PR #260 wired YouTube comment enrichment against
`/v1/youtube/video/comments` with `id=<video_id>`, but the endpoint
requires `url=https://www.youtube.com/watch?v=<video_id>`. Every enrich
call was returning 400 "missing_parameter: you must provide a url", so
no YouTube items ever carried `top_comments`.
The SC transcript fallback (`_sc_fetch_transcript`) had the identical
contract mistake. It was latent because `_fetch_transcript` prefers
yt-dlp and the SC path only fires when yt-dlp is missing, but it would
have failed the same way on hosts without yt-dlp installed.
Switching both callers to `url=` surfaces a second issue in the
response parser: SC returns `author` as `{"name": "@handle", ...}` and
nests like counts under `engagement.likes`, not top-level. The parser
was reading `author` as a string and missing the nested likes, so even
after the param fix every comment would land with an object-shaped
author and 0 likes.
- `_fetch_video_comments`: send `url=` on both urllib and requests branches
- `_sc_fetch_transcript`: same
- Response parser: extract `author.name` when author is a dict, read
`engagement.likes` when top-level `likes` is absent, prefer
`publishedTime` / `publishedTimeText` for date. Legacy string-author
and top-level-likes shapes still work, so existing mocks are unchanged.
Verified live against api.scrapecreators.com: `_fetch_video_comments`
now returns fully-populated comments with real @handles and like
counts (e.g. "@JennyNicholson: ... (49000 likes, 2025-04-15)"). All
tests in youtube_yt/normalize/signals/render pass.
Plan: docs/plans/2026-04-15-002-fix-youtube-comments-scrapecreators-param-plan.md
🤖 Generated with Claude Opus 4.6 (1M context) via [Claude Code](https://claude.com/claude-code) + Compound Engineering v2.56.1
Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Matt Van Horn
53b8e33d13