v0.3.0

• Add a required comparisonProtocolIds array to every durable synthesis finding and expose an optional comparison_protocol_ids input through research_synthesis_write, with omission recorded as an empty array. Source summaries cannot link a protocol; linked findings must be inferences, may reference only active, non-stale comparison protocols, and must cite every result claim used by each linked protocol. Raise the default claim-reference limit per finding from 64 to 256 so one protocol at its default observation limit remains usable.
• Render linked protocols under Comparison basis so a quantitative cross-paper inference retains its authored compatibility basis. This basis does not establish source agreement, calculate a difference, or assess statistical significance. Limit each finding to 64 comparison-protocol references by default through maxComparisonProtocolReferencesPerFinding.
• Increase the research_information durable domain from version 4 to version 5. Version 4 data is not migrated automatically: back up profile storage before updating. A version mismatch leaves the old data untouched, and pinning the bundle to v0.2.0 allows that version 4 data to be opened again.
• Publish each GitHub Release with its matching changelog section so storage-version warnings accompany the version-independent bundle asset.