v0.4.0

• Add durable original-PDF and parser-revision archives, page geometry, and paper_structure for bounded table, formula, and chart extraction inspection. PDF.js remains the default; the optional Docling provider ships a Python worker and requires a separately configured Python environment. Generated extraction remains unverified until checked against the source.
• Add a Research sidebar workspace for the paper library, archived PDF reading, saved positions, notes, evidence matrices, human claim/result reviews, report downloads, and recoverable task checkpoints. A browser bridge registers the plugin's own typed Remote namespace with the published DSH host.
• Add human review decisions with revision checks and history. Add deterministic Markdown, LaTeX, BibTeX, CSL-JSON, and provenance exports; readiness warnings do not constitute publication approval.
• Add three durable research workflows and an opt-in, bounded task execution service with owned logged sessions, stop handling, source-change invalidation, and human-review stops. Execution start/stop/history is available through the trusted workspace API; dedicated browser execution controls are not included yet. The default execution preset is the installed host's minimal preset with the profile's research tools.
• Retain the 0.3 reading pack, digest-paged review renderer, and synthesis comparison-protocol provenance.
• Storage compatibility: the paper-library domain is now version 2 and the research-information domain is version 7, combining human review records with the standalone 0.3 comparison fields. There is no automatic migration from library version 1 or information version 5 (nor from the source application's information version 6). Back up profile storage before updating and use separate fresh storage for 0.4. Unsupported versions are refused without rewriting them; pin v0.3.0 to reopen unchanged 0.3 data.
• DOI/URL/Zotero ingestion, semantic library search, complete scientific-accuracy evaluation, and publication approval remain outside this release.

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.

v0.2.0

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Full Changelog: https://github.com/F1star/dsh-research/commits/v0.1.0