CLEAR Agent Reference Pack

Supporting knowledge for the CLEAR Design Partner & Reviewer agents

Mark Houghton, i-performance

Version 1.0 — 18 October 2026

Front matter

Document: CLEAR Agent Reference Pack
Version: 1.0
Date: 18 October 2026
Owner: Mark Houghton, i-performance
Used by: CLEAR Design Partner & Reviewer (Microsoft 365 Copilot agent) and CLEAR Design Partner & Reviewer (ChatGPT custom GPT)

Purpose of this pack

This pack gives the CLEAR Design Partner & Reviewer agents the supporting material they need to work in both of their modes:

·      Review mode — critiquing existing audit support and training materials against the CLEAR Thinking Design and Implementation Guide v1.0.

·      Partner mode — helping to design and write new audit support and training materials that evoke the intended thinking.

It brings together four pieces of project material that the Guide does not itself contain: the project design rules (the method every design element should follow), a calibration case showing the expected depth of analysis, an independent red-team critique of the CLEAR design system, and the worked pattern library.

Status and precedence

The CLEAR Thinking Design and Implementation Guide v1.0 (12 Aug 2026) is the controlled source and takes precedence. This pack is supporting material: project design rules, a calibration example, an independent red-team critique, and worked patterns. Where terms or definitions here differ from the Guide, follow the Guide.

Note on dating and terminology: The worked patterns (Section D) and the red-team review (Section C) were produced in July 2026, before the Guide v1.0 was approved. They refer to earlier v0.1 project documents (for example the Conceptual Canon, Cognitive Architecture, Language and Design Grammar and Application Standards) and some terminology may differ from the Guide. For example, they use “Record and Reflect” for the fifth CLEAR move. Use them for their method, reasoning and examples, not as a source of controlled definitions.

How the agents should use this pack

·      Section A — apply these design rules when reviewing or drafting any design element.

·      Section B — use as the benchmark for the depth of analysis expected in a review.

·      Section C — use to anticipate the ways CLEAR materials can fail in practice (checklist drift, boilerplate, excessive documentation, performative challenge) and to challenge designs on those grounds.

·      Section D — use as worked examples of weak and redesigned materials, support levels and response quality when drafting or critiquing.

Contents

·      Section A — Project design rules

·      Section B — Calibration case: Setting the Tone meeting pseudo-purpose analysis

·      Section C — Independent red-team review of the CLEAR design system

·      Section D — Worked Pattern Library (revised following red-team review)

Section A — Project design rules

Status: agreed project instructions for the CLEAR Design and Implementation Guide project (reproduced verbatim). These are the design rules the agents should apply to every proposed design element.

This project will develop a practical design and implementation guide for embedding CLEAR thinking within an audit firm.

The guide is for people who design audit methodology, templates, working papers, digital tools, AI interfaces, learning, coaching materials, review processes, communications and other elements of the auditor experience.

Treat CLEAR as the controlled visible framework for critical thinking, challenge and reflection. The guide must also embed the four agreed forms of high-quality thinking, metacognition and counterfactual thinking.

The central design principle is that language, structure, sequence, examples, defaults and interaction patterns prime particular forms of thinking. Materials must therefore be designed with sufficient precision to evoke the intended mental activity and avoid procedural, superficial or performative compliance.

For every proposed design element:

1.        State the audit moment and intended user.

2.        Define the exact form of thinking required.

3.        Identify the unhelpful default thinking to overcome.

4.        Explain the cognitive mechanism by which the wording or structure is expected to work.

5.        Identify any assumptions or conclusions embedded in the wording.

6.        Show what attention and mental operation the design primes.

7.        Test whether it can be answered mechanically or with boilerplate.

8.        Provide examples of strong, partial and superficial responses.

9.        Identify observable indicators that the intended thinking occurred.

10.  Recommend how the design should be tested with auditors.

Do not treat metacognition as end-stage reflection. Treat it as the auditor’s ability to notice, articulate, monitor and regulate their thinking throughout the work.

Do not treat counterfactual thinking only as challenging management’s explanation after it has been received. Treat its mature form as generating and holding multiple plausible hypotheses, anticipating the evidence pattern associated with each, seeking discriminating evidence and updating the hypothesis field as evidence is obtained.

Do not change controlled definitions or approved terminology without explicitly identifying the proposed change and its implications.

Distinguish clearly between:

·      controlled source material;

·      agreed project decisions;

·      proposed interpretations;

·      illustrative examples;

·      external evidence;

·      unresolved questions.

