TuringCorp

DeciderExamples › Writing

Recorded run · Writing · 2026-09-15

A new hire asked how we decide things here, and I want to point them at one short section of the handbook. Which version?

Two candidate answers were submitted. Decider picked one and reported how close the call was. The reasoning below is the run's own output, published verbatim.

Option A

Small, reversible calls belong to whoever is closest to the work. If it is hard to undo, or crosses teams, write a page and get two people to read it before we start.

Option B

Most calls here are made by the person doing the work, so ask them. When something is bigger than one team, or hard to undo, put it on one page and ask two people.

SUGGESTION: A
Confidence: 78.3% · recorded 2026-09-15

Option A is the stronger handbook excerpt because it gives a clearer, more actionable decision rule. It states a crisp principle: small, reversible calls belong to whoever is closest to the work, which directly tells a new hire what authority they have and lets them self-assess. It also defines escalation precisely: when a decision is hard to undo or crosses teams, the process is to write a page and have two people read it before starting. That timing matters because it ensures lightweight review happens before work begins, rather than leaving room for retroactive review. By contrast, Option B is more conversational and has the concrete phrase 'the person doing the work,' but its logic is weaker: 'most calls' softens the rule, 'ask them' can be circular if the reader is the person doing the work, 'ask two people' is vague about what is being asked, and 'bigger than one team' is less precise than 'crosses teams.' Option B also omits the before-start timing. Overall, Option A better mirrors a reversible-versus-irreversible decision framework, gives new hires both clarity about their own authority and a precise escalation procedure, and works better as a short handbook reference.

How to read this

The confidence value is calibrated against outcomes, not a vibe: on JudgeBench, judgements Decider rates 90%+ were correct 99.6% of the time, and the 80–90% band 94.0%. The lowest band flags near-ties where either choice is defensible — an answer, not a failure. Full tables: the product page.

This page is a recorded run published as a worked example. It is not a live call, and nothing here is charged. Raw judgements behind our numbers are published at api.turingcorp.net.

Try your own two options All Writing examples

Prefer the other side? Option B is right there above — the point of a calibrated reading is that you can tell a decisive call from a coin flip.