How PACR Keeps Your Research Metrics Accurate

Same afternoon, same task, two desks.

At the first desk, an assistant professor is updating her annual review. She has three tabs open, each showing a different citation count for the same body of work. One profile includes a conference abstract that was later expanded into a journal article, so that study is counted twice. Another is missing a paper published under her maiden name. A third has quietly absorbed two papers by a chemist who shares her first initial and surname. She spends ninety minutes reconciling them in a spreadsheet, picks the number she can best defend, and writes a footnote explaining why it doesn't match anything a reviewer might look up.

At the second desk, a colleague doing the same review opens one profile. His publications are already linked to his identity. A paper that appeared last month is already there. A small note says one record was held back for review because its author details didn't line up. He confirms it isn't his, and he's done before his coffee goes cold.

The difference between those desks isn't diligence. Both people are careful. The difference is whether the system underneath their numbers was built to be accurate or simply built to count.

That second desk is what we've tried to build in PACR. Since metrics now feed into rankings, hiring conversations and grant narratives, I want to explain how we approach accuracy, where we think the usual approach falls short, and what you should still check yourself.

Why I care about this more than most product people would

I spent a long time in academia before I moved into research product work. I've sat on promotion committees. I've written the letters. I've watched a single bad number follow someone through a review process because nobody had time to question it.

One story stays with me. Our department chair, Ruth, was reading files for a mid-career promotion and noticed that a candidate's h-index had jumped sharply in a single year. That kind of jump usually means a landmark paper or something odd. Ruth was one of the more careful chairs I've worked with, and she did what most people under deadline don't do. She opened the underlying list.

Eleven of the papers weren't his. A researcher with the same name, in a related field at a different institution, had been merged into his profile by an automated aggregator. The candidate hadn't noticed. Honestly, why would he? Who audits a number that's going up?

What bothered Ruth, and me, wasn't the error. Errors happen. It was that the error was invisible until a human with time and skepticism went looking. Most committees don't have Ruth, and most Ruths don't have time for every file.

So my view is simple. A metric is only as fair as the identity and data behind it. If you get those wrong, any ranking built on top of them is unfair, even if the math is perfect.

Step one: making sure a profile actually belongs to the person

Most accuracy problems begin with identity, not citations. Name ambiguity is old and well documented. Common surnames, transliteration differences, name changes after marriage or divorce, and initials-only author lists in older journals all make it hard for any system to know who wrote what. If you want to see how much has been written about this, a PubMed search for author name disambiguation will keep you busy for a while.

That's why PACR verifies profiles before it treats their metrics as trustworthy. The principle is that a profile should be anchored to persistent identity, not to a name string. Persistent identifiers like ORCID exist for exactly this reason. A name is a guess. A verified identifier is a claim someone has stood behind.

In practice, it looks like this:

1. You set up your profile and connect it to your verified research identity.

2. PACR uses that verified identity as the anchor for which publications belong to you.

3. Records that only match on name, without supporting identity signals, aren't silently added to your profile. They wait for you to confirm them.

Some people find step three slightly annoying the first time. I'd ask them to remember Ruth's candidate. A system that asks "is this yours?" a few times a year is much better than one that never asks and occasionally gives you eleven papers you didn't write.

What you should still do yourself

Verification only works if your identifiers are kept up to date. If you've published under more than one name, make sure your persistent identifier reflects that. If you moved institutions mid-career, check that both affiliations are connected. Ten minutes on this once will save you far more time later.

Step two: syncing publication data instead of retyping it

The second source of bad metrics is stale or hand-entered data. I've seen CVs where the "in press" paper had been published two years earlier. I've seen profiles missing a researcher's most-cited work because it came out after the last time anyone updated the page.

Manual entry fails quietly. Nobody sets out to keep an outdated record. People are just busy, and updating a profile never feels urgent until a deadline arrives.

PACR syncs publication data from authoritative records instead of relying on you to type it in. Scholarly metadata infrastructure like Crossref exists so that a publication's core details are registered once and can be reused. If you build on records like that, your profile reflects what was actually published, with the actual title, venue and date, not what someone remembered.

The bigger benefit, in my view, comes from where this sits in your workflow. PACR is an AI-native research workspace that connects searching, reading, writing, collaborating and publishing. Your publications aren't a separate silo you visit once a year. They're part of the same environment where you find literature and draft manuscripts. When a paper you worked on in the workspace is published, the record you see on your profile matches the work you already know. You're not maintaining a second, disconnected version of your career somewhere else.