Prioritise cognitive precision, audit realism and practical usability over polished but generic prose. Challenge wording that merely names a desired behaviour without evoking the required mental action. Challenge structures that encourage retrospective rationalisation, box-ticking, excessive documentation or boilerplate.

The final guide must enable designers to create materials that make high-quality audit thinking more likely, visible, discussable, coachable and reviewable in the flow of work.

Section B — Calibration case: Setting the Tone meeting pseudo-purpose analysis

Status: illustrative calibration example (reproduced verbatim).

This is a gold-standard example of the depth of analysis expected when reviewing a design element. It shows the four moves a good review makes: it diagnoses pseudo-purpose (stated aims that describe activities rather than changes in thinking), explains why each falls short, proposes a genuine purpose, and defines observable evidence that the purpose has been achieved.

The original request was: “please review this doc for pseudo purpose, and recommend an overall purpose for the setting the tone meeting”, with the Setting the Tone meeting agenda template attached. The analysis was provisional, because the template itself was not available to the reviewer at the time; it is therefore an assessment of likely pseudo-purposes rather than a line-by-line review.

Review: Purpose of the Setting the Tone meeting

I would distinguish between the stated purpose of the meeting and its underlying purpose.

The risk with a Setting the Tone meeting is that it becomes an exercise in communicating expectations, reminding the team of responsibilities and demonstrating that the required discussion has taken place.

Those activities may be useful, but they are not necessarily evidence that the meeting has achieved anything meaningful.

From a CLEAR perspective, the more important question is:

What should be different about the team’s shared understanding, thinking and readiness to act as a result of the meeting?

My starting proposition is that the meeting should establish a shared mental model of the audit, not simply a shared awareness of the audit plan.

That means helping team members connect the purpose of the audit, the reality of the business, the significant uncertainties and risks, and their individual responsibilities.

It should also establish the conditions for people to question assumptions, express uncertainty, challenge constructively and adapt their approach as the audit progresses.

Where pseudo-purpose may be appearing

I haven’t been able to locate the actual Setting the Tone template in the available documents, so the following is a provisional assessment rather than a line-by-line review.

Potential pseudo-purpose

Why it falls short

Communicate expectations for the audit

Information can be transmitted without changing understanding or behaviour

Reinforce the importance of audit quality

People can agree without knowing what to do differently

Promote professional scepticism

Names a desired behaviour without activating critical thinking

Introduce CLEAR

Risks making knowledge of the framework the objective

Encourage open communication and challenge

Does not establish whether people feel able to challenge or know when to do so

Confirm roles and responsibilities

Can establish task ownership without shared understanding of the work’s purpose

These are potentially useful means, but none is sufficient as the overall purpose.

Recommended overall purpose

Proposed purpose statement: To establish a shared understanding of what matters for this audit, why it matters and how the team will work together to deliver a high-quality audit — creating the clarity, ownership and openness needed to exercise sound judgement, challenge constructively and respond effectively as the audit develops.

I favour this formulation because it brings together three essential outcomes:

Shared understanding: Team members understand the audit’s purpose, the business reality, the significant risks and uncertainties, and how their work contributes.

Shared ownership: People understand their responsibilities and feel accountable for the quality of their thinking and judgement, not merely the completion of assigned procedures.

Shared ways of working: The team establishes practical expectations for questioning, challenge, seeking support, communicating concerns and adapting when circumstances change.

How would we know the meeting had achieved its purpose?

I would look for observable changes rather than evidence that topics were discussed.

For example, team members should be able to explain the important risks in their own words, connect those risks to the business, identify areas of uncertainty, understand how their work contributes, and describe what they would do if evidence contradicted the team’s current understanding.

Equally importantly, we should see the partner or manager actively invite different perspectives, respond constructively to uncertainty and demonstrate that challenge is expected rather than merely permitted.

One important design distinction

I would avoid making the purpose of this meeting to complete CLEAR thinking across the audit.

At this early stage, the team will not possess the evidence or detailed understanding needed for many later judgements. Trying to incorporate every CLEAR element explicitly could create superficial compliance.

Instead, the meeting should establish the shared understanding and team conditions that enable connected CLEAR thinking throughout the engagement.

That is a more realistic and valuable outcome.

My recommendation would be to use the proposed purpose statement as the design anchor, then review every existing section of the template against one question:

How does this part of the meeting contribute to the intended change in the team’s understanding, ownership or ways of working?

Section C — Independent red-team review of the CLEAR design system

Status: independent critique of the v0.1 design system (July 2026), pre-dating the Guide v1.0. Reproduced in full; only formatting has been cleaned.

