Navigating Conflict

Disagreeing with someone senior, challenging a result, authorship disputes, and misconduct — including the honest cost of formal recourse and when leaving is the correct answer.

Before you start


Read this	Some months before you need it. Judgement is worse from inside a live dispute.
The core skill	Telling apart a scientific disagreement from one about credit or power. Misreading which one you are in is the failure that produces most of the damage.
This page will not	Pretend that formal processes are reliably safe, or that persistence always wins.

First, work out which kind you are in

Three kinds of conflict get treated as one, and they respond to completely different moves.

	Scientific	Credit / power	Conduct
What it is about	Whether a claim is right	Who gets what, or who decides	Behaviour that breaches a rule or a person
Resolvable by	Evidence, and one of you changing your mind	Negotiation, norms, sometimes escalation	Process, documentation, sometimes exit
Good-faith assumption	Usually safe	Sometimes	Do not rely on it
Written record	Helpful	Important	Essential from the first incident
Speed	Take your time	Move before positions harden	Document immediately, act deliberately

The characteristic error is treating a credit dispute as a scientific one. You keep producing better evidence, believing that is what the argument is about, while the other party is negotiating position. You will lose slowly and be confused about why.

The diagnostic: ask what would happen if you were definitively right. If the answer is “the claim gets corrected and we move on”, it is scientific. If you can imagine being completely right and still not getting what you are arguing for, it is not a scientific dispute and evidence alone will not settle it.

Disagreeing with someone senior

The fear is that disagreeing costs you. Usually the opposite is true — disagreeing well is one of the fastest ways to be taken seriously, and most senior people can tell the difference between a challenge and a nuisance within one sentence.

The move

Separate the claim from the person, and give them a route to change position without losing face.

“Can I check my understanding — you’re saying the reciprocity enrichment holds because the degree-preserving null already controls for the hubs? I ran it with a distance-preserving null as well and the effect drops to 1.1×. That might mean I’ve built the null wrong. Could you look at how I set it up?”

What each clause is doing:

“Can I check my understanding” — makes it a question about your comprehension, not their competence. Cheap, and it costs you nothing you actually need.
Restating their position accurately — proves you engaged. It also occasionally ends the conversation, because they hear their own claim and revise it.
The specific evidence, with a number — this is what converts an opinion into something addressable.
“That might mean I’ve built the null wrong” — a genuine escape route for both of you. If you are wrong you have not staked your credibility; if you are right, they can concede by examining your setup rather than by admitting error.
“Could you look at” — an action, so the conversation ends somewhere.

What not to do

Do not do it first in public unless the stakes require it. Give them the chance to update privately; it costs you nothing and it dramatically raises the odds.
Do not accumulate. Six months of silent disagreement delivered at once reads as an ambush regardless of merit.
Do not over-concede. Folding the moment they push back is also a calibration failure — it teaches them your objections are not load-bearing, and next time they will not listen at all.
Do not argue about tone. If they were sharp with you, that is a separate conversation on a separate day.

If they are simply wrong and will not move

Establish what is actually at stake. If it does not affect what gets published, let it go — being right is not a deliverable. If it does, put your position in writing, once, neutrally: “Recording where I’ve got to, so it’s on file: I think X, for these reasons. Happy to be wrong.” That sentence has saved a lot of people later.

When you think a published result is wrong

Proportionality is everything here, and getting it wrong is the most common way early-career people damage themselves.

Escalate in this order:

Assume you are wrong first, and do the work. Re-read the methods for what is not stated. Try to reproduce it. Most apparent errors are misreadings, and finding that out privately costs nothing.
Ask someone who knows the subfield. “Am I misreading this?” Often the answer is that the convention you find bizarre is standard and has a reason.
Contact the authors. Neutrally, specifically, with your reproduction attempt attached. Most people respond well, and a surprising number already know.
Public comment — a preprint comment, a post, a conference question — only if it matters, the authors have not engaged, and you can state it without attributing motive.
Formal correction or retraction request — for results that will mislead the field, not for results you disagree with.

Never open at step 4 or 5. The cost is asymmetric: the field remembers who was gratuitously harsh far longer than it remembers who was right.

The sentence that keeps you safe at every level: describe what you did and what you got, not what they did wrong. “I couldn’t reproduce Figure 3 with the stated parameters; here’s my code” is unanswerable. “Figure 3 is wrong” invites a fight about you.

Authorship and credit disputes

Lab norms covers the conventions. This is about what to do when they break.

How these actually get resolved: almost always by negotiation shaped by who has power, occasionally by a mediator, rarely by formal process. Journals will point at authorship criteria but generally will not adjudicate. Institutions have processes that are slow and that most people find unsatisfying. Knowing this in advance changes your strategy — it means the leverage is in prevention, not in appeal.

Prevention, which is where nearly all the value is:

Agree author order when the work starts, out loud, and write it down. An email to yourself is enough. This one habit prevents most disputes.
Re-agree it when the scope changes, because scope always changes.
Keep a contemporaneous record of what you did. Not for a tribunal — for the ordinary case where three people genuinely remember it differently.

