Stabilize modal rendering and sequential Gemini flow
build-push / docker (push) Successful in 4m13s

This commit is contained in:
AI Assistant
2026-03-16 12:45:12 +09:00
parent 82cead950e
commit c0830b5fde
7 changed files with 145 additions and 125 deletions
+16
View File
@@ -255,6 +255,22 @@
- backend debug broadcasts
## Recent Change Log
- Date: `2026-03-16`
- What changed:
- Search enrichment now runs across the full result set sequentially instead of only enriching a capped top subset in parallel.
- Gemini Vision evaluation now runs candidate-by-candidate with retries and delay, and fallback-only / clearly negative items are excluded from final output.
- Result modal now uses YouTube embed only for Google Video and falls back to source preview video or thumbnail for Envato / Artgrid, avoiding iframe refusal for blocked providers.
- When `GEMINI_API_KEY` is absent, the API still returns ranked fallback results so local smoke tests and non-Gemini environments stay usable.
- Why it changed:
- The user preferred reliability over speed and wanted all candidates processed thoroughly.
- Negative or irrelevant Gemini outcomes were still leaking into the final result set.
- Envato iframe embeds were being blocked, and Google Video modal opening was broken by missing frontend YouTube ID extraction.
- How it was verified:
- code inspection of sequential search/enrichment/evaluation flow
- What is still risky or incomplete:
- Fully sequential Gemini and enrichment processing increases search latency noticeably.
- Some Envato / Artgrid items may still only have thumbnails if public preview URLs are not exposed in source metadata.
- Date: `2026-03-16`
- What changed:
- Result modal layout was rebuilt to match a top `16:9` embedded viewer with bottom-left full AI note and bottom-right action panel.