Outputs do not arrive on common ground.
Each system uses different methods, formats, assumptions, source practices, and documentation standards, making direct comparison difficult.
Vendor-independent operational AI assurance
One evaluation harness across all operational AI systems.
Valid independently evaluates AI-generated military planning outputs across models, vendors, and versions.
Initial use case: per-COA assurance and decision-space assessment for AI-assisted planning.
operational planning assurance consoleRANK · ASSESS · DISTINGUISH · COVER
Valid separates candidate ranking from assurance. Each COA is assessed for constraints, feasibility, suitability, acceptability, completeness, doctrine, and traceability, while the full option set is evaluated for distinguishability, coverage, and duplication.
Rank and assurance are separate: a top-ranked COA can still require human review, while the full option set can pass distinguishability and remain incomplete on coverage.
Multi-vendor operational AI
Multiple AI systems are generating planning content. Who independently assures the result?
Operational staffs increasingly receive planning outputs from different platforms, vendors, foundation models, software versions, and allied systems. Those outputs may conflict, rely on different sources, duplicate the same approach, or omit important alternatives. Without a common assurance layer, staffs are left to reconcile them manually before leaders decide.
More outputs do not guarantee better options.
Several models can produce superficially different plans that share the same assumptions, omissions, dependencies, or operational approach.
Vendors cannot independently assure themselves.
Built-in checks may evaluate one system, but they do not provide a neutral comparison across the full set of planning outputs generated by different tools.
One independent assurance platform
Apply one assurance standard to each planning output and the full decision space.
Valid evaluates individual COAs, assesses the option set as a whole, and preserves the evidence and human adjudication behind every result.
Evaluate individual products.
Check doctrine, mission constraints, factual grounding, unsupported claims, evidence quality, and internal consistency.
Compare systems and alternatives.
Apply consistent criteria across models, vendors, and versions while identifying material changes, redundancy, and missing options.
Preserve the review record.
Link findings to sources, doctrine, model metadata, staff decisions, revisions, and residual risk in one exportable assurance record.
COA assurance use case
More generated options do not always create a better decision space.
In one AI-assisted planning run, Valid evaluated 50 generated COAs, identified 37 distinct operational concepts, collapsed 13 duplicates or near-duplicates, and surfaced the smaller set that warranted further human review.
50
Generated COAs
37
Distinct concepts
13
Duplicates identified
9
Required human review
8
Advanced to the next planning step
2
Missing archetypes
What Valid surfaced
The set contained meaningful alternatives, but it also included redundant concepts and omitted two operational archetypes. Valid therefore evaluates both the quality of each COA and whether the commander is being presented with a sufficiently broad decision space.
MDMP assurance workflow
Validate the process before it becomes orders.
Valid supports an iterative staff workflow at defined decision points. Secure upload is the initial pilot method; connected assurance is the intended operating model. Valid does not replace MDMP, autonomously write the order, or approve a military decision.
Evaluate the current planning package.
During a pilot, staff securely upload mission-analysis products, COAs, wargaming results, comparison products, sources, metadata, or draft orders. Connected systems can deliver the same products directly as the workflow matures.
Surface material planning weaknesses.
Check unsupported assumptions, doctrinal conflicts, weak traceability, redundant or missing COAs, inconsistent analysis, synchronization problems, and material changes across models or versions.
Return findings to the responsible staff section.
Staff accept, reject, escalate, or assign each finding for revision. The responsible section updates the affected planning product and repeats analysis where required.
Preserve the decision and assurance record.
Valid reruns affected checks and records what changed, what was resolved, and what residual risk was accepted before commander approval and order publication.
Recommended assurance gates
Gate 1: before COA approval, evaluate whether the decision space is distinct, complete, supportable, and consistently analyzed. Gate 2: before order publication, verify that the draft order faithfully translates the approved COA and that tasks, timing, resources, control measures, and annexes remain synchronized.
Transparent evaluation
Trace every ranking and finding to the evaluator evidence.
Valid preserves the evaluator trace behind each result, including repeated runs, COA-fit judgments, justification quality, doctrine coverage, constraint checks, unsupported details, and source conflicts. Reviewers can inspect why a result was produced instead of relying on an unexplained score.
How Valid fits
Valid sits between generation and human approval.
Planning platforms and AI tools continue generating operational content. Valid provides a vendor-independent assurance layer that converts submitted or connected outputs into comparable assessments, evidence-backed findings, recommended staff actions, and an audit record.
Start with a bounded pilot
Establish an independent assurance baseline.
Use historical, synthetic, or unclassified planning packages to test whether Valid surfaces material issues missed by the current review process, improves traceability, and adds acceptable review effort at a defined assurance gate.
A bounded planning package
- One or more AI-assisted planning products
- Relevant doctrine, constraints, and source references
- A staff-authored or commander-approved baseline where available
An adjudicated assurance record
- Evidence-backed findings tied to affected products
- Cross-model, version, revision, and decision-space comparisons
- Recommended staff actions and an exportable audit record
Clear operational success criteria
- Material issues found beyond the existing review process
- Finding precision and staff agreement after adjudication
- Time to review, revise, rerun, and advance the product
Deployment path
Begin with secure planning-package uploads. The current product is designed to remain model- and vendor-agnostic. Connected workflows and secure APIs are roadmap capabilities. Secure deployment architecture is designed for future controlled-environment implementation; classified accreditation and additional operational workflows require customer-specific development and validation.
Recommended initial scope: three to five scenarios or planning packages, one defined assurance gate, named staff adjudicators, and pre-agreed measures for issue detection, review effort, and decision usefulness.
Pilot assessment
Get in touch for a pilot assessment.
Tell us what planning AI outputs you need to assess. We will follow up to discuss fit, scope, available data, and a bounded pilot design.