Can a prompt replace Apollo?

Build a prospect database with an obsession about provenance. It contains only what you put in it, and it never lets you forget how old that is.

Stack: your choice, Postgres, one command to start.

Model: Company, Person, and Fact. The twist is Fact. Instead of columns on Person, every attribute — email, title, phone, LinkedIn URL, company — is a Fact row with: the attribute name, the value, a source (free text, required, no default), a captured_at date, and a confidence I set from a fixed list. A Person can hold several conflicting Facts for the same attribute at once; the UI shows the newest first with the others collapsed beneath it, and never silently discards the old one.

Freshness: each attribute type has a configurable half-life — emails 180 days, job titles 120, phone numbers 365. Anywhere a value is displayed past its half-life it renders with a visible stale marker and the age in days. Exports include a column with the age of every exported value.

Add a "verify" workflow: mark a Fact as re-checked today, which writes a new Fact rather than editing the old one, so the history of a changing job title is preserved.

Search and segmentation: filter people and companies on any current Fact value, on staleness, and on tags. Save segments by name. Export a segment to CSV with the provenance columns included, because the point of this build is that the CSV you hand someone is honest.

Import: CSV with a mapping screen that forces you to name a source for the file before it will run. No source, no import.

Write tests for the freshness computation at boundary dates and for the invariant that verifying never mutates an existing Fact row.

Do not scrape, do not buy or embed any third-party contact dataset, and do not send email. This is a filing cabinet with a good memory.