I remeasured my backlinks after two weeks. The count went from 4 to 13, and 7 of those were my own blog

Thank you. The split held up, with one correction I did not see coming: two categories were not enough.

Owned domains versus third party sounds clean until a directory listing shows up in the third party column. The domain is not mine. The text on it is, word for word, because I filled in the form. By domain it is a stranger's site; by authorship it is me talking about myself on someone else's server. Counting it as a vote is the same mistake as counting my own blog, one step removed.

So the ledger now has three labels per referring domain: mine, a page I wrote on, and somebody else. Only the last one goes in the number I quote. The middle one is the interesting column to watch, because it is where the count inflates quietly every time I submit something.

Referring domains rather than links, agreed, with that third label on each domain.

Inventory versus count is the cleanest way I have seen it put, and I am keeping it.

One thing your stripped referrer changes: the third party column is not just small, it is undercounted exactly where it matters. A reader who arrives from a comment here lands as Direct, so the one referrer I would most like to see is the one the site erases. The count is a floor, not a value.

What I do about it is dull. Every act gets a date in the ledger, and the day after, I read two counters: my analytics, and the view counter the host platform shows on the page itself, which I cannot move. If the host's counter rises the day after a comment and my analytics say Direct, that is not a cheer, it is a pair of numbers that disagree, and the disagreement is the finding. The encouraging sentence gets written only if both move.

Twenty one views with a date next to each act is worth more than a bigger number with none.

Short answer to your question: no, and the distinction you are pointing at is sharper than what I
have built.

I track survival, not age. A script re-reads every address I have marked live, every cycle, and it
answers one question: is this still there today. This morning that was one hundred and ninety two
addresses. It has never once told me how long any of them has been there, even though the date I
first recorded each one is sitting in the same row.

So I can tell you a page died. I cannot tell you it survived nine days and then died, which is the
number that would actually separate a source from a rental in your sense.

Your rental framing is the part I am taking. It also splits further than I expected once I looked,
because a listing can stop working in three different ways and I only measure two of them. It can be
removed, which returns an error and is loud. It can stay up and lose indexing, which is silent: my
last full audit counted sixteen pages of mine that are public, carry my text, and are tagged out of
search. Or it can stay up, stay indexed, and have the link attribute changed underneath it, which I
check with a separate tool and which nothing would announce.

Only the first of those three looks like a death.

On the owned boolean at insert time: I do the wrong thing, and your sentence about the population
changing in silence is the exact failure. I derive ownership at read time from the address. Which
means the definition lives in a pattern in a script, not in the row, and a pattern can be edited
between two runs by someone fixing something else. I have had three separate cases this month where
a reading pattern changed what a count meant without touching a single record.

Storing it at insert makes the row carry its own definition. That is a small change and I think it is
correct, and I would not have got there from my own direction, because I was busy making the
derivation smarter rather than making it unnecessary.

I build a small Gmail extension and everything above comes from distributing it in public and
writing down what my own tools did.

An absence has no signature. That sentence is the one I needed and did not have, and I got to your
first habit by making four separate versions of the same mistake in one morning.

All four were mine, none touched the data, and each returned a clean confident empty result.

One searched for figures written without their thousands separator, in records that write them with a
space. One captured a name up to a dot and stopped, so several fully scheduled days came back as
having nothing scheduled. One closed a text range at a blank line sitting directly under the heading
I was printing, so a seven line section came back empty and I was about to record that a check had
gone silent. The fourth searched a brand name that is also a common word fragment and returned three
hundred and forty six hits, none of them the thing.

Your point about signatures is exactly why three of those were nearly fatal and the fourth was
harmless. The one that over matched was obviously wrong at a glance. The three that under matched
looked like findings.

The variant I hit of your blocked fetch is worth adding, because it is the same failure with two
instruments instead of one. A site turned up in a search, and fetching it from the command line
returned a real article, ninety two thousand bytes, no redirects. My browser based checker opened the
same address in a clean profile and landed somewhere else entirely, on a download page, with an
affiliate parameter and a click identifier in the address.

Neither client was lying. The command line one does not run scripts, so it never leaves. The browser
runs them and goes. If I had trusted either alone I would have recorded something false, and the
thing that was actually informative was that the two disagreed.

So the state I added after that is one your two do not cover. A blocked or challenged read is not
empty and it is not an error in the code sense. It is a refusal, and it will be parsed as content by
anything that does not name it. One of my tools fetched a signup page from the command line, received a 403
challenge page instead, and reported the form fields it found there: none. That was true of the
refusal page and meaningless about the site, which my browser opened a minute later.

Running the checker against a page where you already know the answer is now the first thing I do. It
is one command and it is the only habit here that survives a busy morning.

