Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when laying out the set of interview stages for a role — designs the loop as a competency-coverage matrix where every must-have competency has exactly one owner stage and each stage carries a distinct signal, instead of a list of similar-sounding rounds that overlap and leave gaps. Do NOT use for drafting the questions inside one interview (structured-interview-questions), writing up one interview's notes (interview-scorecard-format), or combining finished scorecards into a decision (panel-debrief-synthesis).
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 337% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 367% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 355% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 2278% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 372% | 0% |
Asked to "set up the interviews for role X", a base model lists rounds: phone screen, technical, onsite, culture fit, final. The rounds sound reasonable but overlap heavily — three of them quietly re-test "problem solving" and "communication" — while a competency the role actually depends on gets no dedicated signal. The loop is longer than it needs to be and still has holes.
Design the loop from the competencies, not from a count of rounds. Build a matrix, assign one owner per competency, and only then decide how many stages you have.
competencies (e.g. coding, system design, debugging, cross-team collaboration, ownership). These are the rows of the matrix. Ignore how many interviews you think you want — that falls out later.
that produces the deciding signal on it. That stage is its primary owner. A stage may touch another competency in passing (a system-design round hears the candidate communicate), but only the owner's signal counts toward the hire decision on that competency. One owner per row.
a set of competencies nothing else owns and to emit one output — a scorecard section, a go/no-go on its dimension. Write down what each stage's signal is.
sets are near-duplicates. Merge the stages or move one competency so each has a single owner.
redundant — cut it or fold its unique piece into a neighbor. "Final round" and "culture fit" earn a slot only if they own a competency (e.g. values alignment, executive presence) nothing else tests. The right loop is the fewest stages that cover every must-have with one owner each.
Produce the matrix (competencies down the side, stages across the top, one primary owner marked per competency), then per stage state the competencies it owns and its one signal, then the coverage check: list any GAP (uncovered competency) and any OVERLAP (competency owned by more than one stage, or duplicate stages) with the fix. If the loop is clean, say so explicitly — no gaps, no overlaps.
Don't invent competencies the ask didn't imply, don't grade a specific candidate, and don't write the questions inside each stage — that's a separate step. Design the loop; leave the interviews to the interviewers.
> Role: backend engineer. Must-have competencies: coding, system design, debugging under > pressure, cross-team collaboration.
Matrix (one primary owner per row):
| Competency | Recruiter screen | Coding round | System-design round | Behavioral round | |-------------------------|:---:|:---:|:---:|:---:| | coding | | owner | | | | system design | | | owner | | | debugging under pressure| | owner (live-debug segment) | | | | cross-team collaboration| | | | owner |
Per-stage signal:
and how the candidate recovers when a test fails live.
Coverage check: No gap — all four competencies owned. No overlap — each owned once. Note the loop dropped a generic "onsite culture fit" round the base model would have added: it owned no competency the behavioral round didn't already cover, so it was redundant. Four competencies, three signal-bearing stages plus the recruiter screen.
Counter-example the method catches: a loop with "technical phone screen", "onsite coding", and "take-home" all owning coding is three owners for one row — overlap. Collapse to one coding owner and reclaim two slots (or leave the role with a shorter loop).
Other measured skills in the registry, with their headline benchmark lift.