Skip to content

tycoon data history

Terminal view over tycoon's observability DB. Every dlt ingest and every dbt invocation captures into .tycoon/metadata.duckdb; history is how you read it back without writing SQL.

Synopsis

tycoon data history [OPTIONS]
tycoon data history show <id> [OPTIONS]

Options (history):
  -t, --tool TEXT    Filter to dlt or dbt (default: all)
  -n, --limit INTEGER  Number of recent runs to show (default: 20)
  -s, --source TEXT  Filter dlt runs to one source (config name or schema literal)
  -l, --layer TEXT   Filter dbt invocations to those touching the named layer
  -h, --help         Show this message and exit

history — recent runs

tycoon data history                       # last 20 across dlt + dbt
tycoon data history --tool dlt --limit 25
tycoon data history --tool dbt
tycoon data history --source pokeapi      # only dlt runs for this source
tycoon data history --layer mart          # only dbt invocations that touched marts

--source and --layer are mutually exclusive (you can't ask for a layer of dlt runs — dlt loads aren't classified into layers). --layer requires a compiled dbt manifest; without one, the command exits with a hint pointing at tycoon data transform run.

Valid layer names: source, staging, intermediate, mart, snapshot, seed, unclassified. See the layered architecture recipe for what each layer means and how tycoon classifies your models into them.

Output:

                            Recent runs
┏━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━┓
┃ Tool     ┃ ID                 ┃ Started             ┃ Status ┃ Detail ┃
┡━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━┩
│ dlt      │ deadbeef-cafe-...  │ 2026-04-29 09:15:12 │ ok     │ ...    │
│ dbt      │ inv-abcdefgh-...   │ 2026-04-29 09:18:41 │ ok     │ ...    │
└──────────┴────────────────────┴─────────────────────┴────────┴────────┘

history is a quick "did the pipeline actually run, and when" check. For details on a specific run, use show.

show <id> — drill into one run

tycoon data history show deadbeef           # short prefix is fine
tycoon data history show deadbeef-cafe-001
tycoon data history show inv-abcdefgh

Auto-detects whether the ID is a dlt load or a dbt invocation by ID format. Short prefixes are accepted as long as they're unambiguous; ambiguous prefixes error with the candidate list.

dlt drilldown

For a dlt load, show prints:

  • Pipeline metadata: pipeline_name, transaction_id, started_at, finished_at, duration
  • Total bytes written across all jobs
  • Steps table: extract / normalize / load timings
  • Tables table: per-table row count + bytes column
  • Engine version + success/exception flag

The byte counts and per-step timings come from dlt's trace.pickle, captured by v0.1.3's observability v2a. Captures are best-effort — a missing trace file falls back to the row-count-only view.

dbt drilldown

For a dbt invocation, show prints:

  • Invocation metadata: command, target, dbt version, duration, success
  • Counts: models built, model errors, tests passed, tests failed
  • Nodes table: one row per model/test with status + exec time + rows affected
  • Schema changes vs. previous run table (if any) — model adds/removes, SQL hash changes, column adds/removes/type changes

The schema-changes table comes from v0.1.3's observability v2b. It's only populated when there's a previous snapshot to diff against — the first invocation in a project's history shows nothing here.

What feeds the history

Captured from Tables in metadata.duckdb
tycoon data sources run dlt_runs, dlt_rows_by_table, dlt_trace_runs, dlt_trace_steps, dlt_trace_jobs
tycoon data transform run/test/build dbt_runs, dbt_nodes, dbt_manifest_snapshots, dbt_schema_changes

tycoon run dbt ... (passthrough) is deliberately not captured — it would fire on every ad-hoc dbt show / dbt compile, polluting the history. Only tycoon data transform <cmd> triggers a capture.

Why isn't dashboards just the same as history?

The _tycoon_dlt_usage and _tycoon_dbt_usage Rill dashboards read the same data. They're better for trend lines (success rate over time, average duration); history is better for "what happened in this specific run." Both surfaces stay current — every capture also re-exports the Parquet files Rill reads.

Reset

tycoon data clean --metadata          # wipe metadata.duckdb

Run history is recreated on the next ingest / dbt run. The metadata DB is fully disposable.