Moderation Policy
Last updated: 2026-09-25
This is the public description of how Our Story moderates. It says what is filtered, who can act, what a moderator can see, and how to appeal. The Community Guidelines say what is expected of you; this page says what we do about it.
The short version: most refusals are made by fixed rules that never see who you are, and most enforcement against a person is a removal from one character rather than a ban.
1. What gets filtered, and by what
Text you submit passes through checks that are deterministic: the same input always produces the same answer. There is no model deciding whether you are allowed to speak, no network call, and no stored profile of you feeding the decision.
Four checks run, in this order, and the first one to object is the one that answers:
- Empty, after trimming.
- Length, against a limit that depends on the field. A hero name has a short limit; a command has a longer one.
- Prompt injection. Attempts to instruct the AI to ignore its rules, to reveal its prompt, or to adopt a different persona. This is a list of fixed phrases plus a check for instruction-shaped word pairs appearing close together.
- Blocklist. Slurs and profanity, matched on whole words, including common misspellings, letter substitutions, and spaced-out spellings.
All four match against a normalised form of your text, so changing an i to a
1 does not get a word through.
Two further checks apply to specific fields:
- Third-party marks. A curated list of franchise names, so the generated world does not fill up with characters from somebody else's.
- The verb catalogue. The list of words the game offers you is screened with the same matcher, so the game can never suggest a word it would refuse you for typing.
2. What we check on the way out
The filters above run on what you submit. One check runs on what the AI produces: generated verb narration is checked against the outcome that actually occurred and against the content policy, and a line that fails is not shown.
This is honest about its own limits, so it is worth stating plainly: it covers verb narration and not every generated surface. Location descriptions, objectives, chapter names, mob descriptions and summon blends are protected by the input filters plus the house rules carried in every prompt, and not by this check.
3. Why a refusal does not tell you which rule you broke
When something you submit is refused, the message says it was refused and not what matched. That is deliberate. Naming the rule turns the filter into a tool for finding a way around it, and the people most motivated to do that are the ones the filter exists for.
The trade is real and we would rather state it than hide it: it means an ordinary player who trips a filter by accident gets less help than we would like. If that happens to you, write to us and we will tell you what happened.
4. Who can act against a person
Filtering decides whether a piece of text goes through. It never looks at who wrote it. Acting against a person is a separate system, and the authority to use it is limited:
| Who | What they can do | Against whom | Where |
|---|---|---|---|
| Us, as operator | Ban and remove | Anyone | Everywhere |
| The streamer a hero is bound to | Remove only | Anyone | That hero only |
| An approved community moderator | Remove only | Anyone | Public characters |
| Everybody else | Nothing |
A streamer can never ban an account. They can remove somebody from the hero bound to their channel, and that is the whole of their power. This is enforced in the code that authorises the action rather than by hiding a button.
A removal stops your votes and chat on one character and changes nothing else. A ban is ours alone, and it cuts access to the Service.
5. There is no shadow ban
If you are banned or removed, you are told. You are not left talking to an audience that cannot hear you while the interface pretends everything is normal.
The one exception is at the message level rather than the person level: a single message refused by the filter may be dropped without being shown to others, which denies somebody testing the filters a live signal to iterate against. That is a message being refused, not an account being silently disabled.
6. What a moderator can see
A moderator's view of a character shows a display name and a platform handle, and nothing else. It does not show your email address, your IP address, or your account identifier. It is built that way on purpose, because a moderator's screen can end up on a live stream, and it is tested against exactly that: the test supplies a contributor record carrying an email, an IP and an account identifier, and asserts that none of the three appears anywhere in the rendered page.
The list of who recently took part in a character is kept for a limited time and is capped per character. It is a moderation tool, not a permanent record of who played with whom.
Moderators can read what you submit in the character they moderate, because that content is visible to every participant anyway. Nothing you submit inside the game is private. The Privacy Policy sets this out in Section 11.
7. What we record when something is refused
We record the category of rule that refused it, which character it was aimed at, and which account sent it. We do not record the text that was refused.
There is one deliberate exception, and it runs the other way: when our own generated narration fails the output check, that generated line and the prompt behind it are stored so the failure can be investigated. That is text the Service produced, not text you submitted.
Actions are logged, permanently. Separately from refusals, we keep one record of every moderation action taken against a person or a character, whether a moderator took it or the Service took it automatically. Each record holds the action, who it was against, where it happened, a reason code from a fixed list, and, when a human took it, that moderator's own stated reason. It does not hold what you wrote, and it does not name which rule matched.
These records do not expire, and lifting a suspension does not erase them. That is deliberate: an account whose ban was lifted is exactly the case where "what was done here, and why?" needs an answer, and an audit trail that disappeared when the enforcement did would answer nothing. It is how we can tell you what happened when you appeal.
8. Images
Image prompts are neutralised before an image is rendered, generated images are checked, and any player can flag a rendered image for review. A flagged image can be pulled from the shared world.
9. Appeals
If a report about your account was decided, the decision and the reason the moderator gave are listed for you at Moderation decisions about you, reachable from your account settings. You can appeal from there. There is one appeal per decision, it cannot be withdrawn, and an appeal that was refused for being too long has not used it up.
You can also write to sam@itsmystory.ai from the email address on the account. Say what happened and roughly when. We aim to answer within three business days. The email route stays open for anybody who cannot use the page, and for anything that is not a decision about an account.
If a filter refused something it should not have, tell us the text: that is a false positive and we want to fix the rule, not argue about the message. If you were removed or banned, we will tell you the reason. We will not share another person's report or their identity with you.
10. Changes to this policy
The moderation stack changes as the Service grows, and this page changes with it. The "Last updated" date at the top reflects the most recent revision.