From equations
to orbital evidence.
r̈ = −μr / |r|³ + perturbations
Keep frames, constants, units, initial conditions, integration method and conservation error attached to the trajectory.

The original NORTHSTAR mark. A guiding star. Connections illuminated. Human purpose at the core.
Specialists advise. The executive core arbitrates.
The executive core arbitrates; specialist regions advise. Connections and online states are illustrative.
DISTRIBUTED INFERENCE COGNITION™
A higher level of synthetic reasoning. Models, mathematics and distributed compute, brought together around one accountable executive core.
Specialists advise. The executive core arbitrates.
Human purpose defines the mission.
THE NORTHSTAR SCIENTIFIC CONTRACT
NORTHSTAR is being designed to return an inspectable evidence path. The question, the method, the calculation and the limits belong with the answer.
Problem. What would count as an answer? Keep the requested result, units and acceptance conditions attached to the inquiry.
Assumptions. Expose what the method assumes. A change in the initial conditions or the scope should be visible in the resulting evidence.
Derivation. Preserve the formulation and the steps that connect it to the calculation. Models may propose a method; the mathematics remains inspectable.
Execution. Record what ran, with which inputs and configuration. An execution receipt identifies an attempt; a successful process alone does not establish a correct result.
Verification. Compare the result with declared tolerances and independent approaches. Preserve failed checks and disagreement rather than smoothing them away.
Uncertainty. Publish the supported conclusion with its uncertainty. Inconclusive, contradicted and insufficient-precision outcomes remain meaningful outcomes.
No material claim without a traceable derivation or compute receipt.
Illustrative evidence contract · select a stage to inspect it. No model or worker is contacted.
THE INTRANET OF CELESTIAL NODES
Models, machines, solvers, data, tools and human expertise. Each admitted participant contributes a bounded capability to one governed mission.
ADMITTED CAPABILITY / 01
Formulate, critique and synthesize complementary lines of inquiry.
Connection is not consensus.
Capacity does not grant authority.
EXPLORE THE CONSTELLATION
Add or remove illustrative participants. Watch inquiry flow through framing, distribution, challenge, verification and synthesis.
DIRECT THE CONSTELLATION
Different questions need different perspectives. Choose a mission to reshape the constellation, then follow an illustrative inquiry from purpose to synthesis.
What evidence would help distinguish competing explanations?
A human question sets the scope. Sources, counterpoints, and methods offer complementary lines of inquiry.
Distributed AI reasoning
NORTHSTAR is Mainely Code’s synthetic reasoning platform in active research and development. Distributed Inference Cognition™ is the approach being developed to coordinate models and specialist contributors around a human-defined question. Contributors investigate, challenge, and verify. The executive core arbitrates and brings supported evidence into one accountable synthesis.
The work is intended for scientific inquiry, institutional research, and other complex questions where evidence, disagreement, and uncertainty deserve examination. Its architecture spans local, private, and cloud compute within explicit permissions. The reasoning vision, model orchestration boundaries, and current evidence and qualification gates explain what is proposed and what remains to be demonstrated.
From the NORTHSTAR application
The same atmospheric lattice used in NORTHSTAR’s chat window. Explore its four activity phases, then shape your own constellation with the interactive nodes.

Choose an activity phase, or let the atmosphere unfold. Add or remove nodes in the interactive constellation to explore a different scale. The node controls change the illustration independently of this footage.
Shape the illustration. The 250-node engineering target awaits final qualification.
SCIENTIFIC RESEARCH DIRECTION
Formulate. Calculate. Simulate. Challenge. Reframe. Independent verification surrounds the loop, and disagreement stays visible.
r̈ = −μr / |r|³ + perturbations
Keep frames, constants, units, initial conditions, integration method and conservation error attached to the trajectory.
Ĥ|ψ⟩ = E|ψ⟩
Make the operator, basis, truncation, normalization, precision and convergence available for expert review.
Ask → compute → interrogate → refine
Engineering, materials, climate, chemistry and public-interest research. Human expertise and real-world validation remain essential.
Research direction, not a claim that these disciplines or executors are production-qualified.
MAKE IT INSPECTABLE
Explore a real browser calculation with its inputs, numerical estimate and residual. Then read the complete public product story.
A larger horizon for AI
NORTHSTAR is being engineered to ship with support for constellations of up to 250 nodes: 1 Commander and up to 249 workers.
Not 250 votes. Not a majority-rule chatbot.
Models, machines, and specialist systems can investigate independently, challenge one another’s conclusions, verify evidence, and return their work to one accountable executive synthesis.
1Commander
The Commander coordinates the mission, preserves authority and provenance, and admits supported findings through one final publication boundary.
EXECUTIVE CONTROL249workers, maximum
Models, machines, and specialist systems can investigate independently, challenge one another’s conclusions, verify evidence, and return their work to the Commander.
