Examples & POC catalog

The playground ships 101 self-contained POCs across 8 categories. A POC here is a small proof-of-concept project that demonstrates one Rocky capability end to end. Each has its own rocky.toml, a models/ directory, and a single run.sh that runs the whole thing. Most run on local DuckDB with no credentials, so you can read the code and run it in one step.

The Quickstart and Playground walkthroughs teach the workflow. The POCs show one capability each, in isolation.

Run one

git clone https://github.com/rocky-data/rocky.git
cd rocky/examples/playground

# Run any POC end-to-end (DuckDB, no credentials)
./pocs/02-performance/01-incremental-watermark/run.sh

Start with these

The demos from the README, each a single run.sh:

What it shows	POC
Schema drift detected and recovered on the next run	02-performance/06-schema-drift-recover
Data contracts enforced at compile time (E010 / E013)	01-quality/01-data-contracts-strict
Named branches, content-addressed run records, and column lineage	00-foundations/06-branches-replay-lineage
Column-level lineage on a branching DAG	06-developer-experience/01-lineage-column-level
AI model generation with the compile-verify loop	03-ai/01-model-generation
PR-time blast radius with rocky lineage-diff	06-developer-experience/11-lineage-diff
Classify columns, mask by environment, gate CI	04-governance/05-classification-masking-compliance
Incremental loads with persistent watermark state	02-performance/01-incremental-watermark
One run, three views: trace + cost + replay	06-developer-experience/17-trace-replay-cost-combo

Browse by category

Each link opens the category folder on GitHub. Every POC in it has a README that states what it shows, the exact command, and the expected output. The line under each heading says what the category covers; the folder holds the full list.

Foundations (17 POCs)

DSL syntax, materialization basics, branches and replay, file ingest, per-tenant routing, and the plan/apply deployment workflow. Credential-free (DuckDB).

Quality (11 POCs)

Contracts, inline checks, anomaly detection, local testing, SCD-2 snapshots, the standalone quality pipeline, and freshness SLAs. Credential-free (DuckDB).

Performance (14 POCs)

Incremental, merge, partition checksums, drift recovery, view intermediates, delete+insert, adaptive concurrency, cost and budgets, and EXPLAIN-based estimation. Credential-free (DuckDB).

AI (8 POCs)

Model generation, intent extraction, schema-change sync, test generation, schema-grounded validation, MCP data-grounding, and agent-policy tests. Most need an ANTHROPIC_API_KEY.

Governance (11 POCs)

Unity Catalog grants, schema patterns, workspace isolation, tagging, classification and masking, and retention. Plus cross-team contracts, auto-created target schemas, tag inheritance from a config group, recipe provenance, and the agent policy plane. Some need a Databricks workspace. Classification and masking, retention, auto-created schemas, tag inheritance, recipe provenance, and the agent policy plane all run on DuckDB.

Orchestration (12 POCs)

Shell hooks, webhook presets, remote state, checkpoint/resume, the Valkey cache, Dagster DAG mode, the circuit breaker, and idempotency keys. DuckDB, with a few using Docker.

Developer Experience (21 POCs)

Column lineage, the HTTP API, dbt import, shadow mode, CI, trace Gantt views, portability lint, PR preview and data diff, and lineage-diff. Credential-free (DuckDB).

Adapters (7 POCs)

Snowflake dynamic tables, Databricks materialized views, Fivetran discovery, a custom process adapter, BigQuery, a Rust-native adapter skeleton, and Trino via Docker. Most need the matching warehouse; the process-adapter and Rust-skeleton POCs are credential-free.

The full catalog

The complete, always-current list, with a one-line description per POC, lives in examples/playground/README.md. It also documents the benchmark suite.