The review was commissioned as an independent red-team of the CLEAR Design and Implementation Guide: it was asked to test (not improve) the integrity and likely practical effects of the Conceptual Canon, Cognitive Architecture, Language and Design Grammar, Application Standards and Pattern Library against 18 criteria (including checklist risk, boilerplate, retrospective rationalisation, anchoring on management’s explanation, cognitive overload, experience levels, time pressure and observability), classifying each finding as Critical, Material, Moderate or Minor.

Independent Red-Team Review — CLEAR Design System v0.1

Overall judgement

The system is conceptually strong but operationally unstable.

Its central architecture is coherent: CLEAR is positioned as a visible framework; the four forms describe cross-cutting qualities of thinking; metacognition regulates thinking throughout the work; and mature counterfactual thinking generates and tests competing hypotheses rather than merely challenging management retrospectively. The documents also consistently reject fixed sequencing, indiscriminate documentation and hypothesis-counting. CLEAR Design Review CLEAR Cognitive Architecture

The principal threat is not conceptual incoherence. It is implementation inversion:

A system designed to overcome procedural compliance could itself become an exceptionally elaborate compliance system.

The Application Standards contain 20 universal standards, numerous mandatory requirements, visibility levels, design tests, sequences and application-specific controls. A judgemental working-paper architecture contains up to 15 components, while task briefings may contain 14. Application Standards Overview The Pattern Library then supplies detailed, polished exemplars across recurring audit situations. CLEAR Language Pattern Library

Without a much stronger proportionality and deployment control, firms are likely to convert this richness into:

·      mandatory fields;

·      standard prompt sets;

·      long task cards;

·      required hypothesis tables;

·      visible “updating” records;

·      standard reviewer questions;

·      AI-assisted completion;

·      measures based on the presence of these artefacts.

That would create more visible CLEAR activity without reliable evidence of improved judgement.

Severity summary

Severity

Findings

Critical

3

Material

9

Moderate

7

Minor

2

Total

21

Critical findings

C1. The design system is structurally capable of becoming the checklist it rejects

Affected control documents: Cognitive Architecture; Language and Design Grammar; Application Standards; Pattern Library

Exact weakness

The architecture repeatedly states that CLEAR is not a rigid checklist and that only material reasoning should be externalised. CLEAR Cognitive Architecture Yet downstream documents provide a very large number of named requirements, standard sequences, mandatory design tests, field orders and expected outputs.

Examples include:

·      20 Universal Application Standards;

·      14 required briefing components;

·      a 15-part judgemental working-paper architecture;

·      repeated use of “must”, “every application” and “mandatory standards”;

·      a standard 14-part format for every pattern. Application Standards Overview CLEAR Language Pattern Library

This is not a prose problem. It is a structural contradiction between the intended flexible cognitive architecture and the control density of the application system.

Likely consequence

Implementation teams will translate the standards into auditable completeness requirements. Auditors will learn that “good CLEAR” means populating the prescribed components. Reviewers will request missing fields. Local teams will optimise for visible compliance because it is easier to monitor than cognition.

CLEAR will become an additional methodology layer rather than a way of improving the thinking already required by methodology.

Required correction

Before further development, establish a controlled deployment architecture that distinguishes:

1.        principles applying to designers;

2.        optional design patterns;

3.        minimum requirements for selected high-risk moments;

4.        material that must never become an auditor-facing requirement;

5.        explicit prohibited implementations.

No auditor-facing application should inherit the complete design-assurance framework.

Set a hard rule that CLEAR may not be introduced as five mandatory fields or as a universal task sequence. The Application Standards should govern designers, not automatically become content presented to auditors.

Nature of correction: Application. The underlying concept should not change.

C2. Proportionality is asserted but not operationally decidable

Affected control documents: Cognitive Architecture; Language and Design Grammar; Application Standards

Exact weakness

The documents repeatedly use qualifications such as:

·      “where relevant”;

·      “where material uncertainty exists”;

·      “for material judgemental work”;

·      “where hindsight risk is material”;

·      “proportionate”;

·      “selectively”.

The architecture acknowledges unresolved questions about when hypothesis mapping is proportionate, how much should be documented, and how the system should align with audit standards. CLEAR Cognitive Architecture

However, downstream standards have already been written as if those unresolved thresholds can be applied consistently. There is no reliable decision rule for determining:

·      when multiple hypotheses are required;

·      when expected evidence should be captured before testing;

·      when updating needs a visible history;

·      when an uncertainty should be discussed versus documented;

·      when a compact prompt becomes a formal hypothesis map;

·      when the costs exceed the likely cognitive benefit.

