Can a stranger tell what your organisation does?

If you build Nigeria’s startup ecosystem, I have a question.

Can somebody who does not know you work out what your organisation actually does?

Not from your deck. Not from what people in the ecosystem already know about you. From the public record.

Over the last few weeks I read the websites of the organisations building this ecosystem, one at a time. The gap between the work happening and the evidence of that work is enormous.

Organisations running programmes with no dated archive of them.

Partnerships shown as a wall of logos, with no links and no explanation of what the relationship is.

Organisations operating for years that publish no named person you can write to.

And the same description repeated across directories, which did not come from the organisations themselves.

That last one was ours too, and it is where this note starts.

We were publishing other people’s words as yours

Every entry on the record carries a quote describing what an organisation does, and the page presents it as their own words.

On 98 of 111 entries, that quote came from a third party’s member directory.

The source line said so on every page, so nobody was deceived. It was still the record’s central claim resting on borrowed copy in nine entries out of ten.

The obvious fix does not survive contact. Relabelling the source while keeping the text would present someone else’s copy as an organisation’s own with the attribution removed, which is worse than the state I was fixing.

So we replaced the words. Every entry, read again from the organisation’s own live site. All 109 now quote the organisation itself, with the page and the date somebody read it.

Two rules came out of that, and they hold from here.

A directory can point us at an organisation. It never writes the entry. How an organisation is found and how it is recorded are two different questions.

An organisation needs a working website of its own to be listed. Thirteen did not have one and came off the public record. They are unpublished rather than deleted, and they come back the day their site does.

That second rule has a bias in it and I would rather say so than have you find it. Plenty of real Nigerian organisations run on WhatsApp and Instagram and have never owned a domain. A record that requires a website is a record that is slightly more London than Onitsha. I chose it anyway, because the alternative is publishing descriptions no reader can check.

So we stopped asking you to trust our database

Here is what changed, and why each piece had to.

If we say an organisation works in manufacturing, you can see the sentence that made us say it. The page shows the line, the address it was read at, and the date. Clintonel Innovation Centre carries manufacturing because their own site says they design and manufacture STEM kits, and that sentence sits under the label with a link to the page it came from.

That was the largest change under the surface. A category on this record has always carried a grade derived from evidence. A sector carried nothing: a word appeared on a page and nothing recorded who chose it or what they were looking at. That is survivable while the values are inherited from an old import. It stops being survivable the moment one person makes 75 judgement calls in an afternoon, which is what filling that field required.

The write refuses a sector with no source. Not a warning, a refusal. The escape hatch is obvious: allow the value, fill in the source later. Later does not arrive. A record that accepts unsourced judgement in bulk is an opinion with a schema.

That rule holds even for “works across sectors”, which was the tempting exception. It is a true answer for a general purpose hub and a lazy one for an entry nobody read, and the only thing that separates those two is whether somebody can see what you were looking at.

A source is not a grade. Nobody has to prove they work in agritech. Sectors are description, like who you help and where you are. The quote does not make the sector true. It makes the choice checkable, which is a smaller claim and the one we can actually keep.

What an organisation says about itself stays marked as that. A claim made by the organisation is published as their word for it until something dated and public backs it. That is the same grade the same claim would get from us.

And if we got you wrong, you can take the entry

Prove the entry is yours and you get a private editing link. No account, no password.

The record decides how you prove it, from what it already holds about you rather than from what you tell it. If your entry’s website is a real domain and you write from an address at that domain, the link goes there. If your only web presence is a social account, you post a short code on it, which anybody can check from outside. If the address we hold has been taken over, or there is no site, or you write from a personal address, it goes to a person, and we tell you why.

Why no accounts. An account is a thing this record would have to protect, reset, support and eventually leak. It puts a barrier in front of a busy person doing us a favour by correcting our data. And it proves less: a password proves somebody set a password, while an address at your own domain proves control of your own domain.

It might change, and I would rather write that down now. Magic links assume a working inbox at a working domain, and thirteen organisations just came off the record for having no working site at all. If the people who most need to correct us turn out to be the ones this locks out, the mechanism is wrong.

