How It's Made

A plain description of how this site gets written, checked, and corrected, including where I use AI tools and where I don't. If you're going to trust any of this with decisions this size, you should know how it was put together.

The short version

This is one person's project. It's free, there's nothing to buy, and there never will be. I made it because nobody handed me this information, and I don't think that's a good reason for someone to lose a decade.

There are mistakes in here. The body of knowledge is enormous, it changes every year, and I can't catch all of it. What I can do is tell you how each part was built, be clear about where I'm least certain, and fix things quickly when someone tells me I've got something wrong.

The three parts, and how each one was made

The site has three surfaces, and they weren't built the same way. That changes how much weight each one can carry.

The Road

The articles are mine. Every article on the Road is something I wrote: the arc of it, the arguments, the judgment calls about what's worth saying and what's kinder to leave out. When a page tells you something is a bad idea, that's my opinion and I'm accountable for it.

I drafted these articles. What goes in, what stays out, the arguments, and the judgment calls about what to warn you about are mine. I use AI tools the way I would use an editor and a research assistant: checking my facts against sources, catching where I have contradicted myself, tightening and formatting, and helping me put an argument more clearly than I put it the first time. Not every word on this page was typed by me. Every judgment on it was made by me, and I sign all of it.

The Sky

This one was built differently.

There are 151 specialty and subspecialty profiles here. No one person can research 151 fields by hand to any real standard, and pretending otherwise would be the first untrue thing on this page. So I built a process instead: a shared structure every profile follows, and AI-assisted research that pulls each field's facts from primary sources, mainly NRMP match reports, ACGME program data, AAMC workforce data, and published compensation surveys.

Then I review them. Reviewing 151 profiles is not the same thing as re-checking every number inside them. That's exactly why the checking described below exists, and why the Sky is the part of this site I audit hardest.

The Money

I drafted these too, and the priorities are mine: what actually matters to someone with no family safety net, and what a person should be warned about before they borrow. These pages need more current research than anything else on the site, because the rules move constantly, so the research underneath them is AI-assisted, pulled against current federal rules and published aid data, and then reviewed by me. The same split applies as on the Road. The words were worked on with tools; the decisions were not.

The shelf life is short. Federal student loan rules were rewritten by a 2025 law whose major provisions landed in July 2026, and both the provisions and their effective dates can still shift.¹ Some of what's on these pages will go out of date before I catch it. Anything here that touches your actual money, meaning a deadline, an eligibility rule, or which repayment plan you're on, should be checked against studentaid.gov or your school's financial aid office before you act on it. I mean that as an instruction rather than as a disclaimer.

Where AI sits

I use AI tools throughout. Research, fact-checking, editing, catching my own inconsistencies, and building the systems that keep 151 profiles in a consistent shape.

I'm not hiding it. It's the only way a site this size gets built by one person, and being cagey about it would be the first dishonest thing on a site whose entire point is not doing that.

There is a tradeoff, though. AI is what makes this possible, and it is also exactly where confident-sounding wrong numbers come from. A tool will hand you a salary figure attributed to a survey that never published it, and it will sound completely authoritative doing it. So the answer isn't to avoid the tools or to trust them — it's to build checking around them, treat the output as wrong until something confirms it, and be honest about what that checking does and doesn't catch.

What the tools didn't do is decide what this site should say. The intention is mine. The vision is mine. Every judgment call about what to tell you, what to warn you about, and what to refuse to sugarcoat came from someone who went through this without a map and remembers exactly how that felt.

How I check things

Where the numbers come from. Where I give you a number, I try to point at where it came from and when. Match data comes from the NRMP. Program and trainee counts come from the ACGME. Workforce and demographic figures come from the AAMC.

And the sourcing I'd rather you knew about. Pay is the softest ground on the site. Some compensation figures come from surveys that sit behind a paywall and reach these pages through outlets that reprint the table. Some come from physicians reporting their own pay in public forums, which is first-hand about their own lives and nothing more; where that's the case, the page says so in the open rather than burying it in a footnote. And 29 pages cite admissions-consulting firms. Those companies sell a service to the people reading this, so I trust them less than I trust a board or a federal agency, and they are not allowed to source a price, a fee, or a requirement anywhere on this site. Where one of them is carrying descriptive detail about a specialty that no official body publishes, I'll use it and name the source. If you find a number here with nothing behind it, that's a bug rather than a style, and telling me is the fastest way it gets fixed.

Crowd panels and recruiter guides. Two more kinds of source carry a lot of the specialty pay figures, and more than half of the specialty pages use at least one. Crowd panels pool salaries physicians submit themselves. Recruiter and advisory firms publish salary guides from the offers they see. I use both because for many subspecialties no survey publishes a line at all, and I'd rather you have a number with appropriate caveats than no number at all. Neither is a survey and neither should be read as one. Pages name the source in the sentence; they do not always say which kind it is, or how small the sample was, and that is a gap I am still closing.

The ⟳ mark. When you see ⟳ next to a figure, that's me telling you this is the kind of number that goes stale: pay, match rates, burnout, the share of women in a field. All of those move every year. Treat it as a nudge to check the current source before you rely on it, rather than as a knock on the number.

"Limited data" instead of a guess. When there isn't a trustworthy number, I say so rather than inventing one. This comes up more than you'd expect, because nobody publishes pay for most subspecialties. Where the honest answer is that the data doesn't exist, that's what the page says, and where I give you an estimate instead, I tell you where it came from and how much to lean on it.

Some numbers I won't publish at all, because people expect to see them and the data behind them isn't real. Divorce rates by specialty. The share of first-generation physicians in a given field. A confident statistic with nothing underneath it is worse than an honest gap, so I leave the gap and I tell you it's there.

