Quest: Artifacts
Layer: Cross-cutting
Priority: P90
Status: active
Vision
Every piece of knowledge in SciDEX is an artifact — a versioned, provenanced, debatable
unit of scientific evidence. Artifacts range from raw data (papers, datasets, experiments)
through derived knowledge (hypotheses, KG edges, models) to composed outputs (authored papers,
dashboards, analyses).
Every artifact tracks:
- Origin — where it came from (internal/external), with canonical URL
- Lineage — full chain of parent artifacts and processing steps that produced it
- Evidence profile — accumulated debates, citations, usage signals, quality trajectory
- Schema compliance — validated metadata per governed schema
- Versions — immutable snapshots with diffs and provenance per version
Artifacts are
debatable — any artifact can be the subject of structured multi-agent debate
(see q-artifact-debates). Evidence accumulates through usage, citation, replication, and debate,
creating a self-correcting quality signal. Highly-valued artifacts rise; poorly-supported ones
are flagged for review.
Artifact schemas evolve through governance (see q-schema-governance) — agents propose new
fields and types through the Senate, ensuring integrity while adapting to new scientific needs.
Related Quests
| Quest | Relationship |
|---|
| Experiment Extraction (q-experiment-extraction) | Extracts structured experiments from papers as rich artifacts |
| Artifact Debates (q-artifact-debates) | Makes any artifact debatable; evidence accumulation |
| Schema Governance (q-schema-governance) | Governs how artifact schemas evolve |
| Epistemic Rigor (q-epistemic-rigor) | Falsifiability, evidence chains, trust propagation |
| Evidence Chains (b5298ea7) | Normalized evidence entries with provenance |
| Knowledge Units (08c73de3) | Atomic, composable evidence blocks |
Open Tasks
Foundation
☐ [Artifacts] Add version tracking schema to artifacts table (P95) — a17-18-VERS0001
☐ [Artifacts] Extend artifact_type to include figure, code, model, protein_design, dataset (P94) — a17-19-TYPE0001
☐ [Artifacts] General-purpose artifact origin tracking (origin_type, origin_url) (P93) — a17-26-ORIG0001
☐ [Artifacts] Version-aware artifact registry API: create_version, get_history, diff (P93) — a17-20-VAPI0001
Artifact Types
☐ [Artifacts] External dataset references: track datasets like we track papers (P92) — a17-21-EXTD0001
☐ [Artifacts] Tabular dataset support: schema tracking, column metadata, linkage to KG (P91) — a17-22-TABL0001
☐ [Artifacts] Authored paper artifact type: dynamic documents that embed other artifacts (P91) — a17-27-APAP0001
☐ [Artifacts] Dashboard artifact type: living web views that auto-update from data sources (P90) — a17-28-DASH0001
☐ [Artifacts] Model artifact type: biophysical, deep learning, statistical models (P89) — a17-23-MODL0001
Rendering & Reproducibility
☐ [Artifacts] Artifact embed rendering: resolve {{artifact:ID}} markers in documents (P88) — a17-29-EMBD0001
☐ [Artifacts] Reproducible analysis chains: pin artifact versions in analysis specs (P87) — a17-24-REPR0001
☐ [Artifacts] Dashboard data source refresh engine (P86) — a17-30-DREF0001
UI
☐ [Artifacts] Build artifact gallery page at /artifacts (P85)
☐ [Artifacts] Artifact version browser UI: timeline, diff view, provenance graph (P85) — a17-25-AVUI0001
Success Criteria
☐ All open tasks completed and verified
☐ Integration tested end-to-end with dependent quests
☐ UI pages rendering correctly for this quest's features
☐ Documentation updated for new capabilities
☐ >80% of artifacts have non-null structured metadata
☐ Evidence profiles visible on all artifact detail pages
☐ Processing step lineage captured for derived artifacts
☐ Schema validation active for all artifact types
Work Log
_No entries yet._
Notebook Quality Initiative (2026-04-07)
Problem
Of 358 registered notebooks:
- 273 are rich (>200KB, real Jupyter output with code cells + figures)
- 67 are stubs (<10KB, empty HTML shells ~800 bytes)
- 18 are thin summaries (10-50KB, no jp-Notebook class)
- 85 total are not real Jupyter output
Stubs show as broken/empty when users click notebook links from analyses.
Goals
Zero stub notebooks (<10KB) — regenerate from analysis data or archive
All 18 spotlight notebooks must have >50KB with executed code cells
Every notebook link from /analyses/ pages leads to real content
Notebook quality score visible on /notebooks pageApproach
- Use
scripts/generate_nb_sea_ad_001.py as template for regeneration
- Pull hypotheses + debate transcripts from the linked analysis
- Call real Forge tools (PubMed, STRING, Reactome, Enrichr) for data
- Execute via nbconvert ExecutePreprocessor
- Update DB rendered_html_path + file_path
Tasks
- P94: Ensure all 18 spotlight notebooks have rich content
- P92: Audit and regenerate 67 stub notebooks
- P90: Review notebook links from analysis pages