01 · Freeze
Specification before outcome
Experiments are pinned to a public specification and conformance boundary before the result is known. A failed test cannot be repaired by silently changing the ruler.
How the architecture is separated →1791 Technologies · Los Angeles, California
Independent research in provenance-faithful autonomous systems.
1791 Technologies is the independent research organization behind the Palimpsestus Research Program, an open architecture and experimental program for provenance, warrant, accountable state transition, and autonomous adaptation.
Can an autonomous system change what it believes and does without losing the path that warrants the change?
Research program
A fixed architecture, increasingly hard tests.
The central research bet is that provenance-preserving constraints may produce behavioral regularities that survive changes of implementation and substrate. That is a hypothesis, not an established result.
02 · Test
Positive, null, and unresolved
The evidence ledger preserves useful effects, null controller results, unresolved attribution, and failure conditions rather than collapsing them into a success story.
Inspect the current evidence →03 · Compare
Make different predictions
The next phase targets experiments where Palimpsestus and competing architectures should behave differently under identical observations, actions, memory, and resource constraints.
See the next experiments →Evidence so far
Claims stay inside their warrant.
The program distinguishes architecture conformance, controlled behavioral evidence, official-evaluation milestones, and unresolved attribution.
ARC-AGI-2
192/192 + 160/160
A frozen implementation satisfied 192/192 semantic obligations and 160/160 derivation-ancestry edges. Confirmatory evidence was mixed, including a null controller endpoint result.
ARC-AGI-3 · controlled crucible
0/250 vs. 35/250
On preregistered official ls20, provenance-faithful recurrence produced zero repeated disproven actions versus 35/250 for a matched learner. A su15 control showed no separation; neither architecture achieved level progress.
ARC-AGI-3 · official pathway
publicScore = 0.08
An official Kaggle submission completed with a non-zero public score. Attribution of that score specifically to Palimpsestus agent execution remains unresolved, so no stronger performance claim is made.
Applied research
Different domains. Same frozen ruler.
External systems are useful only if they expose independent ways to test the existing architecture. They are not proof that Palimpsestus “applies everywhere.”
Candidate Crucibles
Autonomous synthetic worlds, aerospace and robotic autonomy, provenance-intensive distributed systems, scientific and engineering workflows, and other environments where state changes under partial information.
Research engagement model
Start with the domain problem in its own vocabulary, freeze the Palimpsestus reference, derive prospective predictions, compare against alternatives, and publish where the architecture succeeds, adds nothing, or breaks.
The domain must consume Palimpsestus. Palimpsestus must not be rebuilt to fit the domain.
Explore applied research →
Identity & ancestry
The mark has a path.
1791's calligraphy, family crest-style mark, and red seal were commissioned from SANTEN Design in 2025. Their design ancestry is preserved as part of the public record rather than flattened into a logo credit.
Trace the design provenance →
Collaborate
Competent disagreement is useful.
The program especially needs people capable of attacking areas that require deeper independent scrutiny.
Formal methods & mathematical systems
Composition, scale, invariance, and conformance.
Autonomous agents & cognitive architectures
Active Inference, NARS/AERA, RL, strong comparators, and goal-directed control.
Distributed systems & provenance
Ancestry, concurrency, compression, fault behavior, and accountability.
Experimental methodology & domain expertise
Preregistration, independent evaluation, aerospace, robotics, games, simulation, and persistent synthetic worlds.