Introduction
This section establishes a disciplined evaluation framework for AI-assisted professional work. As AI Knowledge Workers generate drafts, analyses, and recommendations at high speed, the standard for adoption cannot be based on fluency, structure, or confidence of presentation. Professional reliability requires systematic scrutiny. Learners learn to evaluate outputs using repeatable methods that reveal hidden assumptions, test reasoning integrity, and confirm alignment with evidence and organisational constraints.
The section introduces evaluation as an active process of interrogation rather than passive reading. Learners are trained to extract the core claim, identify the logic chain that supports it, and locate the assumptions that must hold true for the output to remain valid. They learn to examine whether the output is grounded in credible sources, whether intermediate reasoning steps are complete, and whether any claims require verification before use in decisions or deliverables.
Learners are then introduced to triangulation as a primary reliability technique. Triangulation strengthens evaluation by requiring confirmation through independent checks. This includes cross-referencing outputs against trusted records, policies, and primary materials, as well as testing the reasoning by requesting alternative methodologies and counter-arguments. Divergence between reasoning paths becomes a signal of uncertainty, prompting deeper human review and, where necessary, escalation to domain expertise.
By the end of this section, Learners will be able to apply a structured protocol for evaluating AI outputs, detect weaknesses efficiently, and approve work products with confidence grounded in evidence, logic integrity, and professional standards.
2.1 The Protocol of Interrogation
2.1.1 Purpose of the Protocol
AI Knowledge Workers can produce outputs that appear complete, well structured, and professionally written. This speed and polish increase productivity, yet they also increase the risk of unexamined adoption. The Protocol of Interrogation is the evaluation discipline used in Cyrenza to prevent silent failure and preserve professional standards. It trains users to evaluate outputs as analytical material that must earn trust through evidence, logic, and alignment with constraints.
Interrogation is not scepticism for its own sake. It is a structured method for determining whether an output is reliable enough to inform decisions, stakeholder communication, and operational action.
Passive reading focuses on comprehension of what the output says. Active evaluation focuses on validation of whether the output should be trusted and used. Learners learn to shift from reading for clarity to reading for integrity.
Active evaluation requires the reviewer to answer four questions:
- What is the output claiming
- What evidence supports the claim
- What assumptions are required for the claim to remain valid
- What constraints or standards must the output satisfy
This approach prevents presentation quality from becoming the basis of trust.
2.1.3 Step One: Clarify the Objective and Decision Context
2.1.3.1 Identify the Decision Question
Every evaluation begins by identifying what decision or deliverable the output is meant to support. Learners are taught to restate the decision question in precise terms.
Examples of decision questions include:
- Should this contract clause be accepted, revised, or rejected
- Which scenario should be used for budgeting under new conditions
- Which operational bottleneck should be prioritised for redesign
- Which marketing hypothesis should be tested next
A clear decision question defines what must be true for the output to be useful.
2.1.3.2 Identify the Audience and Standard of Proof
Different audiences require different standards. A draft for internal brainstorming can tolerate more uncertainty than a board pack, a client recommendation, or a regulatory submission. Learners learn to classify the output by the level of consequence and match evaluation depth accordingly.
2.1.4 Step Two: Extract the Core Claims
2.1.4.1 Identify the Claims That Matter
Learners learn to identify which statements are load-bearing, meaning that if they are wrong, the output becomes unreliable or harmful. These claims usually include:
- Numerical values, thresholds, and comparisons
- Legal interpretations, obligations, and risk classifications
- Causal explanations and driver analysis
- Recommendations and prioritisation choices
- Compliance statements and policy alignment claims
The goal is to focus scrutiny where it matters most.
2.1.4.2 Separate Facts, Interpretations, and Recommendations
Learners are trained to separate the output into three layers:
- Facts: what is asserted as true
- Interpretations: what the facts are said to mean
- Recommendations: what action is suggested
This separation makes it easier to validate the foundation before assessing the conclusion.
2.1.5 Step Three: Identify and Challenge Assumptions
2.1.5.1 Why Assumptions Require Explicit Review
Many AI errors are not caused by incorrect writing. They are caused by implicit assumptions that the reviewer never notices. Learners learn that any conclusion is only as strong as its assumptions.
Common assumption categories include:
- Data completeness assumptions
- Stability assumptions about markets, operations, or behaviour
- Policy and permission assumptions
- Typical case assumptions that may not hold in edge cases
- Time horizon assumptions, such as short-term versus long-term effects
2.1.5.2 Assumption Testing Questions
Learners apply a standard set of questions:
- What must be true for this conclusion to hold
- Which assumptions are uncertain or unverified
- Which assumptions are organisation-specific
- What changes would overturn the recommendation
Assumption testing shifts evaluation from surface reading to structural validation.
2.1.6 Step Four: Verify the Evidence Base
2.1.6.1 Identify the Source of Each Key Claim
Learners verify whether key claims are grounded in:
- Provided documents and internal records
- Organisational policies and standards
- Approved datasets or system-of-record sources
- Verified external references where relevant
If a claim is not tied to a source, it is treated as unverified until proven.