Likely consequence

Teams will resolve ambiguity conservatively by doing more. Regulators and internal reviewers may also infer that failure to use the richer structure indicates weak thinking. This creates documentation growth and defensive application precisely in the more difficult audits where time and cognitive capacity are constrained.

Required correction

Create a controlled proportionality gate before any application standard is approved. It should classify audit moments into a small number of support levels based on:

·      judgement significance;

·      ambiguity of underlying reality;

·      susceptibility to management or prior-year anchoring;

·      availability of discriminating evidence;

·      novelty to the auditor;

·      consequence of missing the concerning plausible explanation;

·      documentation already required by methodology.

Each level should define the maximum—not just minimum—permitted scaffolding.

Nature of correction: Application, but essential to preserve the concept.

C3. The system does not yet have a valid way to distinguish genuine capability from artefact production

Affected control documents: Conceptual Canon; Cognitive Architecture; Application Standards; Pattern Library

Exact weakness

The documents correctly state that observable indicators are only traces and that CLEAR headings do not prove cognition. The Application Standards propose categories such as attention, mental model, action, work product, updating and escalation. Application Standards Overview

But no validated measurement logic distinguishes:

·      genuine independent thinking;

·      thinking elicited only by the scaffold;

·      compliance with a familiar response pattern;

·      reviewer- or AI-supplied reasoning;

·      improved documentation without improved judgement;

·      temporary task performance from transferable capability.

The Pattern Library suggests pilots measuring hypothesis quality, evidence selection and changes to work, but no common scoring model or blinded evaluation protocol exists. CLEAR Language Pattern Library

Likely consequence

Implementation success will be claimed from:

·      completion rates;

·      presence of alternatives;

·      more detailed files;

·      positive user feedback;

·      reviewers reporting better narratives;

·      examples of work changing.

These indicators are vulnerable to gaming and do not establish that auditors can perform the thinking independently in unfamiliar situations.

Required correction

Do not approve organisation-wide measures until a controlled evaluation model exists. It should distinguish:

1.        prompted performance — quality while using the scaffold;

2.        independent performance — quality without the scaffold;

3.        transfer — performance in a changed audit context;

4.        retention — performance after time has passed;

5.        judgement effect — whether conclusions and evidence choices improve;

6.        burden and displacement — what useful work is lost to the intervention.

Use blinded scoring of realistic audit decisions, not visible CLEAR language.

Nature of correction: Application and measurement. The concept remains intact.

Material findings

M1. The controlled status of core concepts remains ambiguous

Affected control documents: Conceptual Canon; Cognitive Architecture; Application Standards

Exact weakness

The Canon calls several definitions “proposed canonical definitions” and explicitly retains unresolved matters. CLEAR Design Review The Cognitive Architecture nevertheless describes the four forms as “controlled”, and the Application Standards say they do not change controlled definitions.

This creates circular authority: downstream standards treat as controlled concepts that the Canon still describes as proposed.

Likely consequence

Later changes may be treated either as prohibited redefinition or as legitimate refinement depending on the document being cited. This weakens change control and makes contradiction difficult to diagnose.

Required correction

Formally approve, reject or retain as provisional each concept before the design system is labelled a controlled standard. Maintain a one-page authority register showing:

·      exact controlled term;

·      exact definition;

·      approval status;

·      controlling document;

·      permitted explanatory variants.

Nature of correction: Conceptual governance, not a substantive concept change.

M2. “Disconnected Thinking” is too broad to function as a discriminating causal diagnosis

Affected control documents: Conceptual Canon; all downstream documents

Exact weakness

The Canon defines disconnected thinking as loss of connection to purpose, business reality, evidence implications or conclusion. CLEAR Design Review This captures most forms of poor audit reasoning.

As a result, the concept risks becoming unfalsifiable: virtually any quality failure can be redescribed as a disconnection.

Likely consequence

Implementation teams may overstate CLEAR’s causal reach and position it as the upstream answer to unrelated problems such as:

·      insufficient technical knowledge;

·      poor resourcing;

·      inadequate supervision;

·      flawed data;

·      weak methodology;

·      independence issues;

·      commercial pressure;

·      deliberate misconduct.

This may reduce the precision of diagnosis and lead to cognitive interventions where structural or technical correction is required.

Required correction

Retain disconnected thinking as an umbrella problem, but require every application to identify the specific break being addressed and distinguish it from:

·      knowledge deficit;

·      capacity deficit;

·      authority constraint;

·      methodology defect;

·      data-access defect;

·      incentive or governance problem.

