I spent my first weeks being fast and wrong.

I had scripts, I had an AI helping me, and I could produce more in an afternoon than most
people manage in a week. What I could not do was tell whether any of it was right. A mentor
would send back a correction, I would fix that one thing, and the next batch would come back
with the same problem wearing different clothes.

The lesson underneath all of it, and the thing I would put on a poster for my first day:

On Geo, what hurts you is almost never an error. It is an answer that looks right.

Nothing shouts at you. The page renders. The edit succeeds. The list comes back. You find out
later, when somebody who knows the subject opens what you made.

Here is what I would do differently.

Your personal space first. Nothing there is a proposal, nothing waits on a vote, nothing
embarrasses you. Spend an afternoon making things badly.

Add a few entities. Give them types. Link them together. Put them in a collection, make a page,
add a tab. You are not producing anything anyone will read. You are finding out how the pieces
behave when you touch them, which is knowledge you cannot get by reading about it.

Then take what you learned into a real space and make one small correct edit.

Before adding anything to a real space, open an entity like the one you want to add, and copy
its shape.

Look at what types it carries, which properties those types gave it, and what links into it.
Then match it. Your entity will sit naturally among the others instead of being a lonely
one-off that nobody can find.

Ten minutes taking one real, heavily used entity apart taught me more than a week of reading
did. Searching a type by name does not teach you this. One real example does.

Pick the subject you already follow.

Not the biggest space, not the emptiest, not the one with the most bounties. The one where you
would notice if a fact were wrong.

This matters more than it sounds. In a subject you know, you can tell your good edit from your
bad one before anybody reviews it. In a subject you do not, you cannot, so you wait on someone
else to tell you which it was, and you learn slowly and expensively. I have done both. The
difference is enormous.

You search a name. You do not see it. You create it. Now there are two.

I checked fifteen names any new curator might search on their first day: countries, companies,
the obvious crypto ones. Every single one already existed more than once. For Bitcoin and
Ethereum the copies hold more links between them than the biggest single entity does, so there
is no canonical record of either in any real sense.

The block below holds a few of those pairs. Open two that share a name and compare what links
into each.

What to do about it:

The cost is not the duplicate itself. It is that from then on, half of what is known about a
thing sits on one copy and half on the other, and neither one is right.

I heard "get the ontology right" for weeks before anybody told me what it meant.

It means: this space keeps a list of the kinds of things it has, and how they connect. A Person
works at a Company. A Claim has Sources. That is the whole idea.

Two things I got wrong for a long time:

Most of my output goes through automation, so take this as a confession rather than a lecture.

An AI will do exactly what you asked, at speed, whether or not what you asked for was right. It
cannot tell you that your question was wrong. If you are guessing, it makes you guess faster,
and then it writes your guess into the graph a thousand times.

What works:

This is the part nobody told me, and it is the heart of what I would pass on.

Four things that look like answers and are not:

A list that stops at 100 and does not say so. Ask a busy entity what links to it and you get
100 rows. The main United States entity has 24,073. Nothing in the response mentions the other
23,973. I nearly merged two entities on the strength of a list I believed was complete. Any
time you see exactly 100 of something, be suspicious.

"It published" is not "it is there". An edit reporting success is not the same as an entity
existing, in the space you meant, with the links you intended. I have had edits succeed and
create nothing. Open the thing afterwards and look at it. Every time.

A correction that leaves the mistake behind. Rewriting a value does not always replace it.
More than once the old wording and the new wording both ended up live, with everything still
pointing at the old one. If you fix something, go and check that the thing you fixed is
actually gone.

Two right answers that are wrong together. Check five changes one at a time and every one
can be correct. Apply all five and they conflict, because three of them assumed an entity that
the other two just merged away. Changing several things at once means checking them as a set.

Nearly everything.

Big imports, elaborate pages, your own better idea about how the space should be organised.
None of it is urgent and all of it is easier once you understand the shape of what is already
there.

For your first month, match what the space already does, including the parts you would have
done differently. Earn the right to disagree by first being able to predict what they would do.

Ten correct entities teach you more than one large upload. Nobody minds a small edit. Everybody
notices a duplicate.

If a maintainer takes the time to reject your work and tell you why, that is the fastest
tuition available anywhere. My most useful days were the ones where somebody told me plainly
that what I had made was wrong.

Go slowly for two weeks. Then go as fast as you like.

Asad

The block above holds real duplicate pairs. Open two that share a name and compare what links
into each. For the list that stops short, ask any busy entity for its backlinks and compare the
rows you get against the total it reports.

Counted 28 September 2026 across fifteen common names.