Run one real job from start to finish using everything you have built. This is where it either holds together or shows you what is missing.
What to do
Pick a job you are genuinely hiring for. If you have none, use the practice job and a colleague as the hiring manager.
- Intake and blueprint. The twelve questions, the profile with its gaps marked, the scorecard, three must-haves, weighted scores.
- Sourcing. Widen your search terms, build the search, produce a longlist with the source recorded for each person.
- Screening. Check how well the CVs were read, enter your rules, run it, do the disagreement check, record the rules with the shortlist.
- Ranking. Build the match profile from your scorecard, rank, and write one justification per candidate quoting their CV.
- Outreach. Five-message sequence for anyone you approached, checked before sending.
- Interviews. Structured loop, written score descriptions, scores locked before the debrief.
- Decision. Debrief against the bar, recorded with evidence.
- Records. Rules saved, rejections given with reasons, every AI step covered by your notice.
What to hand in
A one-page write-up of your own run: what you did at each step, what took longer than expected, where the process broke, and the one change you would make first.
Writing it is the exercise. A run you do not write up teaches you much less, because the things that went slightly wrong get forgotten within a week.
Where these runs usually break
| Step | Common problem | Usually means |
|---|---|---|
| Intake | The manager will not trade anything | Go back to question 12. Do not start searching |
| Sourcing | The pool is far smaller than expected | A requirements conversation, not a search problem |
| Screening | Everyone scores about the same | Your rules are not separating people |
| Outreach | Silence | Targeting, not writing. Check who you contacted |
| Interviews | Interviewers score after discussing | Lock the scores. Nothing else fixes this |
| Decision | "Culture fit" appears | Ask which skill. Every time |
| Records | You cannot reconstruct why someone was rejected | Save the rules with the shortlist from now on |
Judge your own run
- Could someone else repeat this using only what you wrote down?
- Could you explain any rejection to the person it affected, in plain words?
- Did the disagreement check change anything? If not, look at why.
- Which step took longest, and was it the step that mattered most?
- What would you do differently — and is it a rule change or a process change?
One job, then improve
Do not run five at once to gather data faster. One run, written up, changed, then the next. Teams who scale before the first write-up scale their mistakes along with everything else.