2.1.6.2 Validate the Use of Evidence
Evidence can be present and still misused. Learners review whether the output:
- Uses the correct source for the correct claim
- Represents the source accurately
- Applies the source within the correct context and limitations
- Avoids extrapolating beyond what the evidence supports
This is especially important in legal, financial, and compliance workflows.
2.1.7 Step Five: Inspect the Logic Chain
2.1.7.1 Confirm the Intermediate Steps
Learners are trained to inspect whether the reasoning includes valid intermediate steps between premises and conclusion. Logical integrity requires that the output shows how it gets from inputs to the recommendation.
Where the chain is unclear, Learners request:
- A step-by-step explanation
- A restatement of reasoning in a structured format
- A decomposition of drivers, trade-offs, and dependencies
2.1.7.2 Look for Common Logic Failures
Learners check for:
- Logical leaps without justification
- Over-generalisation where exceptions should apply
- Unstated trade-offs and missing constraints
- Inconsistent reasoning across sections of the output
Logic inspection is essential for preventing plausible outputs from becoming untested decisions.
2.1.8 Step Six: Evaluate Constraint and Governance Alignment
2.1.8.1 Professional Outputs Must Respect Boundaries
Learners review whether the output respects:
- Policy requirements and compliance rules
- Permission constraints and information access rules
- Authority limits, including escalation pathways
- Internal style standards for official deliverables
An output that violates constraints cannot be approved, even if it is analytically strong.
2.1.8.2 Identify Where Escalation Is Required
Some outputs must trigger escalation to a domain professional, such as legal counsel, risk leadership, or finance governance. Learners learn to recognise these triggers and treat escalation as a control mechanism, not a weakness.
2.1.9 Step Seven: Produce an Evaluation Result
2.1.9.1 Classify the Output
Learners classify outputs into one of four evaluation outcomes:
- Approved for use
- Approved with minor edits
- Requires refinement and re-evaluation
- Not acceptable without additional evidence or escalation
This classification makes review decisions explicit and repeatable.
2.1.9.2 Document the Rationale
Professional evaluation includes brief documentation of:
- What was verified
- What assumptions remain
- What risks are accepted
- What changes were required
This supports defensibility and continuity across workflows.
2.1.10 Applying the Protocol Within Cyrenza
Cyrenza supports interrogation by producing structured drafts and enabling iterative refinement through human feedback. The protocol ensures that review remains systematic rather than reactive. Learners learn to use Cyrenza outputs as high-quality starting points, then apply interrogation to confirm accuracy, strengthen reasoning, and ensure alignment with organisational standards before approval.
The Protocol of Interrogation is the practical discipline that keeps augmented intelligence reliable at scale. It protects against silent failure while preserving the speed and capacity advantages of a digital workforce.
2.2 Verification Through Triangulation
2.2.1 Purpose of Triangulation
Triangulation is the discipline of validating an AI-generated output through independent confirmation. In professional environments, a single reasoning path is rarely sufficient evidence for reliability, particularly when the output influences decisions, client deliverables, compliance outcomes, or financial commitments. Triangulation reduces the likelihood of silent failure by forcing the output to survive multiple tests that reveal hidden assumptions, missing constraints, and brittle logic.
Triangulation is not a sceptical posture. It is a professional verification method that increases confidence through structured checks. It ensures that trust is earned through evidence and consistency rather than through presentation quality.
2.2.2 The Core Principle
Triangulation follows a simple principle: no high-impact conclusion should rely on one source, one method, or one narrative. Learners learn to confirm critical claims using at least two independent anchors:
- Independent sources of truth
- Independent reasoning paths
- Independent methodologies or representations
When these anchors align, confidence increases. When they diverge, uncertainty becomes visible and must be resolved through human judgment.
2.2.3 Triangulation Method One: Cross-Checking Against Sources of Truth
2.2.3.1 What Counts as a Source of Truth
Learners are trained to validate outputs against sources that are authoritative for the specific context. Common categories include:
- System-of-record data, such as finance systems, HR systems, CRM records, and claims platforms
- Approved organisational policies, standards, templates, and playbooks
- Signed contracts, legal precedents, and official correspondence
- Primary documents, such as reports, datasets, meeting minutes, and audit trails
- Trusted external references where appropriate, such as regulators, standards bodies, and verified market data
A source of truth is defined by governance, not by convenience.
2.2.3.2 What to Cross-Check First
Not every detail requires verification.Learners focus on the load-bearing elements, including:
- Numerical claims, calculations, and thresholds
- Legal obligations, timelines, and contractual rights
- Regulatory references and compliance statements
- Definitions and scope boundaries
- Recommendations that trigger action or stakeholder communication
2.2.3.3 Evidence Mapping
Learners learn to map each critical claim to its supporting source. If a claim cannot be mapped, it is treated as unverified. This practice strengthens defensibility and improves review efficiency, because the reviewer knows exactly what has been validated.
