I score 97.80 on recruiting search.
Evaluated on PSB-Recruiting, the 30-query recruiting category of PeopleSearchBench. It measures whether a people-search system finds the right candidates, returns enough of them, and packages what a recruiter actually needs to decide.
$100 of credits on us. No card required.
Ahead of every published baseline.
Laidback scored 97.80 overall, versus the strongest published baseline (Prism) at 89.64. That is an 8.16-point lead.
Averaged across 7 independent evaluator runs so the number isn't a lucky judge day.
'%20fill='%23FCFCFC'/%3e%3cellipse%20cx='304.949'%20cy='107.231'%20rx='43.7363'%20ry='86.7747'%20transform='rotate(-12.9533%20304.949%20107.231)'%20fill='%23FCFCFC'/%3e%3cellipse%20cx='207.296'%20cy='107.321'%20rx='33.7301'%20ry='67.712'%20transform='rotate(-12.2177%20207.296%20107.321)'%20fill='%23121212'/%3e%3cellipse%20cx='301.354'%20cy='103.848'%20rx='31.5494'%20ry='64.0532'%20transform='rotate(-12.2177%20301.354%20103.848)'%20fill='%23121212'/%3e%3c/svg%3e)
Laidback
Prism
Lessie
Juicebox
Exa
ClaudeFour things a recruiter would actually check.
PSB-Recruiting doesn't reward a chatbot for sounding confident. It grades what matters when you're staring at a shortlist on Monday morning.
Are the top-ranked candidates the ones a senior recruiter would actually shortlist first?
Did we surface enough qualified profiles, not just a lucky handful?
Is every profile packaged with enough evidence to review without a second tab?
Judged by an independent LLM against explicit, brief-derived criteria, not vibes.
How the benchmark works.
Same 30 briefs. Same normalized outputs. Same LLM judge grading against brief-derived criteria and public evidence. The underlying paper is on arXiv.
- 01Brief in, live search out
30 recruiting briefs written like real hiring requests. For each one, Laidback returns up to 15 candidates ranked strongest fit first.
- 02Normalize the outputs
Every returned candidate is normalized into the structured format the benchmark evaluator expects. No format bonus, no format penalty.
- 03Decompose the brief
The evaluator breaks each brief into explicit, checkable criteria: seniority, skills, domain, location, disqualifiers.
- 04Ground in public evidence
Live web search gathers outside evidence for every candidate so judgments are grounded in real, verifiable public information.
- 05Judge and score
An LLM judge reviews the evidence and scores each result on relevance, coverage, and information utility.
- 06Run it seven times
The full evaluator runs 7 independent times and averages, cancelling judge variance.
A benchmark isn't a trophy. It's a promise: the shortlist I hand your recruiter on Monday is graded by the same rules, every single time.
Reproducible. Auditable. Defensible to a hiring manager who wants to know why one candidate ranked above another, and that the answer isn't "vibes."
Get $100 of credits and try it on the roles you have open today.
$100 of credits on us. No card required.