The summary.json artifact provides a high-level view of episode outcomes, metrics, and flags. It’s designed for quick access to key information without parsing the full event timeline.
Schema overview { "schema_version" : "1.3.0" , "episode_id" : "ep_2024_abc123_s0" , "task" : "Draft release notes for v1.2.0" , "started_at" : "2024-01-15T10:30:00Z" , "duration_sec" : 5.0 , "adapter_result" : "success" , "outcome" : "success_unverified" , "verification" : { ... }, "flags" : { ... }, "metrics" : { ... }, "insight" : { ... }, "tags" : { ... } }
Root fields Version of the summary schema. Currently "1.3.0".
Unique episode identifier.
The original task or goal.
ISO 8601 start timestamp.
Total episode duration in seconds.
Adapter execution status: "success", "error", or "skipped".
Verification outcome: "success", "goal_not_achieved", "success_unverified", or "error".
Verification summary block. Show Verification properties
Whether verification assertions were provided.
Verification pass/fail status (null when unverified or unavailable).
Assertion results (empty when not provided).
Snapshot diff with added, modified, and deleted lists.
Snapshot paths for pre and post.
Snapshot policy (ignore list + symlink handling).
Verification error reason (e.g., "workspace_unavailable").
Flags Configuration flags for the episode run.
{ "flags" : { "intuition" : true , "mode" : "meta" , "direction" : { "applied" : 2 , "vetoed" : 0 , "policy" : "SafetyPolicy@1.0" , "threshold" : 0.75 , "last_diff" : [ "plan.steps[0].parameters" ] } } }
Whether intuition was enabled.
Execution mode: "meta" or "minimal".
Direction statistics (meta mode only). Show Direction properties
Number of directives applied.
Policy that issued directives.
Confidence threshold used.
Paths modified by the last directive.
Metrics Core execution metrics.
{ "metrics" : { "success" : 1 , "plan_count" : 2 , "act_count" : 3 , "reflect_count" : 1 , "veto_count" : 0 , "learn_proposals" : 1 , "learn_applied" : 0 , "latencies" : { "first_action_ms" : 150 , "total_ms" : 5000 , "planning_ms" : 500 , "execution_ms" : 4000 } } }
Success indicator: 1 for success, 0 for failure.
Number of planning iterations.
Number of actions executed.
Number of reflection passes.
Number of learning proposals generated.
Number of learning proposals applied.
Timing metrics. Milliseconds to first action.
Total episode duration in milliseconds.
Time spent in planning phase.
Time spent in execution phase.
Insight Advanced insight metrics (meta mode only; may be absent in minimal mode).
{ "insight" : { "metrics" : { "plan_adherence" : 0.95 , "tool_coverage" : 1.0 , "branching_factor" : 2 , "plan_revisions" : 1 } } }
Computed insight metrics. How closely execution matched the plan (0-1). Rounded to 4 decimal places.
Percentage of planned tools that were used (0-1).
Count of direction events emitted.
Number of times the plan was revised.
User-provided metadata.
{ "tags" : { "environment" : "staging" , "team" : "platform" , "priority" : "high" } }
Key-value pairs of user-provided metadata.
Complete example { "schema_version" : "1.3.0" , "episode_id" : "ep_2024_abc123_s0" , "task" : "Draft release notes for v1.2.0" , "started_at" : "2024-01-15T10:30:00Z" , "duration_sec" : 5.0 , "adapter_result" : "success" , "outcome" : "success_unverified" , "verification" : { "provided" : false , "passed" : null , "assertions" : [], "workspace_diff" : null , "snapshots" : null , "policy" : { "ignore" : [ ".git" , "__pycache__" , ".venv" , ".noesis" ], "symlinks" : "skip" } }, "flags" : { "intuition" : true , "mode" : "meta" , "direction" : { "applied" : 1 , "vetoed" : 0 , "policy" : "SafetyPolicy@1.0" , "threshold" : 0.75 , "last_diff" : [ "plan.steps[0].description" ] } }, "metrics" : { "success" : 1 , "plan_count" : 2 , "act_count" : 3 , "reflect_count" : 1 , "veto_count" : 0 , "learn_proposals" : 0 , "learn_applied" : 0 , "latencies" : { "first_action_ms" : 150 , "total_ms" : 5000 } }, "insight" : { "metrics" : { "plan_adherence" : 0.95 , "tool_coverage" : 1.0 , "branching_factor" : 2 , "plan_revisions" : 1 } }, "tags" : { "environment" : "staging" , "team" : "platform" } }
Reading summaries Python import noesis as ns summary = ns.summary.read( "ep_abc123" ) # Basic info print ( f "Task: { summary[ 'task' ] } " ) print ( f "Success: { summary[ 'metrics' ][ 'success' ] } " ) # Metrics metrics = summary[ 'metrics' ] print ( f "Actions: { metrics[ 'act_count' ] } " ) print ( f "Vetoes: { metrics[ 'veto_count' ] } " ) print ( f "First action: { metrics[ 'latencies' ][ 'first_action_ms' ] } ms" ) # Insight (if available) insight = summary.get( 'insight' , {}).get( 'metrics' , {}) print ( f "Plan adherence: { insight.get( 'plan_adherence' , 'N/A' ) } " )
CLI # Inspect episode dashboard noesis view ep_abc123 # Resolve artifact directory from JSON output EP_DIR = $( noesis view ep_abc123 --json | jq -r '.episode_dir' ) # Read summary fields directly jq '.metrics.success' " $EP_DIR /summary.json"
File access cat .noesis/episodes/ep_abc123/summary.json | jq .
Use cases Check success rate import noesis as ns episodes = ns.list_runs( limit = 100 ) successes = sum ( ns.summary.read(ep[ 'episode_id' ])[ 'metrics' ][ 'success' ] for ep in episodes ) rate = successes / len (episodes) print ( f "Success rate: { rate :.2%} " )
Find vetoed episodes import noesis as ns episodes = ns.list_runs( limit = 100 ) vetoed = [ ep for ep in episodes if ns.summary.read(ep[ 'episode_id' ])[ 'metrics' ][ 'veto_count' ] > 0 ] print ( f "Vetoed episodes: { len (vetoed) } " )
Analyze latencies import noesis as ns import statistics episodes = ns.list_runs( limit = 100 ) latencies = [ ns.summary.read(ep[ 'episode_id' ])[ 'metrics' ][ 'latencies' ].get( 'first_action_ms' , 0 ) for ep in episodes ] print ( f "Median first action: { statistics.median(latencies) :.0f} ms" ) print ( f "P95 first action: { statistics.quantiles(latencies, n = 20 )[ - 1 ] :.0f} ms" )
Next steps
State schema State artifact reference.
Export metrics Export to observability tools.
