01 — BCDR 02 — Lumbra Risk 03 — MuleSoft PCE 04 — Objections ×8
Technical evaluators · ISSOs · Program architects

Architecture Assumptions

Design decisions, BCDR intent, Lumbra vendor risk, MuleSoft PCE constraints, and Diffuse-Discover-Deliver responses to the eight objections most likely to surface in technical and program office reviews.

01 Architecture Gap

Business Continuity and Disaster Recovery

Issue addressed
No BCDR architecture shown for Game Warden enclave workloads, which are the critical path for the entire platform.

The diagrams show Backup and Restore only for Boundary 02 (Salesforce Government Cloud Plus). No BCDR architecture is shown for the Game Warden enclave workloads. The table at right captures design intent per component.

Note on RTO/RPO targets: these are design intent, not validated commitments. A production BCDR plan requires tabletop exercises, runbook development, and validation testing within the classified environment. This is a mandatory deliverable before initial operating capability (IOC).

Lumbra Nebula
Hard stop: all agentic ops
Active-passive HA via Kubernetes pod scheduling. State is external (Snowflake + CRM). Recovery: pod restart, no data loss.
MuleSoft PCE
Hard stop: cross-boundary calls
Active-active Runtime Fabric across two Game Warden nodes. Anypoint control plane survives individual node failure.
ACDS
Hard stop: all platform ops
Stateless (decisions computed, not stored). HA via Kubernetes replica set (minimum 3 replicas). PAP backed by persistent volume with daily snapshot.
Sovereign Pipeline
Ingestion stops; data intact
Airflow HA via Celery executor with PostgreSQL metadata backend. Airbyte stateless workers restart automatically.
Snowflake Gov
Data warehouse unavailable
Built-in multi-region redundancy. Backup via Snowflake Time Travel (90-day retention).
Elastic
Search and SIEM unavailable
Minimum 3-node cluster. Snapshot and restore to Game Warden object storage (daily).
02 Vendor Risk

Lumbra Nebula Vendor Risk

Issue addressed
Lumbra Nebula is an unverified early-stage vendor. The entire agentic layer carries single-vendor risk with no named fallback.
Risk 01
Sole-source dependency
If Lumbra fails to deliver, is acquired, loses its clearance, or is unavailable for a program re-compete, the orchestration layer has no named alternative. The architecture as drawn cannot survive a Lumbra substitution without re-architecting the agentic layer.
Risk 02
ATO evidence
"In production on C2S" does not constitute a verified ATO for the specific use case described — multi-agent orchestration with Tool Registry PEP behavior, Guardrails enforcement, and ABAC-gated tool calls. Lumbra's C2S ATO evidence package should be reviewed and referenced in any formal proposal.
Risk 03
Pricing leverage
No alternative agentic orchestrator is named. Lumbra has pricing power in any re-compete or sole-source scenario until a credible fallback is named and validated.
Current decision: Architecture proceeds with Lumbra as the single named orchestrator. Salesforce as prime accepts program risk accountability. This decision should be revisited before any formal proposal submission.
03 Procurement Risk

MuleSoft PCE Licensing and Feature Scope

Issue addressed
MuleSoft PCE licensing and feature parity gap not acknowledged in the architecture diagrams. PCE is not a drop-in replacement for standard Anypoint Platform.
Constraint 01
Separate SKU and cost
PCE is not included in standard Anypoint Platform licensing. It requires a dedicated procurement action and is materially more expensive than Runtime Fabric. This is a procurement-level dependency that must be scoped before contract award.
Constraint 02
Feature parity gaps
The following are not available or are limited in PCE: Anypoint Exchange, DataGraph, MQ, and some monitoring and analytics capabilities. Named Credentials, JWT pass-through, ABAC tag propagation, mTLS, and per-call audit must each be confirmed available in the specific PCE version targeted.
Constraint 03
Operational burden
PCE requires the operator to manage Kubernetes infrastructure including control plane upgrades, certificate rotation, and capacity management. Significant overhead compared to CloudHub. The BCDR design in Section 1 must account for PCE control plane availability.
Constraint 04
Version currency
PCE typically lags behind cloud Anypoint Platform by one to two major versions. Features from recent platform releases may not be available in the PCE version at the time of deployment.
Required before proposal submission
  • Confirm the specific PCE version targeted and its feature availability matrix against the architecture's stated capabilities
  • Confirm PCE pricing and procurement vehicle (GSA Schedule, SEWP, or direct)
  • Confirm Kubernetes infrastructure provider (Game Warden, customer-provided, or Second Front-managed)
  • Confirm PCE upgrade and patching responsibility (Salesforce, Second Front, or customer)
04 Pitch Preparation

