Tutorials
Four DSSG-style case studies: the Dirty Duckling smoke test, the full DirtyDuck walkthrough (both problem framings), Chicago 311 with the production surfaces, and DonorsChoose’s deep feature synthesis. Pick your tutorial →
triage-pg
PostgreSQL-native temporal machine learning for public-policy data — cohorts, point-in-time features, append-only predictions, in-database evaluation. A deliberately simplified fork of DSSG triage.
Where to go
Section titled “Where to go”Onboarding one-pager
The whole system on one page: pipeline, schema, capabilities, quickstart. Read it →
Coming from DSSG triage?
The honest dimension-by-dimension comparison — what was kept, moved into PostgreSQL, or deliberately dropped. See the side-by-side →
Run it now
docker pull ghcr.io/ccd-ia/triage-pg:v1.1.4 — the CLI and dashboard,
no local Python needed. Point it at any PostgreSQL 11+.
The problem space
Two orthogonal axes: four problem types (what the model predicts) and three observation regimes (who gets a label and why) — exhaustively, with no combinatorial matrix. Map the space →
Technical architecture
The two-tier database design, the full results-schema ERD, the pipeline’s derivation DAG, and how the cloud profile maps onto AWS. Read the architecture →
Dashboard tour
Every surface, screenshot by screenshot — what question each panel answers and which table feeds it. Take the tour →
CLI tour
The headless-complete command line, workflow by workflow, with real output. Read the CLI tour →