Fact suggestions also ask about pages named in the transcript (#1342) #1357

Merged

claude-bot merged 5 commits from feat/1342-mentioned-pages into main 2026-11-06 21:03:57 +00:00

Conversation 0 Commits 5 Files changed 6 +346 -34

claude-bot commented 2026-11-06 08:12:22 +00:00

Collaborator

Copy link

Closes #1342. After the 4.7.0 deploy, the prod backfill asked about nothing, because no wiki pages were linked to those sessions. Now the pass also asks about pages the transcript names.

• Candidates: the session's linked pages, plus every live, non-proposed page in the same campaign with an allowed type whose title or alias the transcript names. It uses the existing mention_pattern and runs two queries per session, however many pages the campaign has.
• Order and cap: linked pages first (by title), then the others by how often they're named. The cap is MAX_PAGES_PER_SESSION = 12, and anything beyond it is counted as skipped_cap on the run. A linked page the transcript never names takes no place: it was never asked about, so it no longer pushes a named page out.
• Over-matching: a page the session didn't link can be found only by a name of at least 3 characters (MIN_MENTION_NAME) that isn't a small word: the existing stop words plus a _COMMON set (she, they, him, one, man…). Linked pages match as before, so a linked "Al" is still asked about. A real one-word name such as "Guard" still matches.
• Unchanged: quote validation, the written-label guard, the bin, erasure and retention rules, and one pending suggestion per session, page and label.
• Re-running prod after this ships: suggest_fact_changes_backfill(force_zero_asked=True) also takes sessions whose last run asked about 0 pages:
  celery -A app.tasks.celery_app call app.tasks.fact_tasks.suggest_fact_changes_backfill --kwargs='{"force_zero_asked": true}'

Tests:

• an unlinked page named in the transcript;
• an alias match;
• the cap order with skipped_cap, and skipped_cap landing on the run;
• linked pages that are never named take no place;
• trashed, bot-proposed, other-campaign and no-facts pages are never candidates;
• short and common titles;
• a short linked title is still asked about;
• force_zero_asked in the service and the task.

The reader guards (SEGMENT_USES, title readers) list the new reads with reasons.

Checks (docker-test, committed code):

• lint is clean.
• tests/wiki/test_fact_suggestions.py, test_transcript_bin_readers.py, test_title_readers.py and test_body_readers.py: 70 passed.
• checks passes.
• Not run: the full backend suite. CI runs it.

🤖 Generated with Claude Code

Closes #1342. After the 4.7.0 deploy, the prod backfill asked about nothing, because no wiki pages were linked to those sessions. Now the pass also asks about pages the transcript names. - **Candidates:** the session's linked pages, plus every live, non-proposed page in the same campaign with an allowed type whose title or alias the transcript names. It uses the existing `mention_pattern` and runs two queries per session, however many pages the campaign has. - **Order and cap:** linked pages first (by title), then the others by how often they're named. The cap is `MAX_PAGES_PER_SESSION = 12`, and anything beyond it is counted as `skipped_cap` on the run. **A linked page the transcript never names takes no place**: it was never asked about, so it no longer pushes a named page out. - **Over-matching:** a page the session didn't link can be found only by a name of at least 3 characters (`MIN_MENTION_NAME`) that isn't a small word: the existing stop words plus a `_COMMON` set (she, they, him, one, man…). Linked pages match as before, so a linked "Al" is still asked about. A real one-word name such as "Guard" still matches. - **Unchanged:** quote validation, the written-label guard, the bin, erasure and retention rules, and one pending suggestion per session, page and label. - **Re-running prod after this ships:** `suggest_fact_changes_backfill(force_zero_asked=True)` also takes sessions whose last run asked about 0 pages: `celery -A app.tasks.celery_app call app.tasks.fact_tasks.suggest_fact_changes_backfill --kwargs='{"force_zero_asked": true}'` **Tests:** - an unlinked page named in the transcript; - an alias match; - the cap order with `skipped_cap`, and `skipped_cap` landing on the run; - linked pages that are never named take no place; - trashed, bot-proposed, other-campaign and no-facts pages are never candidates; - short and common titles; - a short linked title is still asked about; - `force_zero_asked` in the service and the task. The reader guards (`SEGMENT_USES`, title readers) list the new reads with reasons. **Checks (docker-test, committed code):** - `lint` is clean. - `tests/wiki/test_fact_suggestions.py`, `test_transcript_bin_readers.py`, `test_title_readers.py` and `test_body_readers.py`: **70 passed**. - `checks` passes. - Not run: the full backend suite. CI runs it. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

claude-bot added 5 commits 2026-11-06 08:12:23 +00:00

feat(webapp): fact suggestions also ask about pages named in the transcript (#1342) ... 77f834fc2b

Candidates are the session's linked pages plus every live, non-proposed page
of the campaign whose title or alias the transcript names, capped at 12 a
session (skipped_cap counts the rest). Titles under 3 characters and small
words never find a page. The backfill task takes force_zero_asked.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

fix(webapp): a linked page never named takes no capped place (#1342) ... 56fcc65c6e

A linked page the transcript never names has no excerpts and is never
asked, so it must not count against MAX_PAGES_PER_SESSION or push a
named page into skipped_cap. Test included.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

style: ruff format (#1342) ... 07ee87af2a

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

test: list the fourth title read in _candidates (#1342) ... 75c11556c9

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

refactor: keep the linked-page check inside _candidates (#1342) ... 14ba070dd0

The title-readers guard counts reads per function; a nested helper
read as a new reader. Same behaviour.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

claude-bot merged commit 09401007b8 into main 2026-11-06 21:03:57 +00:00

claude-bot referenced this pull request from a commit 2026-11-06 21:03:58 +00:00

Merge pull request 'Fact suggestions also ask about pages named in the transcript (#1342)' (#1357) from feat/1342-mentioned-pages into main

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

backend

Issue related to the backend

  

bot

Issue related to the bot

  

bug

Something is not working

  

duplicate

This issue or pull request already exists

  

enhancement

New feature

  

frontend

Issue related to the frontend

  

help wanted

Need some help

  

invalid

Something is wrong

  

question

More information is needed

  

recording

Issue related to the recording system

  

security

Security hardening / vulnerability work

  

wontfix

This won't be fixed

No labels

backend

bot

bug

duplicate

enhancement

frontend

help wanted

invalid

question

recording

security

wontfix

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

rbrooks/Quest-Board!1357

No description provided.