GSHDW / Applied research
GSHDW treats research as a working practice: find a consequential boundary, build a system that reaches it, then document what survives contact with reality.
The GSHDW research agenda
How should an autonomous tool represent its plan, detect that reality has diverged, recover without compounding the error, and surface uncertainty early enough to matter?
Working question / What evidence should an agent produce before an action becomes trustworthy?
How do people interrupt, redirect, and resume an agent without discarding useful progress or surrendering control to an opaque automation loop?
Working question / Can steering become a first-class interface instead of a failure state?
How can agents move information between documents and structured context while preserving hierarchy, intent, legibility, and a traceable relationship to the source?
Working question / Which document features must remain invariant across transformation?
Method
These are research directions, not publication claims. GSHDW will attach dated notes, evaluations, and reproducible artifacts when they exist. Until then, the strongest evidence is the behavior of the systems themselves.
Start with shad