Every interview Eric has ever landed came from applying directly or from someone he knew. None of them came from a third-party recruiter, and in July 2026 he went through six years of his own inbox to be certain: more than a dozen agency recruiters, no interviews, no placements. One channel worked and one never had. So he built the thing he had been waiting for the recruiters to be. CandidateZero reads job data straight from the systems employers post with, checks every role is still live, enforces a pay floor, remembers everything he has already ruled out, and never once applies on his behalf.
The audit started as a simple question: had any recruiter ever actually gotten Eric a job? He searched six years of Gmail, 2020 to 2026, and counted. More than a dozen third-party recruiters. Zero interviews sourced. Zero placements. Every interview in six years traced back to a direct application or a warm introduction from someone he knew personally.
The clearest case is a company called Revin. In 2025 an agency pitched Eric the role, stayed vague about the company, and produced nothing at all. In 2026 he applied to Revin himself. He reached the final round and they called his references. Same candidate, same company, same resume. The only variable was the middleman.
The uncomfortable part of finding a pattern like that is realizing how long you assumed the problem was you. The research says that reaction is close to universal: most job seekers were ghosted by an employer in the past year, a majority call the silence the hardest part of searching, and a third of ghosted candidates blame themselves for it. Meanwhile four out of five recruiters admit their own employer posts jobs that do not exist or are already filled.
Then there is the finding that reframes the whole industry. Glassdoor studied more than a million interview reviews and found that candidates who got their interview through a recruiter were ghosted more often than candidates who applied cold on a website. The paid channel performs worse than doing nothing.
None of that is a story about bad people. It is a story about how the money moves. A contingency recruiter gets paid only when a placement happens, and submitting a candidate costs the agency nothing, so spraying names at companies is the rational strategy rather than a personal failing. The costs land on everyone else. The employer eats the screening. The candidate becomes the expensive version of themselves, since hiring the identical person who applied directly saves the company the fee.
CandidateZero is not a job board and it is not a search tool. It is a pipeline that runs unattended and produces one short list. It reads job data straight from the systems employers already use to post roles, which are public and require no permission, no integration, and no login.
It pulls postings directly from the employer's own hiring system across six vendors. No scraped job board, no reseller, no listing that has been sitting in someone's index for three weeks.
Title families, then location, then pay. The order is deliberate and it is a cost decision: the expensive checks only ever run on the handful of roles that already survived the free ones. That discipline is what makes reading seventeen thousand postings cost pennies.
A title is not a job. The engine reads the description looking for roles wearing a costume: a go-to-market title whose duties are actually software engineering, or a systems role that is a sales quota in disguise. It quotes the giveaway phrase back to you rather than silently deciding.
Roles already applied to, rejected from, or ruled out never come back. That memory is the difference between a search tool and something that works for you over months. It lives in a private file that is not part of the public code.
A digest with pay, location, verdict, and a direct link. Roles that died since the last run get listed too, with how long they were alive, because an answer is a deliverable. On a quiet night it says so instead of padding the list.
Of the 33 roles that cleared every filter that night, 22 were ones Eric had already applied to, been rejected from, or ruled out, so they never reached his morning. Earlier the same evening the engine surfaced two roles the old eight-company version had never seen, one of them a $220,000 to $300,000 revenue operations architect job.
Finding jobs is close to free in 2026, so the code is not the interesting part. The interesting part is the index behind it.
There is no single place to ask a hiring system for every company's jobs. Each employer has its own address inside that system, and the address frequently has nothing to do with the company name. Chainalysis lives at one slug, Gong at another that ends in "io", Telnyx at one with a number stuck on the end. That is why the first working version of this engine read eight companies: eight was how many addresses Eric knew by hand.
So the real build was a discovery pass. It takes a company domain, finds its careers page, works out which hiring system it uses, extracts the address, verifies the address returns real postings, and records the answer with a date. In one evening it probed 3,257 domains and produced 719 verified company boards across six hiring systems, correctly burying 55 addresses that were dead. The count of postings visible to the engine went from 957 to more than 28,000.
That index is the asset, not the pipeline. The pipeline is a few hundred lines anyone could rewrite. The index was built by crawling and verifying, it heals itself when an address goes stale, and it gets more accurate every time it runs. Publishing the code gives away the easy half.
Because it is the same job, pointed at a different problem. Go-to-market engineering is finding the real buyer, verifying the signal, and routing to the right human before the moment passes. This does that for careers instead of pipeline: source from the system of record, verify liveness, filter on real criteria, dedupe against history, and deliver something a human can act on in ten minutes. The domain happens to be job listings. The craft is identical.
It never applies for you. The obvious feature request is auto-apply, and every competitor ships it. This will not. Volume is the disease, not the cure, and a tool that submits on your behalf recreates exactly what the agency did: your name arriving somewhere you did not choose, in a form you did not read. The engine arms you. You take the shot.
Everything personal stays out of the public code. The engine is meant to be open. The pay targets, the companies already ruled out, and the application history are not, so they live in files that are excluded from the repository by an automated check that fails the build if anything private is ever staged. That check was the very first commit, before any working code existed.
Nothing gets shown that has not been checked. Every listing is verified against the employer's own system before it appears. Every zombie listing caught before a human sees it is a rejection nobody has to internalize. That is the honest version of the promise, and the current limit is worth stating plainly too: because the engine reads employer systems directly, its inventory is live by construction rather than by clever detection. Catching ghost listings out in the wild, on aggregators, is the next piece of work rather than a thing it already does.
It cannot reach what was never posted. The one thing agencies genuinely sell is the unposted role, and no amount of reading public data will surface it. The counterargument is the audit itself: across six years and more than a dozen recruiters, that access produced zero interviews. It was not being delivered anyway.