Series

One product loop, born from a failed lookup screen: signals → grain → lists. That loop is a spine — one per grain. The pipeline is shared, so a later product does not get a second scrape. Three reading paths so a post is not asked to be vision + map + ops at once. Each post stands alone; use this page for order.

Full sequence (author SoT): BACKLOG.md. Terms: Glossary.

Lists from public records

Spine: signals → grain → lists (signal layering). The map names the four shared layers. Read in order when you can; any single post should still make sense alone.

#	Post	Job
1	Signals on a grain	Bet — combine signals on the property grain; match before an event joins the mix
2	Four layers (draft)	Map — shared pipeline; who owns land, match, lists
3	Why gather and augment are separate (draft)	Decision — land raw vs extract/match
4	Why gold stays denormalized (draft)	Decision — wide property row (OBT) for one list query
5	When Postgres is not enough for search (draft)	Decision — search as derived read model
…	See BACKLOG	Deep-dives after the map

How we run it

Where boxes live, how jobs run, how the repo is cut. After the map. Does not change the three moves.

#	Post	Job
7	Crontab over a scheduler (draft)	Batch orchestration
8	Many containers, one compose file (draft)	Service layout
9	Compose profiles for a small cluster (draft)	Host roles
10	One repo for a data platform (draft)	Monorepo boundary
18	Scale sources, not requests	One polite job per source; not faster crawls
19	Queue grain, not raw HTML	Queue parcel-matched signals; not raw HTML

Serving the buyer

Auth, one list query path, observability. The Serve side of the same grain contract.

#	Post	Job
11	OIDC for a B2B marketplace (draft)	Auth
16	One query path for lists (draft)	Search, preview, export, masking