A mutual provisioning layer for coordinated research.
Research programs routinely need scientific capability at a finer resolution than a job, grant, collaboration, or new organization. Samarkand proposes a transaction and routing layer for those needs.
The missing mechanism between distributed capability and transient coordination.
Recent work at Speculative Technologies describes a healthy research ecosystem as requiring both bottom-up “bubbling cauldrons” and transient mission-oriented strike teams, plus a mechanism that can recognize promising activity and rapidly nucleate the right team around it. Samarkand is a candidate implementation of that third element.
A bubble becomes legibly ripe through successive successful exercises of bounded scientific options.
An Opportunity Ask buys a scientific state transition, not an entire role.
An ask specifies the question, inputs, requested transformation, authority boundary, forbidden actions, acceptance criterion, maximum commitment, and required return object. If the work cannot yet be bounded, the next transaction is paid scoping rather than unbounded labor.
The performer may be an independent scientist, academic laboratory, core facility, CRO, company, federal user facility, computational service, or a temporary bundle of several capabilities.
After verification, the buyer chooses: close, repeat, escalate, compose a team, or internalize. Persistent organization is not forbidden; it is invoked when the work demonstrates persistent coupling.
Programs give the market demand; the market gives programs reach.
A PARPA-, ARPA-, FRO-, foundation-, company-, or investigator-led program provisions Samarkand with mission, real unresolved problems, capital, and program-manager judgment. Samarkand provisions the program with people, facilities, verification, price discovery, and optionality.
Each transaction improves the network by revealing who can do what, which facilities can produce which transformations, what those transformations cost, and which work classes should remain external versus become organizationally persistent.
Do not build the marketplace first.
Manually broker roughly 20–30 low-risk scientific asks among a small number of real buyers and qualified performers. Begin with public-data analysis, replication, pipeline forensics, computational reconstruction, literature-to-experiment synthesis, and similar work whose interfaces can be made explicit.
Measure acceptance-test pass rate, scope drift, coordination cost, turnaround, follow-on option exercise, and what the buyer would actually have done otherwise.
Falsifiable question: Is there a scientifically and economically important region between “leave the problem undone” and “hire or build the whole organization”?