README.md

Example 3 — fork a published dataset

A second agent takes a published release and builds a different dashboard — no RPC, no Postgres, no original credentials (acceptance A10).

Prerequisites

Example 1 (examples/transfer-traffic) built and published. Publishing works to a directory target or an S3-compatible target (see its chainplot.yaml).

Note on URL forks: fork --from https://… requires the bucket to allow public (unauthenticated) reads. The example R2 bucket is private, so this walkthrough forks from the local publish output. The fetch rules are identical (checksums, caps, no redirects).

Fork

chainplot below is the CLI from a checkout of this repo: pnpm build, then alias chainplot="node $PWD/dist/cli/main.js" (or pnpm link --global).

chainplot fork --from ./dist/releases/local --output ../forked-dashboard --json
# or, against a publish root with latest.json:
chainplot fork --from ./published --output ../forked-dashboard --json
# or, against a public https release root:
chainplot fork --from https://your-bucket.example.com/prefix --output ../forked-dashboard --json

The fork copies the pinned snapshot (datasets/<id>/tables/*.parquet), the recipe (chainplot.yaml, queries, tests), and strips the chain/event sources — a fork has no chain access. Expanding history or changing contracts is a new ingest project, not implied by the fork.

Second agent adds a query

cd ../forked-dashboard
mkdir -p queries
echo "select count(*) as transfer_count from usdc" > queries/second_agent_count.sql
# add the query + a dashboard panel to chainplot.yaml, then:
chainplot build --json

validate and build run fully offline over the pinned snapshot.