08 Wrap up

Final proof

SELECT
    (SELECT uniqExact(condition_id) FROM polymarket.markets FINAL) AS markets,
    (SELECT count() FROM polymarket.price_ticks) AS ticks,
    (SELECT count() FROM polymarket.trades_clean) AS trades,
    (SELECT count() FROM polymarket.market_midpoints_1m) AS midpoint_minutes,
    (SELECT max(event_at) FROM polymarket.price_ticks) AS newest_tick;

You are done when markets, ticks, and midpoint_minutes are greater than zero and newest_tick matches this run. A quiet live feed can legitimately have zero reconciled trades; fixture mode always produces trades.

What you built

• public market discovery with no credentials;
• a live WebSocket path with heartbeat, stall detection, reconnect, and REST BBO fallback;
• a reconciled public trade path with overlapping windows and deterministic IDs;
• typed ClickHouse tables designed around actual filters;
• an insert-time one-minute quote-midpoint aggregate; and
• a Cloud dashboard or equivalent saved SQL visualizations.

Stop the local collector

docker compose --env-file .env.polymarket down

This removes the stateless container. Cloud data and saved queries remain.

Optional Cloud cleanup

Only run this if you no longer want the workshop data:

DROP DATABASE polymarket;

Delete or idle the Cloud service through clickhousectl or the console according to your organization policy.

How this scales in production

The workshop writes directly because the source is an HTTP/WebSocket API and the volume is small. If a production relay already publishes to Kafka, Kinesis, Pub/Sub, or another supported stream, use ClickPipes for managed offsets, schema mapping, backpressure, and error handling. Do not add a broker only to make the workshop look more distributed.

Return to the workshop catalog or compare this path with the AI SRE track.