Common Objections and Responses

The eight objections most likely to surface in technical evaluation and program office reviews. Each is addressed in three steps: Diffuse (take the heat out), Discover (ask what's really behind it), and Deliver (the substantive response). Click any card to expand.

01
This architecture has too many pieces. We will never get this through our ISSO or CAB.
Diffuse Discover Deliver
+
Diffuse

The diagram earns its density. No component was added for elegance. Each one exists because a specific capability gap required it.

Discover

How many systems does your team touch today to produce a single target package? How many are integrated, and how many require a human to re-key data from one to the next?

Deliver

The architecture shows approximately 15 named components because it replaces 5 to 7 disconnected systems and adds an authorization fabric and agentic layer. Each component has a single job.

The ISSO and CAB concern has a specific answer: Second Front's Game Warden is a continuous ATO (cATO) environment. Each container gets an ATO as part of the platform boundary, not through a separate program review. The customer submits one system boundary, not 15. Second Front has operationalized this model across multiple programs. The apparent complexity is in the problem, not in what this architecture invented.

02
This is a CRM. We are not selling anything.
Diffuse Discover Deliver
+
Diffuse

Fair. "CRM" is the wrong frame and Salesforce should retire it for mission contexts. The label creates a false premise before the conversation starts.

Discover

What does your team currently use to track the lifecycle of a target development package from first report through action and final disposition? Who owns that record, and what happens to the institutional knowledge when the analyst who built it rotates out?

Deliver

Sales Cloud is being used as a mission workflow OS, not a sales tool. What it provides: record lifecycle management for targets, sources, and operations; Flow-enforced process steps that cannot be skipped; role-based visibility scoped to clearance and need-to-know; an audit trail from first record creation; and a stable API every component can write to.

The IC has tried Access databases, SharePoint, custom GOTS, and home-built case management. The problem is always the same: who owns the record, who sees which parts, and where does institutional knowledge go when the team rotates. Sales Cloud solves all three in a FedRAMP-authorized, IL5-certified platform with commercial support.

03
If we are buying Salesforce, why does Agentforce not come with it? Why are we paying separately for Airbyte, Snowflake, and Lumbra?
Diffuse Discover Deliver
+
Diffuse

This is exactly the right question. If the Salesforce portfolio were complete and available at IL5 in-enclave deployments today, this architecture would look materially different and be significantly simpler.

Discover

When do you need operational capability? This fiscal year? FY27? The answer changes which architecture is correct.

Deliver

Data Cloud and Agentforce do not currently have IL5 ATO for in-enclave, air-gapped deployment. The compensatory stack exists because the Salesforce portfolio has not yet closed those gaps for classified environments. This architecture is built on what is available, cleared, and deployable today.

If the Salesforce national security team advances Data Cloud and Agentforce to in-enclave IL5 deployment, the compensatory architecture shrinks substantially. That is a business case conversation worth having, and the roadmap questions document frames it explicitly.

04
MuleSoft is the single path for every cross-boundary call. That is a single point of failure.
Diffuse Discover Deliver
+
Diffuse

It is a single path by design. This was an explicit architectural decision, not an oversight.

Discover

What is your current cross-boundary call path? How many paths exist today, how many are fully audited, and how many apply consistent policy enforcement on every call?

Deliver

Every enterprise integration platform has a central hub or it has a mesh. A mesh of point-to-point connections between four accredited boundaries means inconsistent authentication, partial audit coverage, and no single place to enforce or change policy. A mesh also fails silently and asymmetrically.

