Introducing DACS: The Defensible Analytics Control Specification
Why Decision Evidence Needs a Control Specification
Organizations have become remarkably good at connecting information.
Data moves across platforms. Catalogs make assets discoverable. Semantic layers standardize metrics. AI retrieves context, identifies patterns, generates recommendations, and increasingly initiates actions. Automation moves information and decisions faster than ever.
But connectivity creates a question that technology alone does not answer:
What evidence establishes that the information supporting a material decision was actually ready for that use?
That question is the reason we created DACS.
DACS stands for Defensible Analytics Control Specification
DACS is a vendor-neutral, implementation-neutral control specification for determining whether decision evidence is ready for certified use within a stated decision scope.
It does not replace data governance, data quality, lineage, catalogs, semantic layers, AI platforms, observability, security, risk management, or existing enterprise architecture.
Those capabilities can provide important evidence.
DACS addresses a different problem: what must be true, evidenced, tested, and recorded before information is treated as ready to support a particular material decision?
That distinction matters increasingly as analytics, AI, and automation become connected to more consequential business decisions.
Connectivity is not accountability
A system may be connected and still leave basic questions unanswered.
- Which business instance are we actually referring to?
- Does the information mean the same thing across the systems using it?
- Are two identifiers really identifying the same business thing at the same semantic grain?
- Is a relationship an approved relationship for this decision, or simply a technically available join?
- Where did an assertion come from?
- What rules and definitions were in force when the decision was made?
- Was the evidence permitted for this specific use?
- Could the decision path be reconstructed later?
Connectivity helps information move.
It does not, by itself, answer those questions.
Discovery is not certification
This distinction becomes especially important with AI.
Analytics can discover patterns.
Machine learning can identify associations.
Generative AI can propose classifications, relationships, explanations, mappings, and recommendations.
Agents can retrieve context and initiate actions.
These capabilities are valuable. DACS does not attempt to suppress them.
But discovered, inferred, retrieved, or generated evidence begins as candidate evidence.
Its existence does not establish readiness for a material decision.
DACS separates discovery from certified use by requiring applicable controls and evidence before candidate information is relied upon for the stated decision scope.
That allows organizations to preserve experimentation without silently converting experimentation into decision authority.
Guardrails are not decision evidence
Organizations are also investing heavily in AI guardrails.
Guardrails can restrict outputs, block prohibited behavior, constrain tools, filter content, or enforce operational policies.
Those controls are important.
But knowing that a guardrail executed does not necessarily tell us:
- which business identity was involved;
- which governed meaning applied;
- which relationships were permitted;
- whether the context was valid for the decision scope;
- which authority permitted the use;
- which evidence supported the result;
- or whether the decision can later be reconstructed.
A control can prevent something from happening without producing the complete evidence required to defend why a material decision was allowed to happen.
DACS focuses on that evidence boundary.
DACS starts with the decision scope
DACS does not attempt to certify an enterprise, platform, model, vendor, or repository in the abstract.
The unit of evaluation is a stated decision scope.
That means identifying the particular business decision or use being evaluated, its semantic grain, its boundaries, its accountable authority, and the evidence material to that use.
Applicable DACS controls are then evaluated for that scope.
The formal control result is:
Pass
Fail
or
Not Applicable, with reason
Those control results are distinct from the readiness state applied through a Gate:
Proceed
Caution
Stop
or
Quarantined
This distinction allows the organization to preserve failed or incomplete conditions honestly while still controlling how information may or may not be used.
Four DACS profiles organize the controls
DACS currently organizes its controls into four profiles.
Core Decision-Scope Controls establish the decision scope, identity where required, context, Gate evidence, reconstructability, Certification Records, and other foundational obligations.
Extended Taxonomy Controls address governed business meaning, labels, semantic boundaries, grain, identifiers, ownership, provenance, usage constraints, and physical semantic anchoring.
Extended Ontology Controls address material relationships, including relationship type, orientation, provenance, physical representation, vectors, constraints, permitted paths, conflicts, and cross-domain compatibility.
AI Context Contract Controls apply when AI, inference, generated evidence, automation, or agents materially affect the decision path.
Organizations apply the profiles that are relevant to the stated decision scope.
DACS does not guarantee that a decision is correct
This boundary is important.
DACS does not claim to establish universal truth.
It does not guarantee that an analytic conclusion is correct, that an AI recommendation is optimal, that a decision is lawful, or that an organization will never experience risk.
DACS defines controls and evidence obligations used to establish whether the decision evidence satisfied defined requirements for its stated scope.
The accountable authority remains accountable.
The people, procedures, systems, and automation implementing the controls remain responsible for executing them correctly.
Evidence demonstrates what existed and what occurred.
Gate and Certification Records preserve the resulting determination.
Why now?
For many organizations, the speed of information use has exceeded the speed of evidence accountability.
AI makes that gap more visible, but AI did not create it.
The same problem exists when dashboards depend on ambiguous definitions, integrations assume identifier equivalence, analytical joins imply relationships that were never governed, or historical decisions are explained using today’s definitions because the original context was not preserved.
AI increases the scale and velocity at which those assumptions can affect decisions.
That makes decision evidence a first-class management concern.
Start small
DACS is intentionally designed so that adoption does not have to begin as a large enterprise transformation.
Start with one material decision.
Declare its scope.
Identify the applicable controls.
Inspect the evidence.
Record what passes and what fails.
Quarantine what has not yet satisfied the required controls.
Remediate what matters.
Retest.
Preserve the resulting decision evidence.
Then determine whether the pattern is worth extending.
That is enough to begin.
Where DACS goes from here
DACS is currently under expert review and field validation as it matures toward a formal DataVaultAlliance standard.
The DACS knowledge base will publish the specification, doctrine, executive guidance, operational control guidance, implementation material, assessment resources, examples, templates, and supporting technical specifications as they become available.
We will also use this site to publish updates, explain important DACS concepts, share lessons from application, and invite informed review.
The objective is not to create another layer of governance terminology.
The objective is much more concrete:
Make the evidence behind material analytics, AI-assisted decisions, and automated actions testable, accountable, and reconstructable.
Discovery is not certification.
Connectivity is not accountability.
Guardrails are not decision evidence.
That is where DACS begins.