We hid your job description in a database. Go get it.
No form. No résumé black hole. One read-only Postgres, four tables, one live posting for 1005 Labs. Its body is the JD and your claim token. Nobody said it was tidy.
Connect ↓
$ psql "postgresql://candidate.<ref>:findme@aws-0-ap-south-1.pooler.supabase.com:5432/postgres?sslmode=require"
erp=> \dt erp.*
companies postings posting_versions audit_log
erp=> SELECT id, title FROM erp.postings
erp-> WHERE state = 'published' AND company ILIKE '%1005%';
(4 rows) -- four. we have one job.
erp=> \d+ erp.postings -- read the column comments.
erp=> -- it's in there. nobody said it was tidy.
Schema erp, read-only, connection string above.
Table and column comments are part of the puzzle.
Find the one posting that is live right now for 1005 Labs. Its body is the JD and a claim token. Ten minutes is the budget.
Email with the token as subject: how you found it, what you rejected on the way and why, and the first thing you'd fix in this schema.
Fast.
Scientific.
Flexible.
Any tool counts.
- psql, DBeaver, TablePlus, DataGrip
- Claude Code, Cursor or Codex with a Postgres MCP
- A notebook: psycopg + pandas, 20 lines
- Your own agent: give it the connection string and the question
Agents are expected, not tolerated. If one did the work, send its transcript or the script. We care what you noticed and what you rejected, not which tool you picked. A right answer with no rejected rows is a weak answer.
Prefer physics to SQL?
Puzzle 016 · The logger says fine. The insulin disagrees. Four reefer trucks, 2,400 rows of telemetry, one spoiled shipment the loggers never saw. Six static files, five minutes, any agent.
The database is a throwaway project with synthetic data: no real customers, invoices or people. The role is SELECT-only on one schema. Nothing you can do will break it; nothing you break will help. Dash, Vision and Flash are the agents we build at 1005 Labs.