DISTRIBUTED REASONINGLAN ⇄ CLOUDomnidirectional resource scaling
From a stack of old laptops on a LAN to private infrastructure and FRONTIER cloud compute, NORTHSTAR is designed to scale reasoning omnidirectionally while keeping authority, provenance, budgets, recovery, and final publication under control.
RESOURCE CONTINUUM“Software engineers once sat at the top of the technology food chain. Then agentic coding arrived—and suddenly we understood how the megalodon must have felt.
The difference is, we get to evolve. The generalist engineer who can direct agents, recognize bad output, connect disciplines, and turn a massive idea into a trustworthy working system has more leverage than ever.”
250-node operation is an initial-release capability being engineered and must complete final qualification before release. This website does not represent a live 250-node constellation or release certification. Many intelligences. One greater purpose. That’s NORTHSTAR.
02 / THE COMPUTE CONTINUUM
From a local machine to private infrastructure and frontier compute, contributions are scoped by capability, policy, and the mission.
01 / DISTRIBUTED INQUIRY
Complementary lines of inquiry unfold across model families and specialists. Evidence travels back toward one accountable synthesis.
The celestial lattice
The lattice represents more than connected machines. It represents a research direction: models contributing to, interrogating, and improving one another’s work.
Divide a consequential question into complementary lines of inquiry.
Have different models challenge claims, assumptions, and missing evidence.
Test claims against sources, experiments, constraints, and independent checks.
Bring supported findings together with provenance and uncertainty intact.
More model outputs do not automatically mean more truth. Correlated errors, shared sources, and model lineage matter.Step inside the lattice
Why this work matters
NORTHSTAR’s horizon is not another productivity shortcut. It is a deeper capacity for scientific inquiry, institutional understanding, and responsible decisions about complex human systems.
Explore competing hypotheses, connect evidence across disciplines, and design more revealing experiments.
Investigate interacting constraints across production, distribution, infrastructure, and resource access.
Examine scenarios, challenge assumptions, and compare trade-offs across interconnected systems.
Help institutions examine complex choices with plural perspectives, traceable evidence, and accountable review.
Illustrative research directions—not claims of demonstrated outcomes. Human expertise, institutions, and real-world action remain essential.
03 / RETAINED AUTHORITY
Identity, data, policy, and evidence stay inside explicit boundaries. Human purpose remains at the center.
Sovereignty is more than location
The ambition is broad participation without a requirement to surrender data, policy, or control to a single model or provider.
The principles behind the workExplicit residency and disclosure rules, with local-only and private-only boundaries when a mission requires them.
Approved capacity, identity-bound nodes, scoped tools, and topology-aware paths—not unrestricted access.
Human-defined purpose, an accountable executive boundary, and results that remain open to inspection and challenge.
Ambition is not a benchmark
NORTHSTAR is in active research and development. This is the direction we are working toward—not a claim that universal interoperability, sovereign deployment, or superior reasoning has already been demonstrated.
04 / EXECUTIVE SYNTHESIS
Supported findings converge. Contradictions, provenance, and uncertainty remain visible to the executive core.
Follow the work
AI researchers, institutions, infrastructure teams, and people who care about where this science is going: follow the work and connect with Mainely Code.
Questions worth asking
NORTHSTAR is Mainely Code’s synthetic reasoning platform in active research and development. Its intended architecture coordinates models and specialist contributors around a human-defined question, returning their evidence to one accountable executive core.
Distributed Inference Cognition™ is Mainely Code’s name for NORTHSTAR’s approach to coordinating inquiry across models and distributed systems. Specialists investigate, challenge and verify findings; the executive core arbitrates and synthesizes the supported evidence.
The intended workflow assigns complementary questions, checks disagreement and source independence, and preserves uncertainty. Repeated agreement among models is not treated as independent proof. The research must be compared with strong single-model and simpler multi-model baselines.
This site describes active research and development, not general availability. The 250-node design is an initial-release target awaiting final qualification. Current implementation states, local evidence, and unresolved external gates are described on the development-status page.
The target is one Commander and up to 249 admitted workers. A node is a logical participant, not necessarily one server, GPU, or model. Node capacity and simultaneous inference concurrency are different limits.
The lattice is a local illustration. It does not contact models, connect to workers, or report measured performance. The application footage and interactive scenes explain the research direction.
NORTHSTAR and Buildroom ADVANCED are Mainely Code’s co-flagships. NORTHSTAR focuses on coordinated synthetic reasoning; Buildroom ADVANCED focuses on ambitious engineering workflows. They remain distinct systems. Any future connection needs an explicit, versioned boundary and its own qualification.
Mainely Code’s co-flagships
NORTHSTAR explores coordinated synthetic reasoning. Buildroom ADVANCED focuses on ambitious engineering workflows. Both are Mainely Code platforms in development, with distinct responsibilities.
A research conclusion might inform an engineering task, but it is not permission to execute one. Any future handoff must preserve an explicit, versioned boundary and separate qualification. Product systems remain external arms; NORTHSTAR’s executive core arbitrates while specialist regions advise.