Nature of correction: Application qualification; the central concept need not be abandoned.

M3. Counterfactual thinking is becoming a near-universal architecture rather than a selectively useful discipline

Affected control documents: Cognitive Architecture; Language and Design Grammar; Application Standards; Pattern Library

Exact weakness

Multiple-hypothesis generation and discriminating evidence are strongly and repeatedly embedded across Clarify, Look Beyond, Evidence, Ask, Record and Reflect, task allocation, templates, coaching, review and training. CLEAR Cognitive Architecture Application Standards Overview

The documents say the approach should be bounded and proportionate, but its repeated presence makes it appear to be the normative structure of almost all audit thinking.

Some audit tasks are better characterised as:

·      verifying a specific factual condition;

·      applying a clear technical criterion;

·      testing an operational process;

·      executing a prescribed response to a known risk;

·      identifying whether a defined exception occurred.

These do not always benefit from a formal hypothesis field.

Likely consequence

Auditors may manufacture alternatives around settled or tightly bounded questions, adding time and pseudo-scepticism without improving assurance.

Required correction

Define the entry conditions for counterfactual architecture. It should be activated particularly where:

·      the same observed condition can plausibly arise from materially different realities;

·      management judgement or explanation is influential;

·      evidence patterns can genuinely discriminate;

·      the consequence of premature closure is significant.

Elsewhere, connected and purpose-led thinking may be sufficient.

Nature of correction: Application; no underlying conceptual change.

M4. The “most concerning plausible explanation” may systematically distort evidence search

Affected control documents: Conceptual Canon; Cognitive Architecture; Pattern Library

Exact weakness

The system repeatedly pairs “most likely” with “most concerning plausible”. This is useful for risk focus, but no equal control is provided against:

·      possibility inflation;

·      severity bias;

·      fraud salience;

·      over-weighting high-consequence but weakly plausible explanations;

·      continuing to test a concerning hypothesis after evidence has materially weakened it.

Likely consequence

Teams may over-test dramatic explanations, treat scepticism as asymmetric suspicion, and consume time on low-probability concerns. This could also make conversations with management more adversarial without improving diagnostic accuracy.

Required correction

Require explicit assessment of both:

·      plausibility on current facts; and

·      consequence if missed.

A concerning explanation should remain live only where both justify audit attention. Include a stopping rule for de-prioritising it.

Nature of correction: Application qualification.

M5. The system under-specifies how management’s explanation should be used as evidence

Affected control documents: Cognitive Architecture; Language and Design Grammar; Pattern Library

Exact weakness

The documents appropriately resist allowing management’s explanation to become the organising anchor. However, there is a risk of overcorrection: management’s account may be treated mainly as a cognitive contaminant rather than as potentially important evidence about:

·      processes;

·      intentions;

·      events;

·      controls;

·      assumptions;

·      internal inconsistencies;

·      the origin of accounting treatments.

The Pattern Library frequently places management’s explanation after independent hypotheses. CLEAR Language Pattern Library

Likely consequence

Auditors may artificially delay useful inquiry, create alternatives without sufficient entity knowledge, or treat independent generation as inherently superior even where management’s information is necessary to define the factual field.

Required correction

Distinguish:

1.        obtaining management’s factual account;

2.        accepting management’s causal interpretation;

3.        adopting management’s conclusion.

The control should be against unexamined adoption, not against hearing management early. Independent framing may occur before or after management input depending on the audit moment, but management’s explanation must remain one evidential input rather than the default conclusion.

Nature of correction: Application; underlying anti-anchoring concept remains.

M6. Grounding can become plausible storytelling rather than audit evidence

Affected control documents: Conceptual Canon; Cognitive Architecture; Pattern Library

Exact weakness

The system explicitly states that a plausible story is not grounded until expected traces are tested. CLEAR Cognitive Architecture Nevertheless, the repeated use of mental simulation, narrative, “possible realities” and business stories creates a substantial risk that vividness is mistaken for evidential strength.

A confident auditor can construct a highly coherent but speculative causal account.

Likely consequence

Narrative coherence may displace:

·      source reliability;

·      completeness concerns;

·      audit sampling;

·      control testing;

·      technical requirements;

·      contradictory data;

·      uncertainty about unobserved events.

AI will amplify this risk by creating realistic business narratives from sparse facts.

Required correction

In every grounding application, separate visibly:

·      observed event or condition;

·      inferred causal connection;

·      predicted trace;

·      actual evidence;

·      unresolved inference.

A grounding response must fail if it is vivid but unsupported.

Nature of correction: Application.

