The missing transaction layer in scientific research.
PARPA, Speculative Technologies, and Science: A New Golden Age approach research organization from different directions. Together they point toward a common missing object: a transaction and routing layer that converts unresolved scientific needs into bounded opportunities executable by distributed capability.
Scientific demand exists below the granularity of jobs, grants, and organizations.
A program may need six hours of specialized judgment, an independent replication, three days on an instrument, a provenance reconstruction, or a four-month team that should disappear when one uncertainty closes. Current research systems often force these needs through much larger persistent containers.
The opposite mismatch occurs simultaneously: useful scientific capability may exist outside the buyer’s institutional network or outside a conventional full-time role. Supply exists. Demand exists. The interface between them is weak.
PARPA supplies the upstream compiler.
Ben Reinhardt’s PARPA proposal centers empowered program managers who identify blockers between the present state and an ambitious objective, formulate hypotheses against those blockers, and coordinate heterogeneous external performers. Early work can remain distributed; tighter internal organization becomes appropriate as dependencies and integration costs rise.
Samarkand inserts below the program manager. The program retains mission ownership and sequencing judgment. Samarkand provisions execution.
Source: PARPA.
Spec supplies the ecosystem topology—and leaves the transition mechanism open.
Unbundling the University argues that many research functions need not remain packaged inside one institutional form. Research Has Customers makes the complementary demand-side point: independent capability matters only when someone wants the resulting scientific output enough to support it.
The more direct adjacency is Reinhardt’s 2026 essay on Bubbling Cauldrons and Transient Strike Teams: healthy ecosystems need mechanisms that can recognize promising bottom-up activity, rapidly compose transient mission-directed teams around it, and let those teams dissolve rather than automatically persist.
Samarkand is neither the cauldron nor the strike team. It is a candidate implementation of the mechanism between them.
The distinctive proposal is that recognition need not depend on a committee deciding ex ante that a bubble is “important.” Sequential successful option exercise can itself produce evidence of ripeness.
Sources: Speculative Technologies library.
Public policy is moving toward individual-first, callable scientific infrastructure.
Science: A New Golden Age calls for broader performer pathways, more flexible allocation mechanisms, wider use of major scientific infrastructure, and sketches a future market-mediated scientific economy involving decomposed problems, contracted execution, remotely accessible laboratories, milestone verification, and reputation.
Samarkand narrows that vision into something testable now: build the human-governed transaction layer before assuming autonomous scientific agents are necessary.
Source: Science: A New Golden Age.
Programs provision Samarkand; Samarkand provisions programs.
A coordinated program contributes mission, real unresolved problems, capital, and program-manager judgment. Samarkand returns performers, facilities, specialized services, verification, price discovery, and option value.
Each transaction also trains the network: what a performer can reliably do, what a facility actually provides, what a transformation costs, and whether repeated work should remain external or become persistent.
Own the inference graph, not necessarily the physical graph.
A program that needs microscopy does not necessarily need a microscope. It needs a scientifically valid microscopy transformation. The same applies to sequencing, fabrication, computation, specialized statistical judgment, and many other capabilities.
Samarkand treats facilities as capabilities in the same addressable graph as people and services.
Do not enumerate every pitfall. Detect departures and route them.
A scientific transaction layer could easily accrete the same remediation bureaucracy it is meant to bypass. Samarkand instead uses a small hazard taxonomy with explicit state transitions.
- Scope ambiguity → paid scoping.
- Performer mismatch → rematch.
- Verification disagreement → independent adjudication.
- Regulatory escalation → qualified institutional host.
- Repeated tight coupling → compose team / internalization review.
- Transaction cost greater than scientific value → exit Samarkand for that work class.
“This should not be a market transaction” is an intended output, not a system failure.
The first Samarkand should be an experiment, not a software marketplace.
Manually broker approximately 20–30 low-risk scientific asks among real buyers and qualified performers. Measure acceptance, scope drift, coordination cost, turnaround, option exercise, and the buyer’s credible alternative.
If the region is small, Samarkand should remain small. If it is large, it is missing scientific infrastructure.