fix(channel): keep spam out of topic channel feeds

[mashbean]

Why

Topic channel feeds could surface spam articles because findTopicChannelArticles only filtered article.is_spam, ignoring high spam_score rows where is_spam is still null. The channel assignment path also cleared article.is_spam to false whenever an article was added or re-enabled in a channel, which could override moderation/model state.

This is especially risky after the article spam model was updated and deployed, because high spam_score should now be honored consistently in channel feeds.

What changed

• Use the shared excludeSpam modifier in topic channel article queries, so unpinned channel feeds honor both is_spam and spam_score >= spam_threshold.
• Stop clearing article.is_spam when assigning topic channels.
• Update tests to lock the new behavior and cover the high-score channel exclusion case.

Production audit SQL

with spam_threshold as (
  select value::numeric as threshold
  from feature_flag
  where name = 'spam_detection'
    and flag = 'on'
  limit 1
)
select
  tc.name as channel_name,
  count(*) as leaked_count,
  count(*) filter (where a.is_spam = false and a.spam_score >= st.threshold) as manual_false_high_score_count,
  count(*) filter (where a.is_spam is null and a.spam_score >= st.threshold) as null_high_score_count
from topic_channel_article tca
join topic_channel tc on tc.id = tca.channel_id
join article a on a.id = tca.article_id
cross join spam_threshold st
where tca.enabled = true
  and a.state = 'active'
  and a.channel_enabled = true
  and tc.name in ('生活', '書影音', '時事', '還有')
  and a.created_at >= now() - interval '7 days'
  and (
    a.is_spam = true
    or (a.is_spam is null and a.spam_score >= st.threshold)
    or (a.is_spam = false and a.spam_score >= st.threshold)
  )
group by tc.name
order by leaked_count desc;

Verification

• npm run build
• npm run lint
• npx prettier --check src/connectors/channel/channelService.ts src/connectors/__test__/channelService/topicChannel.test.ts src/connectors/__test__/channelService/findTopicChannelArticles.test.ts

Connector Jest was attempted but blocked in this local environment because Postgres is not running, no MATTERS_PG_* env vars are set, and Docker is unavailable. The failure occurs during genConnections() setup before test assertions run.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.12%. Comparing base (89f67f5) to head (73dd7a6).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4864      +/-   ##
==========================================
- Coverage   73.12%   73.12%   -0.01%     
==========================================
  Files        1081     1081              
  Lines       21644    21641       -3     
  Branches     4735     4735              
==========================================
- Hits        15828    15825       -3     
  Misses       5339     5339              
  Partials      477      477              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mashbean]

Hotfix exception approval recorded on 2026-06-21 Asia/Taipei.

Reason: production topic channel feeds can expose spam-scored articles, and channel assignment can clear existing spam state. User-facing topic channel surfaces including 生活, 書影音, 時事, and 還有 are affected.

SOP evidence:

• Base branch: master
• Head commit: 73dd7a6
• Regression guard after rebasing onto latest origin/master: TARGET_AHEAD=0, HEAD_AHEAD=1, hot-zone empty, deletion-only sentinel scan clean
• GitHub checks: WIP pass, build / build pass for push and pull_request runs, codecov/patch pass, codecov/project pass
• Production action boundary: approval covers hotfix exception and merge path; post-merge production smoke should remain read-only unless a separate moderation mutation is explicitly approved.

[mashbean]

Clarification after reviewing publicationService.ts line 627:

The automatic post-publication path already computes spam from score before channel classification and returns early when the article is spam, so high-score spam should not be sent to automatic channel classification from that path. _detectSpam persists spam_score; the _isSpam decision in post-processing is local unless another path has persisted article.is_spam.

The remaining production risks addressed by this PR are narrower and still valid:

• topic channel feed queries must exclude is_spam IS NULL AND spam_score >= threshold for articles that reach topic_channel_article through other paths
• admin/manual topic-channel writes and topic-channel feedback acceptance must not clear an already persisted article.is_spam = true to false

This also means historical rows with is_spam = false AND spam_score >= threshold may need a separate read-only audit before any data correction, because the query intentionally treats persisted is_spam = false as ham.