Unseenet Working Group                                     A. Lorensen
UN-RFC: 0052                                              October 2026
Category: Standards Track
Status: Draft
Implemented: unseenet 0.6.0
Updates: 0004, 0051

                     Unseenet: Tags

Abstract¶

An author may tag their own post with a few words: its topics (politics, linux) and, where it is needed, a content warning (spoiler, violence). Tags are the author's own, signed with the post, like the Keywords: header of Usenet or the hashtags of Mastodon. A group's owner may give a group default tags. Tags let readers follow, find, filter and score posts by topic with their own rules (UN-RFC 0051), and let clients blur what an author warned about. Tags never change postage, ripening or what a node carries. They live only in a header, never in the text, so # keeps meaning a group reference.

1. Motivation¶

Feeds mix many kinds of content, and discussion groups drift. A reader who follows two hundred feeds may want everything about Linux and nothing about football; a reader of a film group wants to be warned before a spoiler. The author knows best what a post is about and what it shows, so the author says it.

Words others attach to a post - by hand or by a program - are labels, with rules of their own (UN-RFC 0053). The same word can be both: politics written by the author is a tag, politics set by a labeller is a label. The content warnings are one shared set (section 4), so a client treats an author's own warning and a labeller's warning alike.

2. Terms¶

Tag : A word an author puts on their own post.

Topic tag : A tag that says what a post is about.

Warning : A tag from the fixed set of section 4, saying what a post shows.

Group tags : Tags a group's owner gives every post in a group (section 5).

3. The Tags Header¶

An article MAY carry

Tags: politics, linux, spoiler

1 to 8 tags, separated by commas, each surrounded by optional white space;
each tag 1 to 32 characters, in Unicode NFC: lower-case letters (any script), the digits 0-9 and -, beginning with a letter or a digit
so økonomi and 3d are tags, Linux, c++ and two words are not;
no tag twice.

A header that breaks these rules is malformed (section 6). An author's client lower-cases and normalises what the author types, and drops what cannot be a tag, before it signs.

The header is part of the signed article (UN-RFC 0002). In a closed group (UN-RFC 0004) and in a private post (UN-RFC 0009) it goes inside the ciphertext with the other inner headers, so nodes never see it.

Tags are set where a post is written: as chips in the compose box, or by writing #word in the subject or the text, as on Mastodon. The author's client puts both in the Tags header when the post is sent - lower-cased and normalised, each once, at most 8 - and leaves the text as the author wrote it. A # that follows a letter, digit, _, @, #, !, /, &, ., + or - does not start a tag; # heading, with a space, is Markdown; and a #word needs a letter, so #170 stays a number.

The header is what counts. Other clients and nodes MUST NOT read tags from a post's text; a client MAY show a #word in the text as a link to everything with that tag. Groups are referred to with ! (UN-RFC 0031 section 3.4.2).

Tags belong on posts, chat lines, feed entries and comments. A manifest, a control article and a reaction (UN-RFC 0048) have none.

4. Warnings¶

The warnings are a fixed set, shared with content labels (UN-RFC 0053):

Warning	The post shows or tells
`profanity`	Swearing, slurs, crude language
`sexual`	Sexual content
`nudity`	Nudity, not sexual
`violence`	Violence, also described
`gore`	Blood, injury, death in detail
`spoiler`	The plot or outcome of a film, book, game or match

The set is small on purpose: a client must be able to offer a choice for each. A new warning needs an amendment to this document.

What a client does with a warning is the reader's choice: show it, blur it behind Show anyway, or hide it. Clients SHOULD blur by default, and MUST NOT hide a post silently (as UN-RFC 0045 section 7.3): what is hidden or blurred says why, one action from being shown. The reader's own posts are never blurred or hidden for them.

A warning is not a sanction. It does not make a post less visible to anyone who did not choose that, and it is never a reason to report a post (UN-RFC 0045).

5. Group Tags¶

A group's manifest (UN-RFC 0004 section 4) MAY hold

Field	Type	Meaning
`tags`	list of tags	1 to 8 tags every post in the group has

for example dk.politik with ["politics"], or an adult group with ["sexual"]. A client treats a post as having its own tags and its group's, and shows them apart: the author's as the author's, the group's as the group's. A group's tags are the owner's statement about the group, not the author's about the post.

5a. The Language of a Post¶

An article MAY say which language it is written in:

Content-Language: da
Content-Language: en, da

as in mail and HTTP (RFC 3282): 1 to 4 primary language subtags of BCP 47 (da, en, de, fil) - 2 or 3 lower-case letters, separated by commas, each once. Like tags it is the author's own statement, signed with the article, inside the ciphertext in a closed group and a private post. A node checks its form and nothing else (section 6).

