logs-traces-and-evals-difference/Logs record events, traces connect an execution, and evals test behavior against an explicit expectation.Thu, 16 Jul 2026 09:04:00 GMThttps://perseval.dev/blog/what-is-opentelemetry-for-ai-agents/https://perseval.dev/blog/what-is-opentelemetry-for-ai-agents/OpenTelemetry provides a vendor-neutral way to create, transport, and interpret telemetry from agent runs.Thu, 16 Jul 2026 09:03:00 GMThttps://perseval.dev/blog/what-is-agent-observability/https://perseval.dev/blog/what-is-agent-observability/Agent observability connects individual executions to recurring behavior, evidence, and changes across builds.Thu, 16 Jul 2026 09:02:00 GMThttps://perseval.dev/blog/what-is-an-ai-agent-trace/https://perseval.dev/blog/what-is-an-ai-agent-trace/A trace records how an agent reached an answer, including model calls, tool use, handoffs, retries, and verification.Thu, 16 Jul 2026 09:01:00 GMThttps://perseval.dev/blog/deterministic-first-ai-when-useful/https://perseval.dev/blog/deterministic-first-ai-when-useful/Keep failure identity deterministic, use similarity only for navigation, and treat hosted judgments as unproven assistance.Thu, 16 Jul 2026 00:00:00 GMThttps://perseval.dev/blog/from-production-traces-to-regression-evals/https://perseval.dev/blog/from-production-traces-to-regression-evals/A practical loop for turning recurring agent failures into evidence-backed eval definitions.Thu, 16 Jul 2026 00:00:00 GMThttps://perseval.dev/blog/why-agent-trajectories-matter/https://perseval.dev/blog/why-agent-trajectories-matter/Outcome-only scoring misses the decisions, tool calls, and recoveries that make agent behavior reliable or fragile.Thu, 16 Jul 2026 00:00:00 GMT