TIER 4 // KERNEL
Almighty traces every sensor return, effector engagement, and commander order through a directed acyclic graph. Exercise controllers and white cell operators see the full causal chain — not just the timeline.
UNCLASSIFIED // DEMO
Operating Environments
Detect, track, classify, lose_track.
Sensor officers emit radar fans, electronic warfare (EW) cones, MASINT cells, keyhole footprints, and satellite swaths. Each detection commits as a node in the DAG with its causal predecessors named — the radar bearing that produced the contact, the jamming event that degraded confidence. Sensor output feeds Commander decisions; it never engages a target directly.
Engage, suppress, destroy, disable.
Effector verbs are the primary subject of override gateway review and white cell adjudication because they are irreversible. Every engagement records its causal predecessors — the sensor return that cued it, the commander order that authorized it — and emits an indirect fire arc or IR plume artifact for After Action Review (AAR). The white cell can replay the causal chain backward from any effect.
Issue orders, request support, delegate, escalate.
Commanders cannot bypass capability profile constraints — an order to engage out of effective range is rejected at the validator. Intent is auditable end-to-end: every order resolves to the sensor returns that triggered it and the effector engagements it authorized downstream. The DAG makes the chain of decision queryable, not anecdotal.
Three override scopes intercept events before commit.
Per-event always wins. Per-agent-per-turn beats per-turn. The default is review — an empty policy table means everything pauses for white cell decision.
Officer types, action verbs, parameter ranges.
Every entity is bound to a capability profile. Edits create a new profile_id rather than mutating in place. Replays remain bit-identical across thousands of turns.
After Action Review reconstructs a fresh PyRapide DAG.
Replay reads the per-tenant Postgres events table up to any turn N. There is no scrubbing of in-memory state — replay is a re-walk, not a rewind. Every officer decision is reproducible.
Most simulation tools focus on: events.
We focus on: causality.
A timeline tells you what happened. A directed acyclic graph tells you why.
Exercise Control (EXCON) consoles, white cell control surface, AAR replay. Every CZML packet shown to the renderer corresponds to an artifact in the kernel. The unclassified banner reproduces on every visual surface and every exported AAR PDF.
Tenant and scenario registry with role-based access control (RBAC). Turn controller. Override gateway and policy plane. WebSocket fan-out filtered on tenant_id. Per-tenant AWS isolation: Virtual Private Cloud (VPC) subnet, Relational Database Service (RDS) instance, Simple Storage Service (S3) bucket, Key Management Service (KMS) key.
Game time pauses between turns. Blue battalion, red Opposing Forces (OpFor), and the white cell adjudicator agent run sequentially inside a tenant-scoped worker. Each CrewAI agent uses the native PyRapide connector to subscribe to the kernel DAG and adapts its next moves in real time as new causal events commit — sensor return → override decision → effector outcome — instead of replanning from a flat event log. The override gateway filters every emitted event before it reaches the DAG.
Each (tenant_id, scenario_id) pair gets its own Poset. Every event commits with its causal predecessors named — sensor returns, override decisions, effector outcomes — so the kernel records why a step happened, not just when. Cross-namespace leakage is structurally impossible. Officer types, action verbs, and effect artifact taxonomy are all schema-enforced. Replay re-walks the graph in causal order, not insertion order.
Almighty deploys at three postures: synthetic exercise control, live-fire augmentation, and forward-deployed hardened edge. Each tier inherits the previous one. The kernel does not change across tiers — only the surface area, the integration footprint, and the deployment posture.
Exercise Control Tier
For exercise controllers running synthetic wargames inside a closed range LAN.
Live-Fire Tier
For live-fire training where real device telemetry feeds the same DAG.
Hardened Edge Tier
For environments where Python and the cloud are not options.