tarne75/sorethumb

How tarne75/sorethumb works

sorethumb is a local-first tabular anomaly ranker and review shortlist generator. The principal workflow is CLI or Python invocation → source loading → column profiling and feature preparation → detector ensemble scoring → original-column explanations → persisted run history and reports. The pipeline module is documented as the sole subsystem sequencer; source excerpts confirm several internal responsibilities, but not every call site, so uncertain links are omitted. The graph also preserves the documented score-forward and resumable-history capabilities.

Main entry points: _pipeline.py Run pipeline · cli.py CLI · source.py Source resolution · readers.py Format readers

Components 22

Entry and orchestration

CLI src/sorethumb/cli.py
Configuration src/sorethumb/config.py
Run pipeline src/sorethumb/_pipeline.py

Data preparation

Source resolution src/sorethumb/io/source.py
Format readers src/sorethumb/io/readers.py
Column profiling src/sorethumb/profiling/profile.py
Feature planning src/sorethumb/profiling/plan.py
Feature construction src/sorethumb/features/build.py

Detection and explanations

Detector suite src/sorethumb/detectors
Score calibration src/sorethumb/scoring/calibrate.py
Ensemble ranking src/sorethumb/scoring/combine.py
Record explanations src/sorethumb/explain

History and reporting

SQLite store src/sorethumb/store/db.py
Models and results src/sorethumb/store/models.py
Run ledger src/sorethumb/history/ledger.py
Period backfill src/sorethumb/history/periods.py
History contrast src/sorethumb/analysis/contrast.py
Reports src/sorethumb/report

Other components

Analyst
Dataset
HTTP source
Ranked results

Connections 25

Analyst → CLI: invokes
CLI → Configuration: loads settings
CLI → Run pipeline: runs or scores
Analyst → Run pipeline: calls API
Run pipeline → Source resolution: resolves source
Source resolution → HTTP source: downloads
Dataset → Source resolution: provides URI
Run pipeline → Format readers: reads data
Format readers → Dataset: loads records
Run pipeline → Column profiling: profiles columns
Run pipeline → Feature planning: plans treatments
Run pipeline → Feature construction: builds features
Run pipeline → Detector suite: fits and scores
Run pipeline → Score calibration: calibrates scores
Run pipeline → Ensemble ranking: combines scores
Run pipeline → Record explanations: explains records
Run pipeline → Models and results: persists artifacts
Run pipeline → SQLite store: records run
Run pipeline → Run ledger: tracks completion
Run ledger → SQLite store: reads and writes
Run pipeline → Period backfill: selects periods
Run pipeline → History contrast: compares history
Run pipeline → Reports: renders report
Reports → Ranked results: writes report
Run pipeline → Ranked results: returns ranking

More Python diagrams: promtengineer/localgpt · tornadoweb/tornado · airbytehq/airbyte · xiaomi/ha_xiaomi_home · mkdocs/mkdocs

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·