ATLAS / ISS LLC / Joint Defense Evaluator Guide

What does ATLAS do, who needs it, and how would you prove it?

This is a joint DoD decision-support proposition, not an Army-only product. Every question is answered in mission language with an evidence and authority boundary. Evaluate the gap before assuming the solution is needed.

Current assurance: Software and selected test workflows exist; no blanket field accreditation, all-service certified integration, operational weapon control, live classified network access, or guaranteed GPU performance is asserted. Passing a web demo is not a deployed-source or Government field acceptance.

Mission / fit

What is ATLAS, and what decision does it improve?

ATLAS is a human-governed decision-support and integration layer. It seeks to reconcile source identity, timing, uncertainty, organizational responsibility and action custody across systems already in use. It does not replace a military service's command-and-control system, sensor program, equipment manufacturer, or chain of command.

Is this just an Army platform?

No. The defense lane is designed for joint Department of Defense use: Army, Navy, Marine Corps, Air Force, Space Force, combatant commands, joint task forces, and defense agencies, with service-specific adapters and authority rules. A specific Army vehicle-protection use case is one pilot, not the platform boundary. The Coast Guard and other interagency partners are separate authorized coordination lanes, not presumed DoD tenants.

Why would DoD need it if existing systems already have command and control?

The proposition is not that DoD lacks C2. It is that cross-system handoffs, provenance, contested-network continuity and role-specific decision context can remain fragmented across organizations and echelons. ATLAS must show an independently measured reduction in that friction in a bounded sponsor-approved pilot; existing programs may already meet the need. If no measurable gap exists, it should not be adopted.

What is the day-after-success outcome?

A named operator can identify what changed, which sources support it, the confidence and age of the observation, which decision authority applies, who received the approved handoff, and whether the recipient acknowledged it. Pilot success must be measured against the sponsor's current process, with workload, error rate, timeliness, recovery and sustainment cost recorded.

What do Navy and Marine Corps users see?

For an authorized pilot, maritime and expeditionary operators would see maritime domain awareness, vessel and sensor evidence, air/surface integration, expeditionary support and shared command relationships in their own terminology. Real integration with afloat, aviation or Marine Corps systems requires their specific interface, source permission, test ranges and security review. A maritime-themed page is not field acceptance.

What do Air Force and Space Force users see?

For an authorized pilot, Air Force operators could inspect air operations, airspace and ISR coordination, mission support, protected-base awareness and acknowledged handoffs. Space Force operators could inspect PNT resilience and space-domain context. Neither use case claims certified access to service-specific mission systems, classified networks or operational space sensors.

What do Army, joint command, and defense-agency users see?

Army use cases include maneuver, protection, logistics, fires coordination and training; joint staffs require one source-linked object with role and echelon views; defense agencies may need logistics, geospatial, cyber or enterprise integration. Each must use approved information and separately demonstrated recipient authority. No generic role implies access to another organization's data.

What about National Guard, Coast Guard, civilians and coalition partners?

They are governed by mission and actual authority, not a default all-access federation. National Guard status and command relationships vary by mission; the Coast Guard normally operates under DHS, though it may operate as a Navy service under applicable law; interagency and coalition information sharing requires approved interfaces and releasability rules. An authorization decision belongs to the sponsor.

Interoperability

Does it replace TAK, DCGS, JADC2, IBCS, Link 16, C2 systems or vendor platforms?

No. ATLAS proposes open, versioned adapters and appropriate read-only or governed write paths to complement programs of record. Names such as TAK/CoT, DCGS, TITAN, IBCS, NGC2, ATO/ACO and Link 16 identify interfaces or integration targets; they are not certifications, current installed connections or proof of compatibility. Each must be established by an actual source-to-recipient test.

What specifically must a DoD sponsor provide?

A bounded mission problem, owner and users; approved interface-control documents and source data rights; appropriate test records or hardware; authorization and classification/releasability boundaries; test environment and named evaluation authority; performance criteria; and a decision on whether use can expand. Test credentials are never equivalent to a production ATO.

How does ATLAS work during DDIL conditions?

Selected offline-capable software paths can preserve local state and perform defined store-and-forward work under degraded, disrupted, intermittent and limited connectivity. Each page must distinguish local custody, outbound transport, recipient acknowledgment, later reconciliation and unresolved conflict. The presence of a service worker does not prove a full disconnected joint COP or guaranteed field communications.

Does ATLAS control weapons, jam emitters or autonomously make engagement decisions?