2.2.4 Triangulation Method Two: Alternative Reasoning Paths
2.2.4.1 Why Alternative Reasoning Matters
An output may appear coherent while depending on weak assumptions or missing steps. By forcing the system to produce a second reasoning path, the reviewer can detect brittleness. If the model produces inconsistent conclusions when reasoning differently, this indicates that the answer is not stable.
Learners learn to treat stability across reasoning paths as a reliability signal.
2.2.4.2 Common Alternative Reasoning Techniques
Learners use several structured approaches:
Stepwise derivation
Request the reasoning as explicit steps, showing intermediate conclusions and dependencies. This exposes leaps, missing assumptions, and logic gaps.
Constraint-first reasoning
Request a reasoning path that begins with constraints, policies, and risk posture, then builds the conclusion inside those boundaries. This tests whether the output respects governance.
Assumption-first reasoning
Request the model to list assumptions first, then derive the conclusion only from those assumptions. This makes hidden dependencies visible.
Quantitative versus qualitative framing
Request the answer using a different representation, such as a table, a decision tree, a risk register, or a scenario matrix. Representation changes can reveal weak links that a narrative format hides.
2.2.4.3 Interpreting Divergence
Divergence across reasoning paths does not automatically mean the output is wrong. It means uncertainty is present and must be resolved. Learners learn to respond by:
- Identifying which assumptions cause the divergence
- Verifying those assumptions against sources
- Narrowing the decision question
- Refining constraints and requesting a new output
- Escalating to domain experts when the consequence level requires it
2.2.5 Triangulation Method Three: Counterpoint and Adversarial Testing
2.2.5.1 The Role of Counterpoint
Professional decisions require awareness of alternative viewpoints, hidden risks, and unintended consequences. Learners learn to request the counterpoint to an argument or recommendation. This tests whether the output can withstand challenge, and it surfaces trade-offs that may not be visible in a single narrative.
Counterpoint testing is especially useful when the output includes recommendations.
2.2.5.2 Structured Counterpoint Prompts
Learners apply disciplined counterpoint requests such as:
- Provide the strongest arguments against this recommendation
- Identify risks and failure modes that could invalidate the conclusion
- List scenarios where the opposite decision would be preferable
- Highlight stakeholders who may object and why
The purpose is not debate. The purpose is completeness and risk visibility.
2.2.5.3 Using Counterpoint to Improve Final Deliverables
Counterpoint outputs often become inputs for refinement, allowing the final deliverable to:
- Include explicit risk notes and mitigations
- Strengthen justification for the chosen direction
- Improve defensibility with stakeholders and governance bodies
- Reduce blind spots and overconfidence
2.2.6 Triangulation Method Four: Independent Replication Within Cyrenza
2.2.6.1 Using Multiple Role-Based Agents
Cyrenza supports triangulation by enabling work to be replicated by different role-based Knowledge Workers. Learners learn to request independent replication from a second agent with a different professional orientation.
Examples include:
- Finance analysis replicated by an Internal Auditor agent
- Contract review replicated by a Compliance and Ethics agent
- Marketing recommendations replicated by a Market Research Analyst agent
- Insurance triage replicated by a Risk Assessment Officer agent
Replication improves reliability because it introduces alternative reasoning lenses and increases the chance that weaknesses are caught.
2.2.6.2 Peer Review as a Digital Workforce Pattern
Learners learn to treat second-agent checks as a digital equivalent of peer review. This reduces individual agent bias and improves consistency across outputs.
2.2.7 A Practical Triangulation Protocol
2.2.7.1 The Minimum Standard for High-Impact Outputs
Learners apply a minimum standard based on task consequence. For high-impact outputs, the minimum includes:
- Source cross-check of all load-bearing claims
- Alternative reasoning path for the main conclusion
- Counterpoint or risk challenge for recommendations
If any of these steps surface uncertainty, the output returns to refinement before approval.
2.2.7.2 Time-Effective Implementation
Triangulation can be efficient. Learners are trained to prioritise effort by focusing on:
- Claims that trigger action
- Claims that carry compliance or reputational consequences
- Claims that depend on uncertain assumptions
- Claims that cannot be easily reversed once acted on
This keeps verification rigorous without becoming operationally burdensome.
2.2.8 Interpreting Results and Taking Action
2.2.8.1 When Triangulation Aligns
When sources and reasoning paths converge, confidence increases. The output can move toward approval, with any remaining uncertainty documented.
2.2.8.2 When Triangulation Diverges
When results diverge, Learners do not force agreement. They isolate the cause and respond with:
- Additional verification against sources
- Clarification of constraints and objectives
- More specific task framing
- Escalation to domain expertise when required
Divergence is treated as useful information. It signals where the reasoning is fragile and where human judgment must lead.
2.2.9 Role of Triangulation in Reasoning Control
Triangulation is a core mechanism of reasoning control in Cyrenza. It operationalises professional scepticism without slowing productivity. It ensures that AI outputs remain inputs to human judgment rather than substitutes for verification. When applied consistently, triangulation increases reliability, strengthens defensibility, and prevents silent failure from entering decisions and deliverables.