A group's manifest MAY say which languages are written there:

Field	Type	Meaning
`languages`	list of subtags	1 to 4, the first the group's main one

so dk.snak says ["da"] and an international group ["en"].

How a client chooses. People write in more than one language, by group and by subject, so a client suggests and the author decides. Where the author writes, the client suggests, in this order, the first that says anything (#686, 27 October 2026):

the text itself, when it tells for sure - recognised locally, nothing leaving the machine: by its small words, which tell Danish, Norwegian and Swedish apart and say "don't know" rather than guess, and, where it is installed, by a larger model afterwards, taken only when it is at least 80 % sure;
the group's languages;
the languages the author chose the last time they wrote in that group, remembered in their own settings (UN-RFC 0044), never published;
the author's profile (lang, UN-RFC 0022), only when it names one language.

What the author chose for their chat lines and comments in a group wins over all of this for them there, until they take it back; it is kept with their settings, never published. A comment is too short to tell, so it is suggested the language of the post it answers. A chat line keeps the conversation's language - its newest line's - and only a longer line that the text tells for sure is in another language changes it; "ok" and "haha" never do.

The suggestion is shown under the text, by the tags, and is changed with one action. When nothing says anything, the article gets no header: better none than a wrong one.

The header is a suggestion the author sets, never evidence. Nodes, moderators and labellers MUST NOT rely on it: one can say Danish and swear in English. A labeller decides the language of a text from the text, as a client's recogniser does.

Readers' rules can use it (UN-RFC 0051, lang). A labeller MAY fill in the language of posts without the header as a topic label lang-da (UN-RFC 0053), which counts only for readers who follow it for topics. Gateways map the other network's language tags to the header.

6. Nodes¶

A node checks the form of a Tags header it can read (section 3), and of a Content-Language header (section 5a), and refuses an article with a malformed one (441 syntax), as it checks any other header. It checks a manifest's tags and languages the same way.

That is all a node does with tags. It MUST NOT carry, hide, forward, delete or price an article differently because of its tags, and tags MUST NOT change postage or ripening (UN-RFC 0025). A node MAY let newsreaders ask for the header (HDR Tags, UN-RFC 0005).

7. Clients¶

A client:

shows a post's tags, and its group's, near the post;
lets the author set tags where they write - for example as chips under the text - with the warnings offered first;
can show everything with a tag in the groups and feeds the reader follows;
blurs or hides warnings as the reader chose (section 4);
lets the reader's rules use tags (UN-RFC 0051, tagged).

8. Gateways¶

A gateway (UN-RFC 0024) maps the tags of the other network into the Tags header: Nostr t tags, Mastodon and Matrix hashtags. It lower-cases and normalises them, drops what cannot be a tag, and keeps the first 8. Content warnings of the other network (Mastodon's spoiler_text, Nostr's content-warning) map to the nearest warning of section 4 where one fits, else to none.

9. Rationale¶

Why a header, and # too. The header is what counts: it is signed with the post, read the same by every client, and is what Usenet's Keywords: was. #word in the text is how people tag everywhere else, so the author's client takes it into the header; nobody else reads tags from the text, so a quoted or misread # never changes what a post is tagged with. For that, groups moved from # to ! (UN-RFC 0031).

Why so few and so plain. Eight tags is enough to say what a post is about; more is keyword spam. A simple character set lets rules name tags without quoting (UN-RFC 0051) and makes Linux, linux and LINUX one tag.

Why warnings are fixed. A client offers a setting per warning. With an open set, a reader could never know what they had chosen not to see.

Why tags never sanction. Tags are the author's own words. If they could make a node or a price treat a post differently, honest authors would stop warning.

10. Security and Privacy Considerations¶

In open groups tags are public, like the post. They say what a post is about in a form a program reads easily; an author who does not want that leaves them out.
In closed groups and private posts tags are inside the ciphertext.
An author can mis-tag. Readers' rules weigh tags as the reader wants, and labellers (UN-RFC 0053) can say otherwise.

11. Changes to Other Documents¶

UN-RFC 0004 section 4: a manifest may hold tags (section 5) and languages (section 5a).
UN-RFC 0044: the languages last chosen per group are kept with the user's settings (section 5a, 27 October 2026).
UN-RFC 0051 sections 4.6 and 4.7: the condition tagged (a separate amendment).