SupernovaAgentic Workflow Analysis and Optimization
Last import · completed

Import

Bring a dataset in and run analysis.

  1. Choose a source and preview

    Point Supernova at a real S3 prefix using bucket, prefix, region, and a bounded recent-session limit.

    Dataset name

  2. Check the preview

    completed

    Metric scope

    Full population, preview sample, and full-run sample are separate

    Full source pool996 sessions996 discovered source objects; projections use this population.
    Preview sample100 threadsFast preview readout. This is capped at 100 threads regardless of session limit.
    Full analysis sample300 threadsThreads selected for the full pipeline after Start import; this is still a sample of the source pool.
    LLM deep dive target100 threadsWide metrics target 120 lightweight records.
    Preview objects read100Raw objects inspected in the fast preview sample only.
    Full source pool996996 objects discovered for population estimates.
    Preview threads100Preview is capped; full analysis sample is 300 threads.
    Coverage window0114-01-01 to 2025-12-17Workflow confidence · low
    Coverage qualitymediumMapping confidence · high

    Workflow readout

    What kinds of work are in this slice?

    low

    Derived from all 100 selected threads using user-prompt intent clustering. Major workflow types cover 83% of the slice; the long tail is grouped into Other. Claude consolidated the final major types.

    Flight Reservation Modification33 threads
    33% of slice
    Flight Cancellation Refund26 threads
    26% of slice
    Flight Booking11 threads
    11% of slice
    Flight Delay Complaint8 threads
    8% of slice
    Insurance Refund5 threads
    5% of slice
    Other17 threads
    17% of slice
    Local folderTau2-Airline-customer-service-v2
    Technical profiling details

    Preview plan

    generic records · complete

    Preview reads 300 threads, full import reads 300 sampled threads, and dashboard projections use 996 source-pool sessions.

    Source profile

    45 shapes · 4 id · 4 time

    Derived from all 100 selected objects in the current preview. Path patterns: 100. Top shapes: createdAt, id, role, segments (3878) | id, inputText, kind, label, metrics, model, +5 more (1854) | inputTokens, latencyMs, outputTokens (1854). Relationship signals: workflowKey, workflowLabel.

    Inferred entities

    trace[] -> content

    Spec thread-map-0687d07bb9 is validated with high confidence. Record path $; transcript path trace. Cache hit validated; model review not required cached. 100% row acceptance, 100% transcript coverage, 100% outcome coverage.

    File groups

    1 included · 1 skipped

    data/train-00000-of-00001.parquet: included, 996 rows | README.md: docs, 0 rows

    Validation gate

    validated · high

    500/500 sampled rows accepted. Transcript 100%, timestamps 100%, tool calls 99%, outcomes 100%.

    Identity signals

    id, toolUseId, task_id, payment_id

    Repeated identifiers that look like thread, session, account, or record keys.

    Time signals

    createdAt, date, endedAt, savedAt

    Timestamp-like fields detected while scanning the raw objects.

    Profiler notes (12)
    • Profiled 1 supported data file under Tau2-Airline-customer-service-v2 1 file passed the mapping gate.
    • 996 validated records were wrapped into generic thread bundles for preview/import.
    • Mapping spec thread-map-0687d07bb9 uses trace as the transcript path and reused a validated cached spec.
    • Import selected 300 inferred records from a recency-decayed slice.
    • Newer records are weighted more heavily, with deterministic decay into older records instead of taking only the latest 300.
    • Every supported data file belonged to the selected primary record group.
    • No parse errors were encountered while profiling generic files.
    • Mapping spec thread-map-0687d07bb9 uses trace as the transcript path and validated a new spec.
    • Preview sampled 100 inferred records from a recency-decayed 300-record import slice.
    • Raw source objects were profiled before thread assembly and workflow derivation.
    • Workflow clusters are derived from the imported slice rather than from source-specific label tables.
    • Large tool outputs may still require structural token proxies when source-native accounting is missing.

  3. Pipeline

    • QueueDone

      Add the job to the pipeline.

    • DownloadDone

      Copy raw objects locally.

    • NormalizeDone

      Build canonical records from the raw data.

    • AnalyzeDone

      Find optimization opportunities.

    • PublishDone

      Make the dashboard available.

  4. Dataset ready

    Completed
    Sessions
    300
    Operations
    10,082
    Workflow groups
    7
    Findings
    14
    Security signals
    0
    Open dashboard

History

Recent imports

1 jobs
ImportUpdatedStatusActions
Tau2-Airline-customer-service-v2Local archive · imp-60f8fc3dUpdated4/24/2026, 11:45:30 PM
Completed100%
Dashboard ready for inspection.Dashboard