AI Disclosure

1. The one AI call

Clan-Man calls a model in exactly one place: the match statistics import. After a scoreboard is imported, the app's own code works out what looks off about each squad group. If it found something, and if an API key is configured on our side, the server makes a single HTTPS request to OpenAI's API. No key, no call, and the import still succeeds. No findings, no call: of the three real matches we measured, the one the clan won produced none, so nothing was sent.

The call is made by our server, never by your browser. There is deliberately no AI SDK in our dependencies: the request is written out by hand, which is what lets us describe its contents field by field below. Which model answers it is a setting we control.

What the model is for

One sentence. It joins a finding to what that group's admin wrote in their own pre-match brief: the brief said to watch for armour on the left, the finding is about armour losses, so the sentence says the group was briefed on exactly the risk that landed. A model does that because a brief is prose, and no template can read prose.

Every number and every judgement on the report is ordinary code, deterministic and under test. The model does no counting, no ranking and no comparing. It is not deciding what deserves your attention. Code decided that before the call was made.

What comes back

The answer has to match a strict schema, and it is narrowed back to exactly the findings we asked about, so anything invented is thrown away. Any failure at all (no key, a timeout, a malformed answer) produces no sentence, and never a failed import.

2. What we send it, and what we never send

For each squad group with something worth flagging, four things leave the app:

• The group's internal id.
• The group's name.
• The brief its admin wrote.
• The findings the app computed itself: a machine key, plus the sentence the app has already written and already prints on the card in front of you.

That is the entire payload. It is an allow-list written out field by field in one function, not a filter run over a larger object with everything else still attached. That is the difference between a claim you can check and a promise you have to take on trust.

Never sent

• Gamertags. Not your players', not the opposition's.
• Any player row at all.
• Discord identities.
• Kill and death counts, weapon tables, the score, the map, which side you played, member tasks.
• The raw scoreboard payload from the stats site.
• Groups with nothing worth flagging. Those are not sent at all.

We do not use any of it to train anything. Training is OpenAI's to do or not do, so rather than warrant their behaviour: their API terms say data sent through the API is not used to train their models, and we have not opted in to any data sharing.

The one gap in that list

A brief is free text an admin typed, and we do not control what goes in it. An admin who writes a player's name into a group brief sends that name. We cannot strip it, because joining the finding to the brief is the entire point. If that bothers you, keep names out of group briefs.

3. What we commit to

Shipping an AI feature here is deliberate and announced. This one is the first, and the rules written down for it apply to whatever comes after.

• You hear it here first. This page is updated before a feature reaches anyone, and the Last updated date above changes with it.
• Whatever leaves the app gets listed field by field, the way the list above is. If we cannot write that list, the feature does not ship.
• Machine-written text is always marked as such where it appears. Each sentence this produces sits in its own dashed box, labelled Machine read, in a different typeface from the figures around it, and carries the line "Written by a language model from the numbers on this card" underneath it. Model output can be wrong or invented, and nothing we generate is ever presented as a measurement.

Whether this one call earns its place is still an open question, and we are still working out what else AI could usefully do here. We are not pre-announcing features we have not built.

4. Who builds Clan-Man

Two brothers.

• Tailwagthedog, who you may know in game: a long-time Hell Let Loose competitive player for SAS and MSU, and a community streamer for TPL, OSCL and other tournaments. The workflows come from running match nights, not from guessing at them.
• His brother, a professional software developer. He writes the code.

Tools in the workshop

The product is engineered rather than generated: the architecture is deliberate, and a person who writes software for a living reviews what ships. We are not going to tell you that no AI tooling was involved in writing it, because developers use AI assistants now and a blanket denial would simply be false. A tool used while building the app sees code. A model wired into a running product sees your data, and in Clan-Man that is the one call above and nothing else.

Independent of every clan

Naming SAS and MSU raises an obvious question, so here is the plain answer. Clan-Man is not operated by, for, or on behalf of any clan, team, league or tournament, including those two: no clan owns it, funds it, or directs it. Those clans are customers on the same footing as anyone else, and no clan gets preferential access, earlier features, or any view into another clan's data.

Playing against a clan that uses Clan-Man gives nobody a look at their plans, through the app or through the people who run it. Operator access exists to run and support the service, never to read a clan's plans and never for competitive advantage. How clans are kept apart is set out in the Privacy Policy.

5. How to check any of this

Two claims here you can check yourself. Discord lists the bot's permissions and intents in your own server settings: look for the absence of Read Message History and of the Message Content intent. And your browser's network tab shows every host a page contacts, where you will find Discord, us, and Google (these public pages load their webfonts), and no model provider, because the one call is made by our server.

For anything else, such as whether a particular field ever leaves the server, email us: you get an answer from the person who wrote the code. The Privacy Policy lists every third party the running service talks to, and there is no analytics, no tracking and no advertising in Clan-Man. What you agree to by using the app is in the Terms of Use.

6. Contact

Questions about AI in Clan-Man, or about anything on this page: ivan@hyggr.ai. Clan-Man is operated by Syncs App Ltd., a company registered in the United Kingdom.