Independent, and Never Alone

Every so often a prospective PhD applicant asks me what my supervision style is, and I give an answer that sounds evasive until you have seen it working: the idea will be yours, and I will be there every week trying to break it.

People hear the first half and think I mean hands-off. They hear the second half and think I mean control. It is neither, and the distance between those two misreadings is most of what I have learned in the last decade. So here is the philosophy stated plainly, the record it has produced, and — because this is the part nobody publishes — what it costs me.

“Independent” does not mean “on your own”

The single thing I want a student to feel, from month one, is that they are a researcher rather than a pair of hands. That means the research question is theirs to choose and theirs to defend. They get their own space, physical and intellectual: a desk, a project that has their name on the intellectual deed, and explicit permission to tell me I am wrong — in group meeting, in front of everyone, with evidence.

What that does not mean is that I hand them a topic and reappear at the viva. I have watched that model produce three miserable years and a thesis nobody wanted, and the sector data suggests it is not rare enough to ignore: UKRI's own review of doctoral training noted that only around 66% of supervisors have mandatory induction training and just 29% have mandatory continuing professional development as supervisors.3 Most of us were never trained to do this. We are copying whoever supervised us, which is a transmission mechanism for bad habits as much as good ones.

The honest sector picture is better than the folklore, mind. The 2025 Postgraduate Research Experience Survey — more than 35,000 doctoral researchers across 93 institutions — put satisfaction with supervision at 89% and overall satisfaction at 83%, the highest in over a decade.2 Supervision is one of the things British doctoral education already does well. I am not writing this because the average is bad. I am writing it because the average hides two very different failure modes at either end, and I have strong views about which one to fear.

So the arrangement is this. They own the question. I own the standard. They are free to spend a month chasing something I think will not work — and they should, because I have been wrong about that more than once. What they are not free to do is take a half-argued claim to a top venue with my name on it, because the reviewers will be less kind than I am and the rejection will teach them less.

What “elevating the idea” actually means

Here is the part that sounds like arrogance and is really just arithmetic. NeurIPS 2025 took 5,290 papers out of 21,575 valid main-track submissions — an acceptance rate of 24.52%, judged by 20,518 reviewers, 1,663 area chairs and 199 senior area chairs.4 Three quarters of the work sent there was rejected, and a large fraction of it was good.

A student's raw idea is almost never the problem. In my experience the idea is usually the best thing in the room. What separates it from an accepted paper at the very top of the field is a set of unglamorous things a first-time author has no way of knowing yet: which baseline a reviewer will demand and be annoyed you omitted; which ablation turns a plausible story into a demonstrated one; which sentence in the introduction is doing all the work and is therefore the one that must be bulletproof; which venue's community will recognise the contribution at all.

That knowledge is mine to give away, and giving it away is not the same as taking the idea over. The test I apply to myself is simple: at the end, can the student explain every decision in the paper as their own reasoning? If the answer is no — if there is a section they cannot defend under questioning because it was really mine — then I have failed, no matter where it was published. That rule also happens to be the fourth ICMJE authorship criterion doing its job, which I wrote about at length in MyTh 003.

The practical shape of it:

• They pick the problem; I stress-test the framing. Most weeks the useful question is not “is this true?” but “who would be surprised by this, and why would a reviewer care?”
• I am the hostile reviewer first. Better that the objection arrives from me in a Tuesday meeting than from Reviewer 2 in four months. I have a reputation in my group for asking the same question five different ways. It is deliberate.
• Aim high on purpose, early. A student who has only ever submitted to venues they were sure of has not learned what the ceiling looks like. First-year rejections from a top conference are tuition, and they are cheap.
• Their name, first. On the work they led, they go first. This is not generosity; it is accuracy.

The record

Philosophy is easy to assert, so here is the evidence, all of it, including the people whose work has not landed yet. Everything below is drawn from my own supervision record;1 where an achievement has a public source, it is cited.

The fuller version of this record — every publication, every destination, laid out person by person — is on the mentorship page.

KTP: the same philosophy, with a company and a deadline attached

Knowledge Transfer Partnerships deserve their own section, because they are the version of this that cannot hide behind a long time horizon. A KTP Associate is a graduate employed to embed a piece of research inside a real business, with real money and a real board expecting something to work. Over 50% of Associates are offered a job by the host company at the end.23

People assume the independence has to go out of the window in that setting. My experience is the opposite: it is the setting where independence pays fastest. My associates were not handed a specification. They were handed a business problem, a supervisor who asked hard questions weekly, and the authority to tell the company that its preferred approach would not survive contact with the data. That is why the models ended up in production rather than in a report — at Horus, at BT, at Mersea Homes.

One honest note on credit, since this entry is a claim about my own record. On the Mersea Homes partnership I was a co-investigator at a 40% role, not the lead academic; the university's public write-up rightly names the colleagues who led it. On the Horus KTP I was the lead academic and principal investigator — that one is mine to claim, and the university's announcement records my role.20 On another KTP I wrote the entire work package and plan and then handed the PI role to a colleague who had not yet led one, which I would do again. Supervision includes supervising your own appetite for the byline.

What it costs

Now the part I would want to know if I were choosing a supervisor.

This style is expensive. Being a hostile reviewer for eleven concurrent projects means actually reading eleven literatures, not skimming eleven drafts. It means weekly meetings that overrun because a student found something interesting, and it means reading a paper five times across five versions when a faster supervisor would rewrite it once and be done by Thursday. There are weeks when the supervision alone is most of the working week.

I want to be clear that this is a choice and not a virtue. Rewriting a student's paper myself would be quicker, and the paper would probably be accepted at a similar rate. What it would not produce is a person who can do it without me in three years' time. The entire justification for the slower route is that the output of a PhD is not a thesis; it is a researcher. Every time I have been tempted to optimise for the paper instead, the student has been the thing that got optimised away.

And I should say plainly: I love it. Not in the dutiful sense people write in promotion cases. The best hours of my week are the ones where somebody twenty years younger than me explains why my objection is wrong and turns out to be right. That has happened often enough that I have stopped being surprised by it and started designing for it.

What I am not claiming

Three caveats, because a table of achievements is exactly the kind of artefact that invites over-reading — and I spent the whole of MyTh 003 arguing that a list of outputs is a question rather than a verdict.

• These achievements belong to the people who earned them. António's Microsoft Fellowship was won by António, on a thesis led by his primary supervisors in Portugal, where my role was co-supervisory. Every name above had other supervisors, collaborators and mentors. A supervisor who claims a student's award has misunderstood what they did.
• Not everything worked. One KTP was withdrawn mid-project when the company hit COVID-era losses; the associate's work was good and the outcome was still a project that stopped. Papers I was confident about have been rejected. Students have hit dead ends that cost them months, some of which I should have spotted sooner. A record like this shows what survived, which is not the same as what happened.
• The model has limits. It scales badly — there is a headcount past which “in the room every week” becomes a polite fiction, and I have felt the edge of it. And it suits people who want to argue. A researcher who wants a specification and a quiet life would be better served by someone else, and I try to say so honestly at interview rather than discover it in year two.

Come and argue with me

If you are thinking about a PhD and any of this sounds like the way you want to work — you bring the question, I bring the standard and the stubbornness — then write to me. Tell me what you find interesting and why the obvious answer to it is wrong. I do not need a polished proposal; I need evidence that there is something you cannot stop thinking about.

I read every one of these emails myself, and I would rather have a short message with one real idea in it than three pages of my own publications quoted back at me. Funding routes vary and I will be straight with you about what exists and what does not. But the first step is cheap: M.S.Jameel@southampton.ac.uk.