M7. “Narratable thinking” remains vulnerable to becoming a writing standard

Affected control documents: Conceptual Canon; Cognitive Architecture; Language and Design Grammar; Application Standards

Exact weakness

The concept is carefully distinguished from polished writing. CLEAR Design Review However, its observable expression is necessarily a reconstructed explanation. Reviewers and regulators predominantly see the record, not the cognition.

The application system repeatedly asks for reasoning spines, narratives, decisive evidence, updating and residual uncertainty.

Likely consequence

Auditors who write well will appear stronger than auditors who think well but express themselves less fluently. AI may narrow the visible difference further. Reviewers may reward narrative completeness and penalise concise but sufficient reasoning.

Required correction

Require narratability to be tested through multiple channels:

·      oral reconstruction;

·      response to perturbation;

·      evidence-selection behaviour;

·      ability to adapt the plan;

·      concise documentation.

Do not infer capability solely from prose.

Nature of correction: Application and measurement.

M8. The workload implied by the exemplar designs is incompatible with ordinary engagement time pressure

Affected control documents: Application Standards; Pattern Library

Exact weakness

The exemplar patterns are cognitively rich but often require:

·      factual reconstruction;

·      multiple causal hypotheses;

·      expected evidence for each;

·      evidence-function assessment;

·      updating;

·      related-work implications;

·      metacognitive monitoring;

·      a reasoned conclusion;

·      reviewer and coaching diagnostics.

For significant matters, this may be valuable. As a normal expectation, it is too demanding.

Likely consequence

Under pressure, teams will:

·      complete the structures retrospectively;

·      use standard alternatives;

·      copy exemplar logic;

·      ask AI to draft the analysis;

·      reduce time spent obtaining evidence;

·      regard CLEAR as an administrative burden.

Required correction

Test all applications under realistic budgets and deadlines. Measure not only quality gains but time displacement. Develop “micro-CLEAR” patterns that alter one high-risk cognitive move without requiring a complete reasoning architecture.

Nature of correction: Application.

M9. The novice model may ask inexperienced auditors to generate hypotheses they lack the schemas to produce

Affected control documents: Cognitive Architecture; Application Standards; Pattern Library

Exact weakness

The architecture recognises that novices need concrete representations, bounded categories and examples. CLEAR Cognitive Architecture Yet several patterns expect associates or seniors to identify plausible realities, evidential differences and wider implications.

Without adequate business, accounting and audit knowledge, novices may generate:

·      generic possibilities;

·      implausible causes;

·      memorised fraud labels;

·      alternatives derived from exemplars;

·      false confidence.

Likely consequence

The scaffold may disguise missing expertise rather than develop it. Managers may interpret weak hypothesis generation as weak thinking when the true gap is knowledge.

Required correction

For novices, separate:

1.        construction of the business model;

2.        teaching of common causal structures;

3.        guided comparison;

4.        independent generation.

Do not use free hypothesis generation as a proxy for capability before the required schemas exist.

Nature of correction: Application.

M10. Expert use is underdeveloped relative to novice and developing-practitioner scaffolding

Affected control documents: Cognitive Architecture; Application Standards; Pattern Library

Exact weakness

The architecture says experts usually need sparse, high-value interruption. However, most detailed patterns and structures are heavily scaffolded. Expert-specific applications are largely described as general prompts such as weakest assumption, strongest credible alternative and perturbation.

Likely consequence

Senior users may reject CLEAR as remedial, verbose or junior-focused. Alternatively, firms may impose the same structures on experts, reducing efficiency and provoking superficial compliance.

Required correction

Develop distinct expert patterns based on:

·      independent pre-mortem challenge;

·      anomaly interrogation;

·      cross-engagement pattern disruption;

·      evidence-source dependency;

·      decision-threshold testing;

·      compact partner challenge.

Nature of correction: Application.

M11. Compatibility with audit methodology is stated as an aspiration, not demonstrated

Affected control documents: Cognitive Architecture; Language and Design Grammar; Application Standards

Exact weakness

The documents say CLEAR is not a substitute for methodology, but they do not map CLEAR to:

·      risk identification and assessment;

·      assertion-level procedures;

·      sufficiency and appropriateness of evidence;

·      accounting-estimate requirements;

·      fraud requirements;

·      documentation standards;

·      supervision and review;

·      consultation;

·      completion.

The architecture itself lists alignment with formal audit standards as unresolved. CLEAR Cognitive Architecture

Likely consequence

CLEAR may duplicate methodology, conflict with existing terminology or create parallel requirements. Auditors may be uncertain whether a CLEAR conclusion is part of, supplementary to or separate from required audit documentation.

