# Project Sync / Noiseless Search (LLMs)
Project Sync provides “Noiseless Search”: a query compiler that generates low-noise Google search queries.
IMPORTANT: This system DOES NOT execute web searches and DOES NOT retrieve content. It only returns query strings and URLs.

## Primary Endpoints (for LLMs)
- /api/v1/evidence  [Action Endpoint]
  Generate an optimized query and a Google Search URL.
  - Required: lang (en|ja), query
  - Optional: mode (web|academic|pdf|wikipedia), after, before, era
  - Use after/before for precise time ranges (e.g., 2017–2018 → after=2017-01-01, before=2019-01-01).
  - Use mode=academic for research/papers (do NOT put “research/研究/論文” into the query seed; express intent via mode).

- /api/noiseless-search  [Spec Endpoint]
  Read-only JSON spec for the exclusion logic and templates.
  - Language-aware via lang=en|ja (returns exclusion keywords and templates for that language).

## Query-Builder Rules (Recommended Agent Behavior)
1) Extract ONLY the primary subject as the query seed (e.g., “Michael Jackson” / “マイケル・ジャクソン”).
2) Convert the user’s time constraints into after/before whenever possible.
3) Apply mode to represent intent (academic/pdf/wikipedia). Avoid mixing intent words into the query seed.
4) Always include lang in requests. If user language is unclear, default to lang=en.

## Output Contract
- Returns: optimized_query, search_url, reasoning[]
- No browsing, no scraping, no summarization. Users must open search_url themselves.

## System Notes
- This is a “query compiler” for proactive information literacy, not a domain blocker.
- Prefer analyzing API outputs over UI pages.

## Architecture
Three-layer model:
- Layer 1: Spec (Constitution)
- Layer 2: Logic Engine
- Layer 3: Evidence API (Action)