Before the Offer

Why it exists

Most people choose their next role on instinct, urgency and whoever called first, and often to escape what hurts now rather than because something fits. At a career inflection point I wanted to treat the decision like any other high-stakes one: gather evidence, test my own assumptions, and fix the criteria before an offer exists to bend them.

The tools I found were either endless question lists or generic advice, and neither made me collide with facts. So I built the instrument myself, then the tools to act on what it found.

Approach

Replacing the questionnaires. I started from two AI-generated questionnaires, one with 243 questions and one with about 1,700, and wrote a comparative analysis of both. The large one was padded, repetitive and never asked for evidence. I replaced them with a finite, adaptive design: 12 sections with stable question IDs, in a fixed order. Facts come before causes, causes before patterns, patterns before traits, traits before wants, and wants before criteria. Follow-ups are capped at five per section, with a stop rule when two answers in a row add nothing new. Threads that run over budget go into a visible parking lot, so nothing is silently dropped.

Evidence, not narrative. Every claim is tagged fact, recollection, interpretation, hypothesis or unknown, and carries a source. A contradiction ledger keeps tensions visible instead of smoothing them over. Self-blame and self-excuse get challenged with the same two probes, so neither wins by default. Each section ends in a fixed-format summary that I corrected and locked before the next one opened.

Pre-registration. Before answering anything, I recorded my own explanations and predictions. That let me measure how far my beliefs actually moved, and flag any conclusion that simply matched my starting guess.

The outside view. Self-report has limits, so I tested my conclusions in structured conversations with former managers, people who watched me work. I asked open questions first, shared my synthesis afterward, then asked what I was getting wrong. Their evidence was layered onto the record without reopening it, and where it contradicted what I had written, I weighed the conflict explicitly instead of rewriting quietly.

What it produced. A locked, tagged record and an operating brief that sorts findings into Fact, Probable, Preference, Hypothesis and Open. Then a set of working documents I call the Career Decision OS: a role-screening scorecard, a recruiter triage form, an interview diligence playbook, an offer-decision memo template and a developmental evidence plan. Alongside them sit a search and outreach workflow (a company-profile matcher, a target-list process and a daily outreach cadence), screening-call prep, and a resume built so every line survives probing.

How it is built. Plain Markdown in numbered folders under git, so the record is portable and diffable. A design brief fixed the architecture before any question was written. An AI administrator ran the interview in 60 to 90 minute sessions while I answered, corrected every summary and locked each section, and multiple models served as interviewer and auditor. A Notion board tracks the pipeline. The look is document-first, closer to a working paper than an app: evidence tags in brackets, verdicts in blockquotes, and the private record kept apart from the front-facing summary.

Built to end. A project about rigor can turn into endless preparation, so the instrument has a hard ceiling and explicit stopping rules. Finished means specific outputs exist, not that it feels thorough enough, and once it was done the record was sealed with no extra rounds. It ended on schedule.

What I learned

Good decisions come from a method you trust before the pressure arrives. The sharpest result was how often an outside perspective reframed the question I thought I was answering. Building the instrument taught me to separate what happened from what I made of it, and to treat my own conclusions as hypotheses until they survive evidence.

It also taught me where reflection stops. The system exists to produce action, not more system, because some questions only live feedback can answer.

The instrument is finished by design. The next phase is putting it to work: running it against live opportunities, logging what breaks, and improving it only where real use shows friction. So far I am the only person who has run it, and I would generalize it for others only after that evidence is in.