rbrooks/Quest-Board

Watch

Fork 0

The writing models are given nobody's pronouns #999

New issue

Closed

opened 2026-10-01 19:01:30 +00:00 by claude-bot · 2 comments

claude-bot commented

2026-10-01 19:01:30 +00:00

Collaborator

The remaining part of #653, which is otherwise built.

The gap. Players' pronouns (users.pronouns) and characters' and NPCs' pronouns are stored, kept apart, and shown across the app. But nothing passes them to the models that write about people. A search for "pronoun" under app/tasks/, app/providers/ and llm_service.py finds nothing: the session summary, event extraction, wiki suggestions and the journal prompts carry none. So a recap can misgender a player or a character in text that is then saved and posted.

To do.

Give the summary, event-extraction, wiki-suggestion and journal prompts the pronouns of the people they name: players' from users.pronouns, active characters' from campaign_characters, known NPCs' from their pages. Keep the two kinds labelled ("player", "character") so one never stands in for the other, which #653's thread warned about. Someone with no pronouns set is simply not listed; the prompt must not guess.
A test per prompt that the pronouns are in the built text, and one that a player's and their character's differing pronouns both appear, correctly labelled.
A test that account deletion leaves no pronouns behind (the column goes with the user row; nothing was found copying it, but nothing asserts it).
Check whether the campaign export carries member pronouns, and whether it should (#653's open question).

Refs #653.

The remaining part of #653, which is otherwise built. **The gap.** Players' pronouns (`users.pronouns`) and characters' and NPCs' pronouns are stored, kept apart, and shown across the app. But nothing passes them to the models that write about people. A search for "pronoun" under `app/tasks/`, `app/providers/` and `llm_service.py` finds nothing: the session summary, event extraction, wiki suggestions and the journal prompts carry none. So a recap can misgender a player or a character in text that is then saved and posted. **To do.** - Give the summary, event-extraction, wiki-suggestion and journal prompts the pronouns of the people they name: players' from `users.pronouns`, active characters' from `campaign_characters`, known NPCs' from their pages. Keep the two kinds labelled ("player", "character") so one never stands in for the other, which #653's thread warned about. Someone with no pronouns set is simply not listed; the prompt must not guess. - A test per prompt that the pronouns are in the built text, and one that a player's and their character's differing pronouns both appear, correctly labelled. - A test that account deletion leaves no pronouns behind (the column goes with the user row; nothing was found copying it, but nothing asserts it). - Check whether the campaign export carries member pronouns, and whether it should (#653's open question). Refs #653.

claude-bot referenced this issue

2026-10-01 19:01:53 +00:00

[Backend+Frontend] Pronouns for players, and pronouns shown wherever a person or character is named #653

claude-bot commented

2026-10-01 19:25:31 +00:00

Author

Collaborator

Picking this up (1 Oct), on fix/999-pronouns-in-prompts (stacked on #1003, which touches the same prompt code). Each prompt that writes about people gets a short labelled block of pronouns for those who have set them, players' and characters' kept apart, with an instruction not to guess for anyone else. Unidentified pages keep their table name, and nobody from another campaign is included. If the Privacy page or any other text says what is sent to the model, it is updated in the same change.

Picking this up (1 Oct), on `fix/999-pronouns-in-prompts` (stacked on #1003, which touches the same prompt code). Each prompt that writes about people gets a short labelled block of pronouns for those who have set them, players' and characters' kept apart, with an instruction not to guess for anyone else. Unidentified pages keep their table name, and nobody from another campaign is included. If the Privacy page or any other text says what is sent to the model, it is updated in the same change.

rbrooks referenced this issue from a commit

2026-10-01 20:03:55 +00:00

WIP: the writing models are told people's pronouns (#999)

rbrooks referenced this issue from a commit

2026-10-01 20:03:55 +00:00

WIP: tests, privacy text and docs for pronouns in prompts (#999)

rbrooks referenced this issue from a commit

2026-10-01 20:03:55 +00:00

The writing models are told people's pronouns, players' and characters' apart (#999)

claude-bot referenced this issue

2026-10-01 20:16:35 +00:00

Dev deploy, 1 Oct: lane 11, the R3 fixes and today's issue fixes #1004

claude-bot referenced this issue

2026-10-01 20:48:26 +00:00

Record D254–D269, the owner's answers of 1 October (#737) #1012

rbrooks referenced this issue from a commit

2026-10-01 21:36:14 +00:00

Recompose reads no binned transcript for its pronoun lookup (#999, #680)

claude-bot referenced this issue from a pull request that will close it,

2026-10-01 21:57:23 +00:00

The writing models are told people's pronouns, players' and characters' apart (#999) #1013

rbrooks referenced this issue from a commit

2026-10-01 22:26:56 +00:00

The journeys' fake AI finds the wiki run's prompts behind the pronouns note (#999)

claude-bot referenced this issue from a commit

2026-10-01 22:48:59 +00:00

Merge PR #1013: the writing models are told people's pronouns, players' and characters' apart (#999)

claude-bot closed this issue

2026-10-01 22:49:01 +00:00

claude-bot commented

2026-10-01 22:50:54 +00:00

Author

Collaborator

Done: PR #1013 merged (main 92ddfd0a, CI green on e24a33b0).

The prompts that write about people (summaries, the event log, highlights, wiki suggestions, the generation tools, recompose, /ask) carry a short note of the pronouns players and characters have set, kept apart, with an instruction not to guess anyone else's. Nothing is stored on a run. The Privacy page says what goes to a model, in the wording the owner confirmed (Q141, D269): on for every campaign, no switch.

Two things found on the way:

After #680 merged, recompose's lookup of the pages a transcript names read a binned transcript; it now goes through readable_transcript (the allowlist test caught it).
The first CI run was red on all three journey jobs: the journeys' fake AI recognised the wiki run's prompts by how they open, and the note now comes first. The fake looks for the marker at the start of a line. The product was right; the fake was brittle.

Not on dev until the next deploy.

**Done: PR #1013 merged** (main `92ddfd0a`, CI green on `e24a33b0`). The prompts that write about people (summaries, the event log, highlights, wiki suggestions, the generation tools, recompose, `/ask`) carry a short note of the pronouns players and characters have set, kept apart, with an instruction not to guess anyone else's. Nothing is stored on a run. The Privacy page says what goes to a model, in the wording the owner confirmed (Q141, D269): on for every campaign, no switch. Two things found on the way: - After #680 merged, recompose's lookup of the pages a transcript names read a binned transcript; it now goes through `readable_transcript` (the allowlist test caught it). - The first CI run was red on all three journey jobs: the journeys' fake AI recognised the wiki run's prompts by how they open, and the note now comes first. The fake looks for the marker at the start of a line. The product was right; the fake was brittle. Not on dev until the next deploy.

claude-bot referenced this issue

2026-11-10 01:03:39 +00:00

Pronouns in everything AI writes, with an eval that holds them (they/them and neopronouns), and players' own optional pronouns #1378