> ## Documentation Index > Fetch the complete documentation index at: https://quester.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Tuning the Advisor > Five levers, your hard rules, and what it has learned about you: how the advisor reads you and how it sounds. An advisor you cannot argue with about their manner is an advisor you eventually stop listening to, and this page is where you settle that argument. Five sections hang off the sticky sub-nav, covering what your work is about, five behavioural levers, absolute rules that beat everything else, the facts it has inferred about you, and a calibration exercise. Changes take hold immediately, with no save-and-restart and no waiting for the next session. All of it shapes *how* the advisor responds, and none of it can loosen the [boundaries](/manual/advisor), which sit above the lot and hold whatever you do here. The advisor preferences page: the about-my-work statement, the five behaviour levers, and the hard rules below. ## About my work A free-text statement the advisor reads at the start of every session, covering your topic, your stage, your methods and what you care about, with a placeholder that gives you a fair template: *"My dissertation looks at… I'm in my third year… I care most about…"* This field does more for the quality of your answers than any lever on the page, because an advisor that knows you are a mixed-methods researcher in year three who cares about measurement validity says different things from one that knows only your project's name. It starts from what you wrote during [your first ten minutes](/manual/your-first-ten-minutes), so rewrite it whenever the work moves. ## Behavior - the five levers Five sliders, one to five, applied to every session. **Verbosity**, running Minimal, Concise, Balanced, Thorough, Exhaustive, sets how much comes back, so Minimal leads with the conclusion and stops while Exhaustive walks through caveats, edge cases and alternatives. **Directness**, running Very gentle, Gentle, Balanced, Direct, Blunt, sets how criticism gets phrased, and at Blunt it says exactly what is wrong with no cushioning. Some people work better that way and some do not, so there is no correct setting, only yours. **Pushback**, running Affirming, Agreeable, Balanced, Challenging, Adversarial, sets how hard it argues with you, and it is the one to move first. At Affirming it backs your direction and objects only when it must, whereas at **Adversarial** it stress-tests every claim like a sceptical committee member and argues the counter-case, which is where you want it for a week before a defence. **Citation strictness**, running Permissive, Relaxed, Balanced, Strict, Very strict, sets how tightly it ties claims to your library, and at Very strict every claim needs an in-library citation while anything it cannot ground gets marked as such. Use it in drafting season, when an unsupported claim slipping through costs you months later. **Proactivity**, running Silent, Reserved, Moderate, Proactive, Constant, sets how much it volunteers, so at Silent it answers the question you asked while at Constant it brings adjacent observations, risks and suggestions along with every answer. Start with everything at Balanced and move one lever whenever something annoys you, because moving all five at once teaches you nothing about which one was the problem. ## Hard rules Absolute instructions that beat any lever, kept in two lists, **Always** and **Never**, with up to twenty-five rules between them. Discipline-specific and supervisor-specific constraints belong here, and most people never use the section at all. Real examples include *Always use British spelling*, *Never refer to participants as "subjects"*, *Always flag when a source is older than 2015*, and *Never suggest quantitative methods for this project*. Rules beat the levers, so a rule saying never do something holds even at Proactivity 5, though rules cannot beat the boundaries, which means no rule will get it to draft your chapter. ## What it's learned about you Patterns the advisor has inferred from working with you, written out as plain statements you can read. You can edit any of them or forget them, since **Forget** removes a fact for good so it is never inferred again, while **Forget everything** clears the lot after a confirmation that tells you exactly how many. Anything you have edited by hand carries a *"you edited this"* mark, so your corrections stay distinguishable from its guesses. There is a principle underneath this section, which is that an advisor quietly building a model of you that you cannot see or correct is not one you can trust. So everything it thinks it knows sits on this page in words and you get the last say, and if it has learned something wrong you forget it and it stays forgotten. An empty list is normal: *"it notes patterns as you work together."* ## Retrain The calibration exercise from your first ten minutes, available any time. From an Advisor conversation, open **Project context**, expand **About this advisor**, and choose **Adjust guidance** to reach it. It shows sample observations and you mark each one **More like this** or **Less like this**, and applying the batch moves your levers immediately. This gets you to the right tone faster than reasoning about five sliders in the abstract, because most people recognise the voice they want long before they can describe it. Run it again when the work changes shape, since the register that suited you during coursework is rarely the one you want heading into a defence. ## Turning it off The advisor switches on and off per project, in [Project settings](/manual/project-settings) under Advisor, and when it is off this page says so and links you there. Off means gone, with no cards, no observations and no prompts anywhere in the project. The boundaries still govern anything it would say, though it will not be saying anything. Next: [The Journal](/manual/journal).