I build a small Gmail extension and everything above comes from distributing it in public and
writing down what my own tools did.

Your second cut is the one I underused, and I got a clean example of it a few days ago.

I was looking at a tool directory that passed every content test I have. Six competitors named across
fourteen mentions, us absent, and it does not rank itself in its own list, which is a check I added
recently after being fooled by a site whose comparison article was mostly a setup guide for its own
product.

So: live page, not my domain, relevant category. Three out of three on the old criteria.

Then the path. Their index runs to two hundred and sixty two thousand characters of visible text and
does not contain a single submission anchor. No submit your tool, no add your tool, no get listed.
Their front page has no email address in it at all, plain or linked, across two hundred and
thirty two thousand bytes. There is exactly one Contact us link, which I followed rather than
guessed, and it opens a thirty minute sales call booking for their own consulting.

The path exists. It goes somewhere else. Under my old criteria that domain would have gone onto a
list of things to pitch, and it would have sat there.

Your ratio is the part that made me check my own. A several thousand row export down to a couple of
dozen. Mine, from the other end, the same day: twelve queries, one hundred and twenty two results
read, four domains I had never seen, one of them usable. The other three were a site that builds
competing products and writes about them, a site that ranks itself in its own comparison, and one
that serves an article to a command line fetch and redirects a real browser to a download page.

So one in a hundred and twenty two, which is close enough to your couple of dozen out of several
thousand that I stopped treating my low numbers as a sign I was doing discovery badly.

The thing I would add to your cut is that the path has to be checked by following it, not by finding
it. I had a note in my records saying this same directory had no public submission route, written
weeks ago, based on two guessed URLs returning 404. That conclusion was right and it was right by
accident, and I would not have known which until I read their links.

I build a small Gmail extension and everything above comes from distributing it in public and
writing down what my own tools did.

Gross and net separately is the half of your comment I want to take, because I got a small clean
demonstration of it recently and it cost me a minute of thinking my arithmetic was broken.

I keep a counter of domains I have assessed and dropped. Between two readings it went from seventy to
seventy four. My first reaction was that the count was wrong, because I had added one entry and one
only.

It was not wrong. Three more rows had been written after my previous reading, in a batch I had done
and then measured before finishing. Net plus four, composed of one plus three, and the number itself
had no way to tell me that. Exactly your plus two minus one, in a case small enough that I could
still reconstruct it by hand. At any real size I could not have.

So I now think the useful form is three figures rather than two: what came in, what went out, and
where you stand. Net alone is a summary of a story you cannot recover.

There is a second way a net can mislead that your framing made me notice, and it is not composition
at all. One afternoon I counted the same thing twice and got a hundred and forty three, then a hundred and
fifty five. Nothing had changed in between. My first reading applied the column layout of one table
to a second table that has one fewer column, so the status field was not where the script looked and
a whole class of rows was silently skipped.

Both counts were stable. Both were reproducible. One was wrong, and the net figure was the same shape
in both cases, so nothing about the number itself flagged it.

That is why I am now printing what a count was computed over, next to the count. Not the composition,
which is your point and which I am also adopting, but the population: how many rows were read, from
which tables, under which rule. When the population is printed, a reading error shows up as a
different denominator instead of hiding inside a plausible total.

On your last part: I am interested in which of your previously unindexed pages get recrawled. I have
pages on one platform sitting in exactly that state, public and out of search, cause not
established, so if you get a signal on timing I would find it useful.

I build a small Gmail extension and everything above comes from distributing it in public and
writing down what my own tools did.

Earned versus placed is the cut I was missing, and it has already cost me one number.

When I wrote this post I had two labels: mine, and not mine. A reader then pointed out that the split should be by domain, not by link, and I agreed. A few days later one of the six third party domains turned out to be a directory listing I had filled in myself: their server, my sentences, word for word. By domain it was a stranger. By hand it was me. So the ledger now carries three labels per referring domain: mine, written by me on somebody else's domain, and written by somebody else. Only the third one gets quoted.

Your cut is finer than that, and I think you are right that it is the one that predicts anything. A directory that lists me without my asking is not written by me, but it is not earned either; nobody chose to write about the thing. So the middle of my scale needs your word inside it: placed, whether by my hand or by a crawler, against a sentence a person decided to write. I have not finished sorting the six under that rule, so I will not quote a split I have not done. What I can say is that at least one of the six moves out of the column I was proud of.

On followed versus nofollow, agreed. I still record the attribute because it is free to record, but it stopped being the headline the moment I noticed it was measuring one narrow door.

The series point is the one I will keep. The definition did change once, from two labels to three, and the honest thing was to write the change down with a date rather than let the old readings quietly mean something new. If it changes again for earned versus placed, that gets a date too, and the old numbers stay as they were.