What is in here
- The Job Search Problem – how much of a job search is structural overhead and how much of it one person can automate, written while running one. The time estimates come from the author's own tracked weekly cycles.
- The numbers the series rests on: ~65% of active job-search time goes to repetitive, rule-based work; at least 74% of active LinkedIn employers have a dedicated careers page; a recruiter's first pass at a CV lasts 6–7 seconds. Each has a page of its own with the arithmetic and the sources.
- What it builds: a reverse ATS – maintained criteria, a weekly sourcing cycle and the SQLite schema behind it – against the five structural failures of sourcing on LinkedIn, with 13 potential failure modes of the system itself listed in advance.
- Every number is either measured in the author's own cycles or cited, and where a figure is an estimate its arithmetic is shown on the page that states it.
~65% of active job search time is consumed by repetitive, rule-based tasks. This piece breaks down the five structural problems with LinkedIn-based sourcing and proposes a semi-automated system to solve them.
Jun 2026 · Original research
Read the full article →
13 potential failure modes in the job search automation system – URL rot, title standardization, list maintenance, cold starts, feedback loops, and more. Most dissolve on closer inspection; the rest have clear mitigations.
Jun 2026 · Original research
Read the full article →
Most career pages look JS-rendered, but the data behind them is public JSON fetched via API. Handling the top four ATS platforms covers ~80% of the target company universe – no browser automation needed.
Jun 2026 · Original research
Read the full article →
Five lists from "The Job Search Problem" mapped to eight SQLite tables – companies, job postings, CV versions, cover letters, applications, and a complete audit trail. The schema design decisions behind each choice, and why SQLite beats a spreadsheet for a serious job search pipeline.
Jun 2026 · Original research
Read the full article →
What the database found once it was full: 18,436 open postings read from 590 company career pages. Nearly two thirds of what boards file under “marketing” is sales; a B2B company that says marketing means sales, a B2C company means marketing; and every profession next door is exactly as narrow. The problem is the label, not the candidate.
Aug 2026 · Original research
Read the full article →
Sixty-one days and 41,441 postings after the schema was published: which design decisions held without a migration, which ones were torn out, and how the line between what the machine decides and what I decide moved four times.
Sep 2026 · Original research
Read the full article →