Synthetic worlds
Autonomous populations with history
Can a persistent world remain coherent while thousands of situated AI actors learn, disagree, act, inherit consequences, and preserve different evidence-supported views of the same history?
Applied research
Real systems as external Crucibles.
1791 uses external domains to stress the existing Palimpsestus architecture. The goal is not to force every problem into one theory, but to find where the same structural constraints recur and where they do not.
The domain must consume Palimpsestus. Palimpsestus must not be rebuilt to fit the domain.
Candidate domains
Different kinds of stress.
These are research directions, not claims of solved industry problems.
Aerospace & robotics
Adaptation under partial information
Can autonomous systems revise state and policy under uncertainty while preserving inspectable evidence, fault history, authority, and resource-constrained degradation?
Distributed systems
Accountable state at scale
Can provenance-bearing state transitions preserve causal partial order, distinguish current truth from inherited history, and compress safely without destroying evidence needed for future decisions?
Scientific & regulated systems
Decisions that must survive audit
Can automated decisions remain reconstructible from the exact evidence, rules, authorities, and temporal context that justified them at the time?
Engagement model
Applied work should generate evidence, not just deliverables.
1
Specify the native problem
Start in the domain's own language and identify the real operational constraints before introducing Palimpsestus terminology.
2
Freeze the test
Pin the architecture, define strong alternatives, and preregister outcomes that would distinguish them.
3
Publish the boundary
Record where the architecture helps, where existing methods are sufficient, where costs dominate, and where Palimpsestus breaks.
Engagement terms: Background and project-specific IP are agreed before work begins.
Work with 1791
Bring a hard system, not a request for a demo.
Useful engagements involve consequential state change, partial information, provenance, uncertainty, distributed actors, or long-lived autonomous behavior. A narrow, testable research question is better than a broad “AI transformation” project.