Warehouse estate
Three real warehouses sit behind the subagent*.com family. None of them is hypothetical: each schema below was read from its DDL file (or the worker source that writes it) at authoring time, with the file path cited. The live ways in are subagentetl.com (the pipeline dashboard over the D1 warehouse) and subagentdb.com (validated SELECT-only natural-language-to-SQL over the same star).
1 · subagentjobs-dwh — the D1 job-board star
A Kimball-style star written by workers/etl (ATS crawl endpoint) and
crawled on a 6h schedule by workers/cron across
73 Greenhouse + 7 Lever +
1 Ashby boards (board counts from the literal
GREENHOUSE_BOARDS/LEVER_BOARDS/ASHBY_BOARDS arrays in
workers/cron/src/index.ts). The same star also exists as the Postgres copy in
crates/durable-store/migrations/postgres/001_schema.sql, which is where these
columns were read from.
job_post_id (PK)titlelocation_namelocation_typeabsolute_urlcontent_lengthfirst_publishedupdated_atoffice_countis_prospectcompany_nameplatform
columns: 001_schema.sql lines 1–15 + the INSERT column list in workers/etl/src/index.ts
board_token (PK)nameplatformlast_snapshot_sha256last_crawled_atjob_count
columns: 001_schema.sql lines 17–24; upserted via INSERT … ON CONFLICT by workers/etl
2 · durable-store Postgres — 10 migrations
The local Postgres 16 behind crates/durable-store (L2 Redis / L3
Postgres tiered store). File list from ls crates/durable-store/migrations/postgres
— 10 migrations; the newest two add the agent-dispatch routing-telemetry capture (009) and the
repo code FTS index (010). Repo-only fact, dated snapshot — list refreshed 2026-07-10.
| migration | creates |
|---|---|
001_schema.sql | fact_job_posting, dim_board, dim_skill, bridge_job_skill, crawl_log |
002_indexes.sql | indexes (incl. idx_crawl_board on crawl_log) |
003_add_eviction.sql | eviction support |
004_vendor_filesystem.sql | fact_filesystem / dim_file_ast (indexer targets) |
005_task_events.sql | task FSM events |
006_ecosystem_catalog.sql | fact_orgs, fact_repositories, dim_packages |
007_doc_pages.sql | fact_doc_pages (docs-crawler CDC) |
008_fleet_deploy_ledger.sql | fact_fleet_deploy (+ status/wave indexes) — added 2026-07-09 |
009_agent_dispatch_events.sql | agent_dispatch_events — source-shaped capture of the zero-token model router + telemetry hooks (deliberately NOT fact_/dim_: mirrors the real Claude Code hook payload field names) |
010_repo_file_index.sql | fact_repo_files — content FTS (sha256 CDC + GIN tsvector, modeled on 007) over the local repo's own code corpora: workers/, plugins/, crates/, scripts/ |
3 · the WBR star — coworker × week × metric
The weekly-business-review warehouse owned by
plugins/cwc-data-engineering/skills/wbr-warehouse. Canon is one file per table in
data/warehouse/ — each carrying a cube.dev-style CUBE_SEMANTICS block,
an ETLM dependency block, and portable DDL, kept from drifting by
validate.py. Facts are written only by registered deterministic collectors
(anti-hallucination gates G1–G6) — no LLM-produced number may enter the fact table.
| table | role |
|---|---|
dim_coworker | one row per coworker plugin |
dim_week | Sun→Sat weeks in America/Los_Angeles; week_id = yyyymmdd of the Saturday end date |
dim_metric | metric registry — DDL owned by cwc-data-engineering, rows seeded only by cwc-finance's typed seeder |
fact_coworker_week | THE fact. Grain: coworker × week × metric; value is NULLABLE (NULL = 'not measured', never 0, never fabricated); rates carry numerator/denominator so fleet ratios are Σn/Σd |
rpt_wbr | the report view the weekly-business-review page reads |
Honest scope
This page is a static map, not a live console — row counts and freshness live on subagentetl.com, and ad-hoc queries go through subagentdb.com. The durable-store Postgres and the WBR star run on the operator's workstation, not on a public endpoint — they are documented here because they are real, not because you can query them from this page.