A realistic example

Say you're a postdoc with a paper that goes from preprint to journal publication. In a disconnected setup, you'd often end up with two records, two partial citation counts, and a profile that undercounts the paper's real influence, or overcounts it if both versions get added together.

In PACR, the goal is to treat those as versions of one piece of work, not two separate outputs. When there's ambiguity about whether two records are the same work, that's the kind of thing we'd rather surface to you than guess at. Duplicates inflate numbers, and split records deflate them. Both distort your profile.

Step three: flagging unusual metric changes

This is the part I care about most, and the part where I disagree most with how metrics are usually handled.

The common approach is to count everything and let the totals speak. Most researchers don't question it, because higher numbers feel like good news. But anyone who's spent time around research integrity knows that metrics can move for reasons that have nothing to do with influence. Misattributed papers are one cause. Coordinated citation behavior is another, and it's been studied seriously enough that a PubMed search on citation manipulation and citation cartels turns up a real literature. Organizations like COPE publish guidance because these problems are real, not hypothetical.

So PACR flags unusual metric changes instead of quietly absorbing them. When a profile's numbers move in ways that don't fit its history, the change is marked for a closer look before it's treated as settled.

I want to be careful here, because flagging can easily turn into accusation, and it shouldn't.

A flag is a question, not a verdict

Ruth taught me this one too, from a different file a few years later. A young faculty member's citations had jumped dramatically in a single year. Someone on the committee was quietly suspicious. Ruth looked into it. The researcher had co-authored a methods paper that a large field had adopted almost overnight. The spike was real, earned, and the most important thing in the file.

If the system had just suppressed the jump as suspicious, it would have punished exactly the kind of work we say we want to reward.

That's the balance we aim for. Unusual changes get flagged so that a misattribution or a strange citation pattern doesn't flow straight into a ranking. Flagging doesn't mean the change is wrong. Many unusual changes have ordinary explanations:

• A paper becomes a standard reference in its field.

• A large consortium paper with hundreds of authors gets published.

• A record merge brings in papers that genuinely were yours but were missing.

• A highly cited review article starts gathering citations from across a field.

The point of a flag is to make sure someone asks "why did this move?" before the number affects how one researcher is compared with another.

Why this matters for rankings in particular

A personal profile that's slightly off mostly hurts you. A ranking that's slightly off hurts everyone in it.

Rankings are relative. If one profile is inflated by a misattributed paper, everyone below it moves down a place without doing anything differently. Unfairness spreads through a ranking in a way it doesn't through a single CV. That's why we put the accuracy work before the ranking and not after it. Verified identity, synced data and flagged anomalies are what make it reasonable to compare people at all.

I'll also say plainly that I don't think any metric, however accurate, should be the whole story of a researcher. The San Francisco Declaration on Research Assessment makes that case well, and I agree with most of it. Numbers lack context. A clinician-researcher whose work changed practice in a small specialty may never match the citation counts of someone in a huge field. Accuracy doesn't fix that limit. What accuracy does is make sure that when metrics are used, they at least describe reality.

Those are separate problems, and I'd rather we solve the one we can solve well than pretend to solve both.

What I'd tell you to do this week

If you're a researcher with a profile anywhere, PACR or not, here's what I'd actually do, in order:

1. Open your publication list and read it. Read the list itself, not the totals. Look for papers you didn't write, papers you did write that are missing, and duplicates.

2. Check your persistent identifier. Make sure name variants and past affiliations are connected.

3. Look at your metric history, not just today's number. If something jumped, make sure you know why. If you can't explain it, a committee member will ask eventually.

4. Fix problems at the source. Patching a spreadsheet only fixes one document. Correcting the underlying record fixes every place that uses it.

If you're a department chair or sit on review committees, I'd add one more step. When a number surprises you, open the list behind it. It takes five minutes. Ruth did it every time, and I've never seen a careful chair regret it.

Where this fits in the rest of your work

I don't want accuracy to be a feature you think about once a year at review time. It works best when it runs quietly in the same place you already work. In PACR, the profile and its metrics sit alongside the parts of research that take most of your time: finding and exploring the literature, reading, summarizing, drafting and working with co-authors. Because those steps are connected, the record of what you've published grows out of the work itself instead of being rebuilt separately from memory.

That's the second desk from the opening. It's not a researcher with more discipline. It's a researcher whose system was built to be right.

If you want to see what your own profile looks like when identity is verified and publications sync on their own, you can set up your PACR workspace and read through your publication list the way Ruth would. Start with the papers. Then look at the numbers.