Method and limitations
How the research prototype represents and grounds design knowledge.
This page explains the current artifact boundary, what its source validation can establish, and where researcher judgement and further corpus curation remain necessary.
Purpose of the artifact
The DSR Knowledge Library is a research prototype for inspecting and reusing represented design knowledge from a curated paper corpus. It supports exploration and comparison; it does not replace close reading of the original publications.
Native OKF representation
Paper records and producer-defined design-knowledge concepts are stored as native Open Knowledge Format Markdown. The local parser preserves open-ended frontmatter and resolves internal Markdown links into a directed graph without forcing every paper into one fixed DSR taxonomy.
Stored knowledge and generated synthesis
Stored concepts come directly from the curated bundle. A generated flow may combine compatible stored knowledge into a proposed direction; those synthesis nodes are marked separately and must not be read as claims that a paper already contains the proposed artifact.
Grounding and citation validation
Local retrieval selects a bounded set of native concepts. Backend code assigns citation IDs, checks generated citations against that allowlist, and validates every diagram source path. This supports traceability to concept records, not page-level verification of an original publication.
Paper design maps and Raw links
The default design map presents the semantic native types actually represented for a paper in readable columns. Raw links is a separate technical view of resolved Markdown relationships, including relationships that are intentionally omitted from the semantic projection.
No-match behavior
When deterministic retrieval finds insufficient library evidence, the assistant returns an insufficient-context result without asking the model to answer from outside knowledge. Weak or absent context is not treated as permission to invent a response.
Corpus-curation limitations
Native format and link validity do not imply complete semantic coverage. Some papers currently represent only a subset of the design requirements, objectives, principles, features, or other knowledge described by their publications. Coverage auditing reports potential gaps for later paper-by-paper curation; the interface does not fill them automatically.
Researcher-evaluation status
The prototype is being evaluated with researchers. Participants should explore the library and grounded assistant before responding in the external survey. Generated output remains non-authoritative, and generated synthesis remains subject to researcher review against cited concepts and the original papers.
Open evaluation survey ↗Operational privacy
Evaluation access is invitation-controlled. The operational access and quota store records counters, token usage, estimated cost, safe outcomes, and privacy-preserving identifiers. It does not persist questions, answers, conversation history, retrieved source text, or credentials. Chat history remains in bounded browser session storage for the current tab.