[HosungYou]

A research question can now collect evidence, receive a host-authored cited answer and comparison, and resume without repeating searches through lt research or the MCP research tool. Research sidecars preserve source versions and limitations without changing existing researcher decisions or project state.

Source failures continue independently by default; strict coverage remains explicit. A shared source registry removes readiness mismatches. Ordered Korean/short research terms are preserved, host query refinement keeps the original question, and final comparisons include cited evidence while discovery candidates remain in the packet. Credential URL checks, per-run locking, immutable revisions, and source deadlines protect saved evidence. All packages and internal pins are aligned to 0.1.73; this PR does not publish npm packages.

Validation:

• Full npm test: build, smoke/E2E groups and all workspace typechecks passed.
• 20 fixed synthetic research workflow cases and package checks passed.
• Independent regressions verified credential redaction, concurrent evidence import/retry, strict refresh, query preservation and cited-only comparisons.
• Three live bundles (AI/agents, education, Korean policy) produced bounded source-supported drafts and comparisons; resumed with zero network calls. Full-paper review and independent counterevidence gaps remain explicit; these are not a comprehensive research-quality benchmark.