Institutional & sovereign reasoning
Models investigating, orchestrating, cross-evaluating, and synthesizing across local, private, public-cloud, and sovereign environments.
A clear line between vision and evidence
NORTHSTAR in development.
Evidence defines readiness.
NORTHSTAR is in active research and development toward institutional and sovereign-scale synthetic reasoning. This site explains that vision; it is not a live release dashboard or a declaration of general availability.
Public vision · October 1, 202604 / EXECUTIVE SYNTHESIS
The evidence comes together.
Supported findings converge. Contradictions, provenance, and uncertainty remain visible to the executive core.
HUMAN PURPOSE
DISTRIBUTED INQUIRY
EVIDENCE & SYNTHESISNORTHSTAR / SYNTHESISCONCEPT VISUALIZATION
INVESTIGATECHALLENGEVERIFYSYNTHESIZE
What the website represents
Direction, not a blanket
capability guarantee.
An illustrative celestial lattice
Locally rendered stars, connections, and example workflows. The display is not connected to a fleet and does not report real resource counts or model performance.
Exact, reviewable qualification
Runtime behavior, installed-package acceptance, provider interoperability, security, and reasoning quality must be evaluated for the actual version and environment.
How progress should be measured
More than a test count.
A credible result states the source revision, artifacts, model identities, environment, expected behavior, observed outcome, and remaining limitations. Fixture, simulation, loopback, installed-package, and physical-network evidence are different proof levels.
Reasoning quality
Compare with strong single-model and simpler multi-model baselines. Examine evidence coverage, contradiction detection, calibration, reproducibility, and human usefulness under comparable cost and time constraints.
Resource coordination
Show actual, authorized composition across the claimed workers and providers. Test admission, capacity, quota, billing, cancellation, revocation, restart, and uncertain outcomes.
Scale and the control plane
Support for constellations of up to 250 nodes—1 Commander and up to 249 workers — is an initial-release capability being engineered. It remains unqualified until scheduler throughput, bounded memory, coordination overhead, evidence volume, failure recovery, and real multi-node operation pass the release gates. Thousands remain a future horizon as control-plane hardware and orchestration capacity advance. Simulation does not certify production fleet capacity.
Network behavior
Network systems require independent qualification in their own isolated environments. NORTHSTAR may receive scoped advisory evidence through explicit boundaries; it does not determine effective network state or promote network claims.
Sovereign boundaries
Demonstrate the specific deployment’s data-flow restrictions, identity controls, tenant separation, auditability, and operational ownership. Do not assume geography alone establishes sovereignty.
Operator experience
Verify installation, accessibility, first-run setup, upgrade, interruption, and recovery on the actual supported platforms. Interface messages must match current execution evidence.
Unverified stays unverified.
Missing credentials, labs, independent assessment, or release evidence are reported as remaining gates. They do not become passes, and they do not shrink the vision.
Readiness at a glance
The target, the evidence,
and the remaining gate.
This matrix is deliberately categorical. “Pending” does not become a percentage or a pass.
| Capability | Current implementation state | Local evidence | External qualification |
|---|---|---|---|
| 250-node constellation limit | CODED TARGET 1 Commander + up to 249 workers | CONSOLIDATED RUN PENDING Source and regression work must be qualified together | PENDING Real multi-node environment |
| Commander publication boundary | IMPLEMENTED PATH Advisory evidence remains separate from final publication | LOCAL TESTS | PENDING E2E |
| Mixed local / cloud scheduling | IN DEVELOPMENT | FIXTURES & LOCAL TESTS | PENDING PROVIDER ACCESS |
| Node churn and recovery | IN DEVELOPMENT | LOCAL FAULT TESTS | PENDING AT SCALE |
| Frontier-model interoperability | ADAPTER-SPECIFIC | LIMITED / INERT | PENDING CREDENTIALED LAB |
| LAN / private workers | IMPLEMENTATION PATH | LOCAL / LOOPBACK | PENDING PHYSICAL FLEET |
| Sovereign deployment | ARCHITECTURAL OBJECTIVE | NOT CERTIFIED | PENDING INDEPENDENT ASSESSMENT |
Follow the work, not the hype
Official updates.
Specific evidence.
Follow Mainely Code for development announcements. Any public release should carry its own supported scope, installation guidance, and qualification record.
Clear answers
Questions about the architecture
What does “locally tested” establish?
It establishes only the behavior exercised by the recorded local test and its configuration. It does not certify a provider integration, sovereign deployment, physical fleet, or production release.
Where should release readiness be checked?
Use the evidence for the exact release and environment: source revision, artifacts, model identities, observed outcomes, and remaining gates. This public site does not present itself as a live release dashboard.