MuleSoft is not just a reliability choice. It is the policy enforcement point for every cross-boundary call. Removing it means removing ABAC enforcement, JWT pass-through, named credentials, and per-call audit from every integration path simultaneously. MuleSoft is the mechanism by which the architecture keeps its promises about authorization and audit.

05
This integration effort will take three years before we see anything working.
Diffuse Discover Deliver
+
Diffuse

A fully deployed version of this architecture is not a 90-day sprint. That is an honest statement.

Discover

When was the last time your program delivered a working integrated capability in under 12 months? What were the blockers, and are those blockers present here?

Deliver

The architecture phases naturally. Phase 1 deploys Sales Cloud, MuleSoft PCE, and the ABAC fabric. Working mission record management system with policy-enforced cross-boundary access. Timeline: 6 to 9 months to first working capability.

Phase 2 adds the sovereign data pipeline. Phase 3 adds the Lumbra Nebula agentic layer. The 9-minute target package flow is Phase 3. But Phase 1 alone replaces whatever Access database or SharePoint the program is currently using, with commercial support and a continuous ATO. The program sees value before paying for Phase 3.

06
What does this actually cost? We do not have an unlimited budget.
Diffuse Discover Deliver
+
Diffuse

The cost question is completely legitimate and the number is real. This is not an inexpensive platform.

Discover

What does the current fragmented stack cost in licensing, integration contracts, custom development, and analyst hours per year? What is the fully loaded cost per target package today, including the time it takes to produce one?

Deliver

The largest cost variables: MuleSoft PCE (premium over Runtime Fabric), Salesforce Government Cloud Plus (full Sales Cloud licensing), and Lumbra Nebula (cleared startup pricing with limited competitive pressure). Snowflake, Elastic, Airbyte, and Airflow are substantially less expensive than the Informatica IDMC or legacy data warehouse alternatives they replace.

The TCO argument has three legs: (1) this platform replaces 5 to 7 separate contracts, (2) it reduces analyst hours per target package from roughly 9 hours to under 9 minutes across every package the program produces, (3) Second Front's Game Warden bundles ATO and security operations overhead into the platform cost. A precise cost model requires target volume, user count, and data volume.

07
We already have Palantir. Why do we need this?
Diffuse Discover Deliver
+
Diffuse

If Palantir is delivering operational value, that is a real data point and it should be on the table.

Discover

Who owns your data? Is it in Palantir's ontology or in your systems? What are the costs and the timeline if you need to move off the platform or re-compete the contract?

Deliver

Palantir is a capable product. It is also a proprietary data model: your data lives in Palantir's ontology, not in a customer-owned warehouse. Exit costs are catastrophically high by design, which is a known and intentional element of their business model. Palantir AIP does not currently operate in C2S.

This architecture takes the opposite position. Data stays in customer-owned Snowflake and Elastic. Every commercial component is replaceable. Salesforce is the accountable prime but is not the platform monopoly. The right question is not which platform to pick but which data model the program wants to own long-term.

08
Who clears all these vendors? Who supports this in a classified environment after award?
Diffuse Discover Deliver
+
Diffuse

The cleared support model is often more consequential than the technology choices, and it is a reasonable objection to raise early.

Discover

What is the current support model for the program's existing stack? How many of those vendors have cleared support staff on contract today, and what are the response time commitments?

Deliver

The support accountability structure: Salesforce as prime owns accountability for the integrated platform and serves as the single throat to choke. Second Front owns Game Warden operations including cleared staff, continuous ATO management, and underlying infrastructure. Each named subcomponent vendor is responsible for tier-2 support under the prime contract.

Cleared technical staff exist at Snowflake Government and Elastic Federal. Astronomer has federal practice experience. Lumbra Nebula is the most operationally unproven on the cleared support dimension and that risk is acknowledged in Section 2. This is not a stack that requires the customer to develop cleared expertise across eight products. Salesforce prime accountability is the design choice that makes this a managed service.