Why a shorter list of employers works better

A short list of employers whose boards were each confirmed by reading their real job posts beats a long list that was never confirmed at all, because the second list is full of dead pages, duplicate entries, and boards that turned out to belong to someone else. Size is not the thing that makes a list of job sources useful. Confirmation is.

The temptation of a wide crawl

It is not hard to build a large list of employer job boards. Guess an address for every company name you can find, sweep the results, and you will end up with a number in the tens of thousands within a day. That number looks impressive on a landing page. It is also mostly noise, because a guessed address that resolves to something is not the same as a guessed address that resolves to the right something.

Some of those guesses will land on dead pages. Some will land on boards that belong to a completely different company that happens to share a name or an abbreviation, the exact problem covered in the wrong board with the right name. Some will land on boards that used to be active and have not posted a role in years. None of that shows up in a raw count. A crawl counts pages found, not pages that are actually useful, and those are very different numbers.

What a smaller, checked number actually represents

Every employer board we hold was confirmed the slow way, by reading its actual postings and verifying they belong to the company we think they belong to, not by trusting that a URL pattern resolved correctly. Alongside the boards we keep, we permanently record every dead end: addresses that looked plausible and were checked and rejected, so they never get probed again.

That 189 is not a failure rate to be embarrassed about. It is the visible cost of doing the confirmation step honestly. A list with zero recorded dead ends is not a list where every guess happened to be right. It is a list where nobody checked, and the mistakes are still in there, just unlabeled.

Why this produces better results than a bigger number would

When you search against a curated list, every result you see has already cleared a real bar: it is a genuine board, for the company you think it is, still being maintained. When you search against an uncurated crawl, you inherit all of that uncertainty yourself. You do not know, from a result appearing in a list, whether it is a live board or a dead one, the right company or a coincidence of naming. You have to redo the confirmation work that a curated approach already did once, for every single result, every time.

That redone work is expensive in exactly the currency that matters most in a job search: your attention. A curated graph does the confirmation once, centrally, so it does not have to happen again for every person who searches it. Company job boards beat aggregators as a category for a related reason: the source is authoritative and does not go stale the way a copied listing does. A curated graph extends that same principle to the discovery step, before you even get to an individual listing.

What a new account gets on day one

None of this curation happens per user. A new account does not start with an empty list and build coverage over weeks of use. It inherits the entire set immediately, every confirmed board searchable from the first session. The work of finding and verifying each board happened once, ahead of time, and every account benefits from it the moment it exists.

This is the part that a raw board count cannot capture on its own. A board count only means something once you know it represents boards that were checked, not boards that were merely found, and once you know that checking happened before you ever needed it rather than something you would have to do yourself.

None of this scales for free, either. Confirming a board costs real work, whether that work goes into checking a newly discovered address against its postings or rechecking one that has been in the graph for a while. A larger but unchecked list can always claim to be bigger for less effort, because skipping the confirmation step is what makes a fast, large crawl possible in the first place. A curated graph trades that speed for something a raw crawl cannot offer: the ability to say, with actual confidence, that a specific board belongs to a specific employer and was still active the last time anyone looked. That trade is the whole argument for building the graph this way rather than the faster way.

Curation is not a one time event

Confirming a board does not make it permanently reliable. Companies migrate between applicant tracking systems, retire old board addresses when they rebrand, and occasionally go quiet for months at a time without formally shutting the board down. A list that was fully accurate on the day it was built will not stay fully accurate forever, which means a confirmed count is never a static achievement. It is a number that has to be maintained on an ongoing basis, with boards rechecked rather than confirmed once and assumed correct indefinitely. It is also the reason this piece does not quote you a current total: any figure printed here would be a photograph of one afternoon.

This is part of why the gap between a curated graph and a wide crawl only grows over time. A crawl that is never rechecked gets staler with every month that passes, silently, since nothing about its output changes unless the crawl runs again. A curated graph that is actively maintained catches that drift as it happens, which is a meaningfully different guarantee than simply having been accurate once, a while ago.

What curation does not solve

A confirmed board list tells you where to look. It does not tell you which of those employers are actually a fit, and this is where a lot of job search tooling quietly breaks down again after solving the discovery problem. Industry labels look like an obvious next filter, and they turn out to be a weak one in practice. Why industry filters return the wrong jobs covers that specific failure and what tends to work better once you already have a trustworthy set of employers to search across.

iapplyai.app's graph is built on exactly this kind of checking, with every rejected address recorded alongside every confirmed one, and every new account gets the whole set from day one rather than building it up from scratch.