What "people online say" is synthesized, never quoted as fact. When a page describes how a field gets talked about, I'm pulling together a broad read across trainee and physician forums and paraphrasing it. No verbatim quotes, no forum chatter presented as hard truth, and no claim about who was posting. Take it as the temperature of a community, which is useful, rather than as data, which it isn't.

Every page carries a "last reviewed" date at the bottom, so you always know how fresh it is.

Checking, and re-checking

Pages go wrong on their own. Nobody has to touch them. A match rate that was right in March is wrong by the following spring, and a repayment rule that was accurate in June stops being accurate in July, while the page sits there saying exactly what it always said. Writing something carefully once isn't enough.

So there are three kinds of checking here, and they do different jobs.

Automatic checks, every time I publish. A set of checks runs against the whole site before anything goes live, and what they catch is mechanical failure rather than a wrong fact: a link pointing at a page that doesn't exist, a page nothing links to, a footnote defined but never cited (which the publishing tool drops silently, so the citation disappears from the page without anyone noticing), a citation marker with no footnote behind it, a references list with nothing in it, and a source from a category this site doesn't accept. None of that is glamorous, and all of it is the kind of thing that erodes trust when a reader trips over it. Machines are better at it than I am, so I let them do it.

Slower checks I run by hand. Some of the most useful checking can't run on every publish. Confirming that every citation on the site still resolves means fetching roughly 2,400 addresses, so it runs on its own schedule. So does the check that compares a number in a figure's caption against the number in the article underneath it, which flagged 26 places worth a second look across eight profiles the first time I ran it. I'd rather tell you those are manual than let "automatic checks" sound like it covers more ground than it does.

Audits against the outside world, on a cycle. This is the one that actually matters. Every figure on this site came from somewhere: the NRMP for match data, the ACGME for program and trainee counts, the AAMC for workforce and demographics, published surveys for pay, studentaid.gov for the loan rules. Auditing means going back to those sources and asking whether what they say now still matches what this site says. Where it doesn't, the page changes and the date at the bottom moves.

The ⟳ marks are how I know where to look. They aren't only a warning to you. Every ⟳ on this site is also on my list, flagging at the moment I wrote it that this specific number is the kind that expires. When new data lands, those marks are the worklist.

When it happens. I won't invent a schedule I can't keep, so instead of promising a month, here is the trigger list. The checking happens when the Match runs and the NRMP publishes results, when the annual compensation and burnout surveys come out, when the ACGME and AAMC release new workforce data, and whenever the federal loan rules move, which lately has been often and without much warning. Those are the moments the numbers here go stale, and those are the moments I go back through them.

And the honest limit. This is one person doing it. Some things will lag. A page can be checked and still be wrong, because I misread a table or a source updated quietly. Audits reduce mistakes; they don't eliminate them. That's the whole reason the next section exists.

What this site knows about you

Almost nothing.

There are no accounts, no sign-ups, no forms, and nothing to buy. The site sets no cookies, and it stores nothing about you in your browser beyond three anonymous flags that clear when you close the tab: two so that a one-time hint on the Road and on the star map does not replay at you, and one so a failed page load only retries once.

I do count visits. The site runs Cloudflare Web Analytics, which tallies page views and the site you arrived from. It uses no cookies, it cannot follow you anywhere else, and it cannot tell me that one person read four pages. Visitors in the EU are excluded from it entirely. It tells me which pages get read and where readers arrive from, which is how I decide what to work on next. It is the only third-party code on the site.

Cloudflare also hosts the site, and like every web host its servers log requests. I have no way to connect any of that to a person and I don't try.

The only thing I ever learn about a reader is what someone chooses to write to me.

If something looks wrong, tell me

This is the part I actually want you to use.

If a number looks off, if something contradicts what your advisor told you, if a page describes your specialty in a way that doesn't match your experience of it, tell me. You don't need to be certain. You don't need to be polite about it. "This looks wrong," with a link to the page, is plenty.

I'll check it, and if you're right I'll fix it and note the change so the next person gets the better version. If you're a physician and I've described your field badly, you're the person I most want to hear from, and I'll credit you if you'd like.

Recommendations are just as welcome. If something's missing that would have helped you, I'd rather know than not.

Write to [email protected].

What this is, and what it isn't

This is one person, working alone, giving away the map I wish someone had handed me. There's no company behind it, no course at the end, nothing to buy. I'm not trying to build anything except a thing that helps.

Which also means it's just me. I get things wrong. I miss updates. I can't cover everything, and there are real gaps, some because I haven't written them yet, and some because the information genuinely doesn't exist anywhere.

This is educational information, not advice, and none of it is medical, financial, or legal. It's a place to get oriented rather than the last word. For anything that actually binds you, go to the AAMC, to studentaid.gov, to the schools themselves, and to advisors who know your situation.

And if this helps one person, someone who'd otherwise have gone through it the way I did, guessing, a step behind, sure that everyone else got a manual nobody ever handed you, then I'm glad I made it.

Take what's useful. Verify what matters. Tell me when I'm wrong.

— Saad

References

Last reviewed: 2026-08-16.

Federal borrowing for professional degree programs is capped at $50,000 annually and $200,000 in aggregate from 1 July 2026, with Grad PLUS eliminated for new borrowers. Those limits, and the repayment plans that go with them, are the provisions this page is describing as volatile. Federal Student Aid: https://studentaid.gov/understand-aid/types/loans · What the caps mean against a four-year cost of attendance is Can You Actually Get the Money?. ↩