MOONSHOTS
Big research goals, with the open problems left visible.
Eight long-horizon objectives that organize where the Lab's research machinery is pointed without treating ambition as present-tense capability.
HOW TO READ THIS PAGE
Show where the Lab wants to go without pretending it is already there.
The shared test is practical: current footholds should be visible now, progress should be measurable, and stronger capability claims should wait for prototypes, adversarial review, replication, and broader deployment evidence.
RESEARCH DIRECTIONS
Eight things the Lab is trying to become capable of doing.
- 01Hybrid research infrastructureResearch Operating System
A durable research substrate where humans, neural models, symbolic state, formal tools, experiments, and computation can accumulate coherent work over time.
- 02Networked bounded contributionDistributed Scientific Intelligence
A distributed research network where people, agents, formal solvers, simulations, instruments, and institutions can contribute bounded work to shared state.
- 03Formal tools across disciplinary boundariesMathematical Interoperability
Make more of the mathematics and formal machinery that already exists across disciplines practically translatable, composable, and testable.
- 04Machine-addressable scientific workExecutable Science
Move more claims, assumptions, evidence, experiments, theories, contradictions, and proof obligations into inspectable machine-addressable research objects.
- 05Categorical-algebraic formalizationFormal Representation Mechanics
Develop Boundary Theory into a rigorous grammar-centered framework for decomposition, admissible transformation, closure, defect, and lawful recomposition of representations.
- 06Research that improves its own machinerySelf-Improving Research Infrastructure
A governed feedback loop where successful research can improve the representations, methods, tools, and infrastructure used by later research.
- 07Adversarial theorem and proof-program calibrationMillennium Problems Research
Use the seven Clay Millennium Prize Problems as heterogeneous stress tests for theorem-facing derivation, proof-program analysis, computational probes, and explicit closure and defect machinery.
- 08Fundamental-constant derivation under falsification pressureFine-Structure Constant Program
Use the fine-structure constant as a stringent calibration problem for whether geometric and representational hypotheses can become a parameter prediction without hiding normalization freedom, scale dependence, or fitted structure.
Evidence before rhetoric
How the moonshots are judged
The long-horizon program is useful only if it creates measurable intermediate capability rather than aspirational language.
- 01
Track accepted persistent work per unit of human supervisory effort.
- 02
Measure repair rates, evidence validity, provenance completeness, and downstream reuse.
- 03
Test whether independent agents can continue prior work without reconstructing hidden conversational context.
- 04
Compare cross-domain transfer against native disciplinary baselines rather than conceptual resemblance.
- 05
Treat negative results, demotions, and exposed limits as successful research when they block stronger false claims.
- 06
Keep technical ambition separate from claims about scientific validation, market value, external adoption, or inevitability.
Boundary First Labs — Moonshots: Long-Horizon Research Strategy, 2026-09-20HOW TO READ THE PROGRAM
Moonshots organizes difficult work without turning aspiration into evidence.
The program is useful only when its objectives produce smaller, testable, inspectable work. Negative results and exposed limits count as progress when they prevent stronger claims that the evidence cannot support.
NEXT SURFACES