Opportunity library

Niches today

Each niche has its own page with the counts by opportunity type and verification state, the fresh, aging and stale split, the coverage statement and the dates of the crawl and the verification pass. The table summarizes them. A niche with no pool yet says so rather than showing a placeholder.

Read the stale count with its cause. A record is stale when it has never been read by the verification job or when its last read is older than 90 days. Every held record starts stale, and the first verification pass over each crawled niche read 100 records. The remaining records wait for the job, which reads at most 200 records per invocation under a written budget. Stale never means gone. It means unread.

What a record is

The subject of a record is the publisher side: a host, a page or a published submission route. It is never a customer, never a customer’s target page and never a destination we would like a link from. A record carries what we observed on public pages, the verifier’s reads with their dates, six eligibility verdicts with no score, its source class and its rights basis. Whether to pursue it stays the reader’s judgement, the same way the discovery use-case guides hand judgement calls back with the evidence attached.

Six opportunity types describe how a link would be earned on the surface: directory, resource page, listicle or review, guest post program, link insertion and AI citation source. For the crawled niches the type is a stated heuristic on the page’s editorial outlink count, and the record says so with a type basis of heuristic. A type read from a page that states the route, such as a submission page or a contributor page, carries observed.

Records come from five source classes only. Two are ours to read today: our own crawl of named publishers under robots, and our own route-verified directory inventory. Three are defined and empty: a Common Crawl host-graph seed that may point at a fetch and may never fill a field, a customer project lane with written consent, and our own listing outcomes. Bought lists, vendor exports, customer imports, competitor lists, vault captures and the private registry we inherited can never produce a record. The contract refuses them by value, and the strict schema has no field for a vendor metric, a contact address, an operator note or a score.

The two crawled niches were extracted from owned crawls of the kind the niche crawl guide describes, under two rules. A publisher-destination pair that appears on 20 percent or more of a publisher’s crawled pages is template chrome and is dropped. A destination counts as evidence only when two or more distinct publishers cite it editorially, the same agreement idea the link insertion guide applies to one article. Each record is the referring page, with the destinations it cites as evidence that the page links out to comparable sites. Whether that selection is good has not been judged. Until the owner’s blind scoring is recorded, every corpus niche page carries the label “selection not yet blind-scored”, and nothing on this site claims relevance from crawl accuracy alone.

States and ages

Three statuses, and only one of them is ever listed:

• Held: under verification or awaiting a publication decision; not listed
• Published: passed every gate and listed on this site
• Withdrawn: taken down on request or by a decision; never listed

The verification job reads a record with the same verifier the hosted checks and the CLI run: robots respected under our crawler token, a courtesy floor per host, no HTML retained. A complete read that finds the editorial link marks the record present and verified and stamps the date. A read that cannot conclude says unknown with a reason and moves nothing, so unknown never demotes a record and never counts as loss. Loss of a listing needs two complete absent reads at least 30 minutes apart, which is the same rule the methodology page gives for monitored placements.

Age is computed from the last verification. Fresh within 60 days of the last verification, aging from day 60 to day 90, stale after day 90 or when never verified. A stale record is held automatically and returns to published only after a new read passes, so a public listing can never show a record older than its window. The split on each niche page is computed at the data file’s own timestamp, September 15, 2026, and that date is printed beside it.

What the library never does

• It never sorts by authority and never carries a “dofollow” column. Google’s spam policies name low-quality directory links as an example of link spam, and a list sorted by a borrowed authority number with a follow flag reads like an invitation to build them. Link attributes appear only as the raw rel tokens the verifier read on the page, per occurrence, never as a directory’s own marketing claim.
• It never shows a vendor metric, a traffic estimate, a contact address or a score. The schema has no field for them.
• It never names a customer or a customer’s target, and it never publishes a dead-citation list from a customer’s project.
• It never shows a stale record and never lists a held one.
• It never claims that a surface absent from a niche does not exist. Absence from the library means the page was not read or did not meet the rules. The coverage statement on each niche page says what was read.
• It never sends, submits or fills a form. Your agent does the reading, the judgement and any authorized outreach with your own tools, as everywhere else on this site.

Take-down route

Every record carries the take-down route withdraw_on_request. A publisher who wants a surface removed writes to [email protected] naming the host or page. The record is withdrawn with the reason publisher_request and nobody argues the merits. Withdrawn records stay in the store with their reason so the next extraction does not recreate them, and they are never listed.