When Kenaptic says no¶

Most documentation describes what a product does. This page describes what Kenaptic declines to do, and what it says instead of a number when it can't produce one honestly.

It's here because a tool that writes into your live content earns its trust in exactly these moments. "We found nothing" and "we were unable to look" are completely different statements, and a product that blurs them will be believed at precisely the wrong time.

"Not measured" is not zero¶

Several figures in Kenaptic can only be computed once certain conditions hold. When they don't, Kenaptic says so in plain words rather than showing a confident zero.

Content gaps need somewhere to publish. A gap is a topic with demand in one place and no coverage in a place you could publish to. If none of your sources is writable, there is no such place — so the question has no answer, and Kenaptic reports that it cannot be computed rather than reporting no gaps. Connect a writable source and the number appears.

Orphan and siloed counts need a completed crawl. Until Kenaptic has read enough of your estate to see which pages link to which, the Overview says not measured. These counts include your existing links as well as Kenaptic's, so they describe your estate rather than our coverage of it — which only works once the coverage is real.

A capped crawl is reported as a partial view. Kenaptic paces its requests and caps how much it reads from any one site in a single run, so a large estate never lands on somebody's server as a burst of traffic. If a run hits that cap, Kenaptic tells you the picture is partial instead of quietly presenting it as complete.

Thin and empty sources are flagged, not averaged away¶

A source that returned one page, or none, is almost always a configuration problem — the wrong address, an aggressive robots rule, a site that needs JavaScript to render, or an address that redirects somewhere unexpected. Left alone, it silently drags down every estate-wide number you look at.

Kenaptic flags these on the source itself:

Empty — nothing was read at all.
Thin — so little was read that any conclusion drawn from it would be noise.
Absorbed by another source — everything found here actually belongs to a source you already connected (see below).

The flag is the useful output. A source showing zero results tells you something true about your setup, and hiding it behind a tidy dashboard would just make Kenaptic look unreliable later.

Overlapping sources: one page has one owner¶

If you connect docs.example.com and also example.com, the two overlap — and a page counted twice would inflate every total in the product.

Kenaptic assigns each page to the most specific source that claims it. A page under docs.example.com belongs to your docs source even though the broader example.com source could also have claimed it. Addresses that redirect are followed first, so a source configured at an address that has since moved is still matched correctly rather than appearing to own nothing.

The practical consequence: if a source looks emptier than you expected, check whether a more specific source is legitimately claiming its pages. That's usually the answer, and the totals are right even when one row looks surprising.

Pages Kenaptic refuses to write to¶

Some pages are unsafe places to add a link, and getting this wrong is worse than doing nothing — the change looks correct right up until it disappears. Kenaptic declines to write to:

Generated pages — documentation built from code, schemas or configuration is overwritten by the next build.
Translations — a translated page belongs to the people who maintain that language. Kenaptic proposes against the source language and lets translation flow normally.
Frozen versions — old releases that no longer take changes.
Content synced in from elsewhere — a folder mirrored from another repository is erased by the next sync.
Archived projects — something wound down is a place to link to, never a place to propose work.

Every refusal is reported, never silent. See How linking stays safe for the reasoning in full.

Sources that decline to be read¶

If a third-party site's rules ask crawlers not to read it, Kenaptic doesn't — the page is skipped and the refusal recorded. robots.txt is how the rest of the web says no to us, and it is honoured as a complete opt-out.

For your own sources the rule is different, and deliberately so. Kenaptic records what your robots.txt says and shows it against the source, but it does not treat it as a refusal. Your robots.txt is aimed at search engines and other people's crawlers; you have already told Kenaptic to read this site by connecting it, and a Disallow: / meant to keep a staging site out of Google should not silently produce an empty estate. If you want a section of your own site left alone, scope the crawl (below) rather than relying on robots.txt to do it.

Either way the source appears in your list with a clear status, so a missing silo is never a mystery. You can also ask to be told — A domain refused crawling is one of the events available in notifications.

What you can and can't keep out of a crawl¶

Today, you scope a crawl in rather than filtering pages out.

Each source is configured with a starting address, and Kenaptic reads only what lives under that path, on that host. Point a source at example.com/docs/ and /pricing, /careers and everything else on the domain are never read at all. A narrower starting address is the control you have — and for most estates it's the one you want, because it's also faster and cheaper than reading pages you'll discard.

Kenaptic also drops some things on its own, without being asked:

Listing pages — archive indexes, tag pages and paginated "page 2 of 47" pages. They're navigation, not content, and counting them as pages distorts every number downstream.
Another source's pages — where a more specific source claims a subtree, the broader source cedes it, so nothing is read or counted twice.
Platform interface pages — a repository host's own menus-and-buttons routes, as opposed to the content hosted on it.
Non-canonical versions — where a docs site publishes many versions of the same page.
Links no human would follow — hidden traps some sites plant to catch impolite crawlers.

There's no per-page or pattern exclusion list yet

You can't currently say "read this source, but never /internal/*". The workarounds are to connect the narrower path as its own source, or to connect several specific sub-paths instead of one broad one.

If this matters for your estate, tell us. It's a small piece of work, and the shape of the real requirement is what decides whether it should be a path list, a pattern list, or something else entirely.

No publishing route¶

A source with no credential, or no repository destination, cannot receive a link. Kenaptic marks these in the preview list before you publish rather than failing at the end. See Where cross-links are delivered.

Nothing publishes itself unless you asked it to¶

The default everywhere is that a link waits for a person, and that's the setting we'd like you to leave alone. Auto-publish is the one way to change it: per source, above a confidence threshold you choose, switchable only by an administrator, and carrying a warning we make no attempt to soften.

If it's on, that source's links stop waiting for you. Everything else still holds — they're marked, they're reversible, and the audit trail attributes them to the system rather than to a person who never saw them. Turning the switch off withdraws whatever it approved that hasn't gone out yet; anything already live stays live until you explicitly retract it. See How linking stays safe.