bench_agentic

{“what”:“UUID primary key; unique identifier for each run row.”,“why”:“Ensures each run has a stable, unique reference for joins and updates.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.505506707+00:00”}

{“what”:“UUID grouping multiple attempts or iterations of the same benchmark scenario.”,“why”:“Distinguishes repeated runs of the same task to analyze variance and retries.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.521018223+00:00”}

{“what”:“Name of the model provider (e.g., OpenAI, Anthropic) that served the request.”,“why”:“Needed to compare performance and error rates across different vendors.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.531614331+00:00”}

{“what”:“Model identifier as requested by the caller (e.g., gpt-4o).”,“why”:“Records the intended model to detect mismatches with what was actually served.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.542351663+00:00”}

{“what”:“Actual model identifier that handled the request; null if unknown or not provided.”,“why”:“Reveals routing or fallback behavior when the requested model is unavailable.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.577859136+00:00”}

{“what”:“Maximum token count the model context can hold; null if not reported.”,“why”:“Helps interpret failures or performance issues related to context limits.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.592597881+00:00”}

{“what”:“Expected number of tool-call steps in the benchmark task.”,“why”:“Serves as the ground truth to measure how many steps the agent actually took.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.607335142+00:00”}

{“what”:“Count of tool calls that matched the expected schema and arguments.”,“why”:“Measures correctness of tool usage; low values indicate parsing or reasoning failures.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.617643169+00:00”}

{“what”:“Number of tool calls that failed schema validation or had malformed arguments.”,“why”:“Quantifies how often the model produces unusable tool calls, a key quality metric.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.627756323+00:00”}

{“what”:“Boolean flag: true if the model returned plain text when a tool call was expected.”,“why”:“Identifies a common failure mode where the model ignores the tool-use instruction.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.641411412+00:00”}

{“what”:“Boolean flag: true if the agent finished the task within the allowed steps.”,“why”:“Distinguishes successful completions from timeouts or early termination.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.653715504+00:00”}

{“what”:“Boolean flag: true if the final submitted answer matched the expected result.”,“why”:“Measures task success beyond just completing steps; false indicates wrong final output.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.672550838+00:00”}

{“what”:“Total number of steps (tool calls or actions) the agent took in the run.”,“why”:“Shows efficiency; comparing to target_depth reveals over- or under-iteration.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.732955104+00:00”}

{“what”:“Total wall-clock time in milliseconds for the entire run.”,“why”:“Primary latency metric; needed to compare speed across models and providers.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.744054805+00:00”}

{“what”:“Average time in milliseconds per step; null if steps is zero.”,“why”:“Normalizes total time by step count to reveal per-step performance differences.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.760958826+00:00”}

{“what”:“Longest single step duration in milliseconds; null if no steps.”,“why”:“Highlights worst-case latency spikes that could cause timeouts or user-perceived delays.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.773493600+00:00”}

{“what”:“Integer status code (e.g., 0=success, 1=error, 2=timeout) for the run.”,“why”:“Provides a machine-readable summary of run outcome for filtering and alerting.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.783825609+00:00”}

{“what”:“Error message text if the run failed; null on success.”,“why”:“Stores the failure reason for debugging and categorizing common issues.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.794687540+00:00”}

{“what”:“Timestamp (with time zone) when the run record was inserted; defaults to now().”,“why”:“Enables time-based analysis, such as tracking performance changes over releases.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T04:03:57.891653943+00:00”}