Search runbook assumes Boolean query syntax that Semantic Scholar doesn't support

[melek]

Problem

The search runbook (runbooks/search.md) and the protocol template guide users toward writing Boolean query strings with AND/OR operators and quoted phrases for Semantic Scholar. However, the Semantic Scholar API explicitly does not support Boolean syntax — its search endpoint accepts plain-text queries only.

From the MCP tool description:

"A plain-text search query string. No special query syntax is supported."

This means queries like "response calibration" AND ("human-AI" OR "human-computer") are silently treated as a bag of words, producing unpredictable results rather than the intended Boolean filtering.

arXiv does support Boolean + quoted phrases. paper-search-mcp appears to be plain-text.

Impact

• Users will write Boolean queries for Semantic Scholar during Phase 0 protocol definition, believing they work
• Search results will be unpredictable — sometimes adequate (relevant words present), sometimes missing targeted papers
• The protocol documents a search strategy that cannot be faithfully executed, undermining reproducibility (A7)

Proposed Fix

1. Search runbook: Add a per-database query syntax reference table at the top. Flag Semantic Scholar as plain-text only. Recommend using fieldsOfStudy, year, and venue API parameters for filtering instead of query-level Boolean.
2. Protocol template: Split the search terms table to have database-specific guidance. Semantic Scholar rows should note "plain-text, use API filters" vs arXiv rows noting "Boolean + quoted phrases supported."
3. SKILL.md Phase 0 Field 2: Update the search term guidance to explain the syntax difference between databases during protocol definition.
4. Search agent: Could validate query syntax against database capabilities before executing, logging a warning if Boolean operators are detected in a plain-text-only database query.

Context

Discovered during panel review (2026-03-23). The first live review run hit this immediately.