If it has already gone wrong:

Establish what you actually want. Author position? Acknowledgement? An apology? Something different next time? Vagueness here loses arguments.
Ask directly and privately first, framed as a question rather than an accusation: “I want to check how we’re thinking about authorship on this — my understanding was I’d be second, and I want to make sure I’ve got that right before it goes out.”
If that fails, escalate to a person, not a process — a co-author with standing, a graduate director, a mentor outside the group. Processes are slow and blunt; people resolve most of these.
Formal routes exist. They are appropriate for genuine misconduct. Understand before you start that they are costly and that “resolved” often does not mean “resolved in your favour”.

The proofreading case specific to this field: community proofreading contributions are essential and chronically under-credited. If you are contributing at scale, ask early what the credit model is — contribution counts, acknowledgement, authorship thresholds. The projects that do this well state it up front; if a project cannot answer the question, that is information.

Harassment and misconduct

This section is deliberately blunt, because writing that soft-pedals it does real harm.

Document from the first incident. Date, time, what was said or done, who was present, and how you responded. Contemporaneous notes stored somewhere not controlled by your institution. You are not being paranoid — memory degrades, and the value of a record written the same day is very hard to replicate later.

Know the routes that exist, before you need them. Typically: your supervisor; a graduate director or equivalent; a designated harassment or Title IX-type office; HR; an ombudsperson; a funder’s own reporting route. They differ in one crucial respect — confidentiality. Some are confidential; some are mandatory reporters, meaning that telling them triggers a process you cannot then stop. Find out which is which before you disclose anything, because that choice is not reversible.

What protection does and does not exist. Retaliation is generally prohibited by policy. Prohibition is not prevention. Reporting can cost time, standing, and sometimes a working relationship, and the outcome is often not disclosed to you. Many people who report say afterwards that they would do it again; many say the process was worse than they were told it would be. Both are true, and you are entitled to weigh that for yourself rather than be told what the brave choice is.

An ombudsperson is usually the safest first conversation where one exists — typically confidential and non-reporting, so you can understand your options without starting anything.

Do not go alone if you can avoid it. One other person who knows what is happening, ideally outside the power structure, materially changes both your judgement and your safety.

When to leave

Leaving is a legitimate strategic choice, not a failure of resilience. Consider it seriously when:

The behaviour is known to people with the power to stop it, and has not been stopped.
You are managing around a person rather than doing your work.
The cost is showing up outside work, persistently.
The thing you would gain by staying is one line on a CV.
You have run the honest version of “what do I need from here to finish what I’m doing?” and the answer is that you cannot get it.

Leave well if you can. Finish or hand off cleanly, keep the relationships worth keeping, take your data and notebooks (check what you are entitled to before the last week), and line up references outside the situation. This field is small; the person you are in conflict with will be at conferences for decades. That is an argument for leaving cleanly, not for staying.

Check yourself

Your PI presents your analysis in a talk and describes it as work from the lab, without naming you. Scientific, credit, or conduct?

Credit — and probably an ordinary one rather than an act of bad faith.

Apply the diagnostic: if you were definitively right that it is your work, would that settle it? No — you would still need them to do something, which means evidence alone is not the lever.

Proportionate response is a private, specific, forward-looking ask: “I saw the AL/PM analysis in your Tuesday talk — could we add my name on that slide for future versions? Happy to send you a version with it.” Concrete, easy to grant, no accusation, and it establishes the expectation for next time.

What not to do: raise it publicly, treat one slide as a pattern before you have one, or say nothing for a year and then raise it during an authorship negotiation, where it will read as leverage rather than as a real grievance.

You cannot reproduce a well-known result from a paper's stated parameters. What is your first move?

Assume you are wrong and do the work — check versions, seed, preprocessing, and in this field especially the materialization version, which is the single most common source of irreproducible connectomics numbers and is frequently unstated.

Then ask one person in the subfield whether you are misreading it.

Then email the authors with what you did and what you got, attaching your code. Neutral, specific, no attribution of motive.

Only after all three, and only if it matters, does anything public come into it. The asymmetry is worth remembering: if you are right and you went privately first, you lose nothing. If you are wrong and you went public first, that is the thing people remember about you.

Common errors and how to recover

Fighting a credit dispute with evidence. Recover: run the diagnostic — would being definitively right settle it? If not, change instrument.

Opening at maximum escalation. Recover: there is no way back down. Start privately; you can always escalate, never de-escalate.

No contemporaneous record. Recover: start today, even mid-dispute. Note explicitly which parts are reconstructed from memory — a record that distinguishes the two is far more credible than one that does not.

Disclosing to a mandatory reporter before you meant to start a process. Recover: not possible, which is why the order matters. Ask about confidentiality first, as a hypothetical.

Treating leaving as defeat. Recover: cost it honestly against staying, with one outside opinion.

Over-conceding to end discomfort. Recover: if you fold whenever pushed, your objections stop carrying information. Hold the ones that matter and let the rest go explicitly.

What this page does not cover

Legal advice — for anything with legal weight, talk to someone qualified, and note that your institution’s lawyers act for the institution. It also does not cover the specific procedures at your institution, which you should look up now rather than during an incident, nor the ordinary difficult-conversation skills in lab norms.

Lab norms — authorship conventions, escalation paths, and what a PI expects but will not say
Belonging — including how to tell a bad environment from a knowledge gap
Career mechanics — why positions are contingent, which shapes the power in most of these disputes
Reading and judging work — how to establish whether a result is actually fragile before you challenge it