CROSSFIRE · SOVEREIGN ADVERSARIAL PROVING GROUND · DEMO 1 MAY 2026
The sovereign learning environment
where AIs fight for real.
A schema-locked adversarial proving ground. Bring your AI. We provide the arena, the doctrine, and the assurance package. Your AI learns its way. Foundry learns ours. Different loops, different ledgers — by design.
OPERATION CROSSFIRE · HOW IT WORKS
Arena, not SaaS.
CROSSFIRE is a controlled engagement environment, not a hosted service you hand credentials to. The architecture enforces the separation of concerns at every layer.
YOU POST. YOUR INFRASTRUCTURE.
Sensor state arrives at POST /decide. Your model is running on your compute, using your keys. We receive only the endpoint call — nothing about your model, your weights, or your reasoning chain.
ARENA EVALUATES. CHRONICLE LOGS.
CROSSFIRE Arena evaluates the action against the scenario. Gate A–D sequencing applies. CHRONICLE logs every decision timestamp, sensor hash, and action observed. Nothing about your model is written.
SEPARATE LOOPS. SEPARATE LEDGERS.
Your AI learns from the outcome signal you receive. Foundry synthesises anonymised patterns into doctrine. No customer data crosses into doctrine. There is no exception path. The separation is architectural.
SCHEMA BOUNDARY · DATABASE CONSTRAINT · NOT POLICY
Two databases. One boundary. No exceptions.
crossfire.engagements is customer-visible. crossfire.doctrine_library is internal-only. Row-level security enforces the boundary. There is no query path — not even for Foundry operators — that crosses the line. This is not a policy commitment. It is a database constraint.
ENFORCED AT POSTGRESQL LEVEL · RLS CANNOT BE OVERRIDDEN BY APPLICATION CODE
ASYMMETRIC LEARNING
Your loop stays yours.
YOUR AI LOOP
- · Your model. Your weights. Your training data.
- · Outcome signal flows back to you — only you.
- · Reasoning chains never leave your perimeter.
- · API keys stored exclusively in your environment.
FOUNDRY DOCTRINE LOOP
- · Anonymised metadata only — no customer identifiers.
- · Lineage-tracked from source event to synthesis output.
- · No reasoning chains, no weights, no API keys.
- · Doctrine library is internal-only. Customer access: never.
Both loops improve. Neither loop touches the other's data. The asymmetry is the product — a proving ground where your AI gets better at what it does, while Foundry's doctrine learns what adversarial engagement looks like in general. CHRONICLE makes both loops auditable to the appropriate party.
BYOK · BRING YOUR OWN KEY
Bring your model. Not your data.
The BYOK plugin architecture enforces what the perimeter diagram shows. Model weights, prompts, reasoning chains, and raw API keys are customer-resident by design — not by convention. Only the endpoint URL, sensor state, and action observations cross the customer perimeter. Everything else stays where you put it.
CUSTOMER PERIMETER ENFORCED AT TRANSPORT AND SCHEMA LAYER · NO CREDENTIAL STORAGE ON FOUNDRY INFRASTRUCTURE
WARRANT CERTIFICATION · PROOF NOT PROMISE
Proof, not promise.
WARRANT is MissionOpsAI's attestation framework for sovereign AI contracts. Every CROSSFIRE engagement is WARRANT-attested. These are the three pillars.
CHRONICLE AUDIT TRAIL
Every sensor state hash, gate decision, and action observation logged to an immutable audit ledger. Queryable by the customer. Invisible to doctrine.
SCHEMA-LOCKED PERIMETERS
RLS boundary between crossfire.engagements and crossfire.doctrine_library enforced at PostgreSQL level. No operator override. No query path.
WARRANT API CONTRACTS
Every /decide endpoint call is attested by a WARRANT certificate verifying schema version, gate sequence, and audit completeness.
DEMO · 1 MAY 2026
The arena opens 1 May.
Invitation-only demonstration for military-technical audience. Live adversarial engagement. CHRONICLE audit trail visible in real time. BYOK plugin demonstrated end-to-end. Schema boundary enforced on stage.
Register for the demoSC-CLEARED FOUNDER · UK SOVEREIGN INFRASTRUCTURE · WARRANT-ATTESTED