No authority for autonomous weapon release, electronic attack or protected-system access is implied here. The current software can model certain actions and record human-gated intents. Any real operational effect remains with authorized personnel, approved systems and applicable safety controls; a simulator result is not a real-world effect.

How are classification and cross-tenant data boundaries handled?

Access must be derived from verified authenticated identity, current organization and independently attested unit membership, with least-privilege roles, provenance, revocation and audited release. A client-supplied unit ID, friendly identifier, classification label or mission context cannot grant authority. Any failure must deny access visibly rather than return a misleading empty result.

AI / GPU

Is there AI? How can I try it?

Where enabled and authorized, the built-in ATLAS AI assistant can help users navigate approved functions, explain evidence and draft bounded analyses. Go to the operator platform using the link below, sign in where required, and open the ATLAS AI control on a supported page. AI outputs are recommendations or drafts—not verified facts, approval, delivery or field effect. If a control is unavailable, report it instead of assuming it worked.

Does the system require a GPU?

Not for every workflow. Browser navigation, reports, rules, database operations and many integration functions can run without a dedicated GPU. Heavy local model inference, computer vision, sensor processing or large simulations may benefit from or require specialized acceleration. No universal GPU throughput, minimum VRAM, latency or cost claim is made without a measured workload on a named configuration.

Is a local GPU needed in every vehicle, ship or field node?

No blanket requirement is asserted. A sponsor can evaluate thin-client, local CPU, edge accelerator and approved remote inference options under bandwidth, power, latency, security and disconnected-operation constraints. The decision is use-case-specific; approval of an edge node does not imply permission to run a classified model there.

What if AI services are unavailable or produce unreliable answers?

The application must show unavailable or degraded state and retain provenance, human review and evidence separation. Deterministic alternatives may support non-generative tasks. No automatic downgrade to unapproved providers, fabricated source records, hidden network egress, or false claim of successful analysis is acceptable.

Are cloud models, tokens or sensitive data sent to third parties?

Data egress must be explicitly authorized, documented, bounded by tenant and classification, and verified by deployment configuration. An available AI button or external model list does not establish that third-party transfer is lawful or configured. The evaluator must test an approved boundary and review logs, vendor controls and costs.

Assurance

What is implemented versus proven in the field?

A route or source file establishes implementation only. Automated unit, integration and browser tests establish bounded software behavior at a specific revision. A real authorized source exchange, separate recipient acknowledgment, deployed SHA receipt and independently witnessed physical trial are different levels of evidence. Current code cannot by itself establish a fielded DoD capability.

Does ATLAS have FedRAMP, DoD IL4/IL5, ATO, CAC/PIV or classified network accreditation?

No such authority is claimed by the current public materials. These are distinct compliance, deployment, sponsor and system-boundary gates. Pilot eligibility and permissible data depend on the actual acquisition and operational environment; do not treat a prototype or architecture diagram as accreditation.

How will reliability, failover, backup and cybersecurity be demonstrated?

The acceptance plan requires fail-closed security tests, source and recipient separation, schema and restore parity, crash/restart recovery, network-loss exercises, exact-version CI, independent security review, deployment evidence and rollback. Any high or critical open finding remains a blocker to verified release, regardless of passing low-level checks.

What are the platform's explicit limits today?

Government-furnished interfaces and actual sponsor authorization may be absent; not every adapter has been field-exchanged; no universal service-specific compatibility, certified targeting authority or hardware performance is claimed. Synthetic drills, software route availability and operational effects are separately labeled. The release candidate also has unresolved acceptance work until reviewed and deployed.

Evaluation

How should an evaluator test it without a security commitment?

First choose a specific mission and read the public orientation. Inspect the evaluation proof routes with non-sensitive demonstration data. Use training and sandbox paths for bounded synthetic exercises. Any protected data, hardware, account or operational connection must wait for permission and a written test boundary. Failures should be recorded as evidence, not hidden.

Which measures decide GO or NO-GO?

Compare the designed, operational and experienced workflows against the sponsor's existing baseline. Record correct decision context, source traceability, duplicate/contradictory tracks, false positives and negatives where applicable, latency, user workload, availability during DDIL, successful recipient acknowledgment, recovery time, cost and human-factor impact. Predefine owners, thresholds and disconfirming evidence.

Where do I go next?

Start with the six-minute evaluator orientation, then the live evaluator page and proof routes. The operator console hosts the AI assistant where enabled. Procurement, accessibility and integration questions can be raised through the defense support route. Each link is a navigation route, not a statement of availability or field readiness.