Required correction

Complete a requirement-by-requirement methodology crosswalk before implementation. Identify:

·      where CLEAR improves an existing requirement;

·      where it merely explains the cognition behind it;

·      where no new documentation is needed;

·      where a current method or template must be removed to avoid duplication.

Nature of correction: Application and governance.

M12. Regulatory compatibility is assumed but not tested

Affected control documents: Application Standards; Project implementation logic

Exact weakness

The system aims to make reasoning defensible and regulator-visible, but “minimum sufficient externalisation” may conflict with conservative interpretations of documentation requirements. Conversely, visible hypothesis histories may invite retrospective regulatory scrutiny of discarded possibilities.

Likely consequence

Firms may over-document to protect against challenge, while regulators may interpret recorded but rejected hypotheses as evidence that additional work was required.

Required correction

Test prototypes with technical, legal and regulatory stakeholders using concrete files. Determine:

·      what must be retained;

·      what may remain ephemeral;

·      how discarded hypotheses are treated;

·      when coaching records should remain outside the audit file;

·      how AI contribution is attributed.

Nature of correction: Application.

Moderate findings

Mo1. The boundary between Adaptive thinking and metacognitive regulation is not stable in use

Affected control documents: Conceptual Canon; Cognitive Architecture

Exact weakness

The Canon states: “Metacognition detects and regulates; Adaptive thinking changes the response.” This is conceptually defensible, but in practice regulation often is the change: redirecting attention, reframing or seeking help.

Likely consequence

Designers may produce separate prompts for noticing and adapting, multiplying fields without adding cognitive value.

Required correction

Treat the distinction as analytic rather than something that must be separately represented in applications.

Nature of correction: Application clarification.

Mo2. “Ask” overlaps with all other CLEAR elements and lacks a unique stable boundary

Affected control documents: Conceptual Canon; Cognitive Architecture

Exact weakness

Ask includes self-questioning, management inquiry, coaching, challenge, specialist involvement and help-seeking. These functions also occur within Clarify, Look Beyond and Evidence.

Likely consequence

Applications may add an “Ask” stage after the real questions have already occurred, creating token challenge.

Required correction

Define Ask primarily as the social and dialogic mechanism used to improve or test the mental model, while making clear that questions can activate any CLEAR function. Do not require a separate Ask artefact.

Nature of correction: Minor conceptual clarification with application implications.

Mo3. “Record and Reflect” remains visually end-loaded despite the continuous-metacognition explanation

Affected control documents: Controlled CLEAR framework; Cognitive Architecture

Exact weakness

The architecture explains two meanings of reflection, but users will still see “Reflect” under the final R in the mnemonic. CLEAR Cognitive Architecture

Likely consequence

Most auditors will intuitively interpret reflection as a conclusion-stage activity. Training may not overcome the affordance of the acronym.

Required correction

Retain the controlled label, but make continuous metacognitive monitoring visible outside the five-letter sequence in every implementation. Do not rely on explanatory text alone.

Nature of correction: Application; no terminology change required.

Mo4. The standard may encourage alternatives to be generated before sufficient factual clarification

Affected control documents: Cognitive Architecture; Language and Design Grammar; Pattern Library

Exact weakness

The system strongly prioritises alternatives before commitment, but hypothesis quality depends on an adequate factual and process model.

Likely consequence

Teams may brainstorm too early, producing generic or speculative explanations that anchor later work.

Required correction

Require a minimum factual base before formal hypothesis generation: observed condition, entity context, relevant process and known constraints. “Before management’s explanation” should not become “before understanding the facts”.

Nature of correction: Application.

Mo5. Expected-evidence recording may itself create anchoring

Affected control documents: Language and Design Grammar; Application Standards

Exact weakness

Capturing expected evidence before testing reduces hindsight, but it can make auditors search narrowly for predicted traces and discount unexpected but important evidence.

Likely consequence

Prediction becomes a new confirmation frame.

Required correction

Include an explicit “unexpected observations/new hypothesis” route. Reviewers should test whether evidence outside the predicted pattern was noticed.

Nature of correction: Application.

Mo6. Strong exemplar responses are too complete and professionally polished

Affected control document: Pattern Library

Exact weakness

The strong responses combine facts, causal modelling, evidence evaluation, updating and precise audit consequences in highly polished prose. CLEAR Language Pattern Library

Likely consequence

They will function as model answers even though the library warns against copying. AI and template designers will reproduce their structure and language.