You cannot mark your own homework. You can change everything descriptive. You cannot touch whether the entry is confirmed, how it was confirmed, or whether it is published. Those follow from what happened.

And there is one more thing that changed this week, from watching the first person walk the flow. Proving a claim used to confirm the entry immediately. Click a link, and the page said the organisation had confirmed it, with today’s date, on a page they had not yet read.

Confirmed then meant “somebody at this domain clicked a link”, while every other page of this record insists that checked and confirmed are different words.

So now proving it opens your entry, and saving is what confirms it. That is the first moment anybody can say they looked at what we hold and let it stand. Saving with nothing changed is a fine answer. It means you read it and it was right.

What the ecosystem looks like when you count it

We see this differently when we stop relying on what everybody says and count what is publicly visible.

87 of 109 organisations run a programme you can join.

17 write cheques or connect founders to capital.

78 of 109 work across sectors rather than naming an industry. Most of this ecosystem is general purpose hubs, open to whoever walks in.

29 of 109 publish the name of a human being you can write to. The other 80 have a form or a general address.

Five sector values are still empty: oil and gas, hospitality and tourism, sports, insurance, water and sanitation. 109 organisations were read and none of them work there.

Those are not rankings. They are a picture of what this ecosystem looks like from the outside, today, using only what organisations publish about themselves.

The lopsidedness is the finding. This sector runs things and rarely writes cheques, and a founder looking for money should know that before spending three weeks on applications.

Two things reading 109 websites turned up that I did not go looking for

Five organisations have compromised websites. Four are serving casino link spam inside their own live pages, with their real content intact underneath. The fifth is worse: its own content is buried, and it is associated with a prompt telling visitors to paste a command into their terminal to prove they are human. That command downloads a script, runs it, then clears the screen and deletes itself from the shell history.

If you take one thing from this note and you have nothing to do with this sector: no legitimate website has ever needed you to paste a command into a terminal to prove you are a person.

Four entries pointed at addresses their organisation no longer uses. One hub had moved from a .com to a .ng while the old domain became a parking page. Another had done the same, and its real site sat behind a bot check that made it look dead. I had published two of these as gone. Both are running.

An entry pointing at a dead address usually means the record is stale, not that the organisation is. Every one of those four returns a perfectly healthy response code, and so does the gambling site, which is why anything that monitors this sector cannot be built on status codes.

So: go and find yourself

If you run a hub, a fund, an accelerator, a community, a training organisation, a policy organisation, a coworking space or anything else that supports people building here, open the record and look yourself up.

Read what we say you do. Check the evidence under it. Look at the sectors, the city, and who we say you serve.

Then tell us where we are wrong. Claim the entry and correct it. If your organisation is not there, add it.

One number puts all the others in their place. 109 entries are checked. One is confirmed. That one is ours, from testing our own flow.

Everything here is read by us and dated. Almost none of it has been confirmed by the people it describes, and the machinery for closing that gap now works end to end. Nobody has been asked yet. That is the next note, and I will publish the reply rate whatever it is.

Because the goal is not for SENPAI to become the authority on Nigeria’s startup ecosystem. It is to build a record good enough that the ecosystem can become the authority on itself.

What I got wrong

Four mistakes this week, and they share one shape.

I published two working organisations as gone, on research five days old, without re-checking. I declared three working websites unreadable after giving them three seconds to load instead of six, and reported that as a fact about their sites. I drew the same chart twice to suit the drawing before letting the data set the scale. I wrote a loader that was unsafe to run twice and ran it twice, which put the same paragraph on every entry three times.

Each time I turned something I had not checked recently enough into something the record stated as true.

That is the failure this project exists to fix in everybody else. Three of the four are now impossible by construction rather than by care, which is the only kind of fix that survives a busy week.

---

The data here is limited and does not represent the true state of the ecosystem. It is what 109 organisations publish about themselves, read on a particular week. We will gather more over time and get closer to the real picture, and the fastest way to speed that up is for the organisations themselves to correct it.