Required correction

Use more process-visible exemplars:

·      brief contemporaneous notes;

·      incomplete but improving thinking;

·      oral dialogue;

·      evidence decisions;

·      alternative valid strong responses.

Do not make the “strong” response synonymous with a perfect final paragraph.

Nature of correction: Application.

Mo7. The Pattern Library’s coverage is too small to establish general audit realism

Affected control document: Pattern Library

Exact weakness

The library covers nine patterns concentrated in familiar high-judgement areas. It does not yet test the system in:

·      routine controls work;

·      IT audit;

·      group audit;

·      estimates involving specialists;

·      sampling;

·      completeness testing;

·      service organisations;

·      regulated entities;

·      small entities;

·      data analytics;

·      component teams;

·      late evidence and clearance pressure.

Likely consequence

The system may be overfitted to narrative analytical problems and less effective in audit moments governed by different cognitive demands.

Required correction

Expand testing before claiming general application, especially into areas where the relevant mental operation is not primarily causal explanation or multiple hypotheses.

Nature of correction: Application and validation.

Minor findings

Mi1. “Counterfactual thinking” may be an imperfect label for the controlled construct

Affected control documents: Conceptual Canon; Cognitive Architecture

Exact weakness

The defined construct includes abductive hypothesis generation, diagnostic reasoning, evidence prediction and Bayesian-style updating. “Counterfactual” commonly suggests imagining alternatives to an established fact or past event.

Likely consequence

External readers may misunderstand the concept as “what else could have happened?” rather than parallel diagnostic hypothesis testing.

Required correction

Retain the agreed term unless formally changed, but consistently include the controlled definition or an explanatory descriptor such as “parallel diagnostic hypothesis thinking”.

Nature of correction: Terminology explanation. A formal rename would be a conceptual change.

Mi2. Some controlled claims exceed the current evidence base

Affected control documents: Conceptual Canon; implementation materials

Exact weakness

The materials make broad claims that disconnected thinking is an upstream leverage point and that the design mechanisms will improve quality. These are plausible project hypotheses, not yet empirically demonstrated across applications.

Likely consequence

The implementation narrative may sound more certain than the evidence permits.

Required correction

Mark causal claims as hypotheses to be tested until pilot evidence exists.

Nature of correction: Evidential status, not concept.

Cross-document contradictions

1. Flexible framework versus mandatory control system

The Cognitive Architecture describes flexible recurring moves and says micro-iteration usually requires no separate documentation. CLEAR Cognitive Architecture The Application Standards then establish 20 universal requirements and extensive mandatory structures. Application Standards Overview

Assessment: Critical operational contradiction.

2. Proposed definitions versus controlled definitions

The Canon labels several formulations “proposed canonical definitions”. CLEAR Design Review Downstream documents treat them as controlled.

Assessment: Material governance contradiction.

3. Minimum sufficient externalisation versus detailed preferred architectures

All documents endorse minimum sufficient externalisation, but the detailed task, working-paper and pattern structures make extensive externalisation the path of least resistance.

Assessment: Critical practical contradiction.

4. Examples are not model answers versus strong responses that are highly reusable

The Pattern Library explicitly says examples should not be copied, while supplying polished, realistic paragraphs that can easily be reused. CLEAR Language Pattern Library

Assessment: Moderate design contradiction.

5. Stable cognitive standard versus practical limits of novice schemas

The architecture says the quality standard remains stable while scaffolding varies. In practice, some expert-level operations cannot be performed reliably without developed entity, technical and causal schemas.

Assessment: Material developmental tension. The aspiration is valid, but expected independent performance cannot be identical.

6. Management explanation as one hypothesis versus fixed sequence after independent alternatives

The conceptual position is that management’s account should not dominate. Some patterns operationalise this as a sequence in which alternatives precede management’s explanation.

Assessment: Moderate over-specification. The principle should control anchoring, not prescribe universal information order.

Assessment against the 18 requested dimensions

Dimension

Assessment

1. Conceptual consistency

Strong internally, with authority-status ambiguity and some blurred construct boundaries.

2. Audit realism

Good in high-judgement examples, insufficiently tested in routine, technical and distributed audit work.

3. Cognitive precision

Very strong, but richness exceeds what can be practically activated at many workflow moments.

4. Additional-checklist risk

Critical. The principal implementation threat.

5. Excessive-documentation risk

Critical, despite explicit warnings against it.

6. Boilerplate risk

High, particularly once prompts, exemplars and AI become widely available.

7. Retrospective-rationalisation risk

High unless initial views and evidence effects are captured selectively and contemporaneously.

8. Alternatives generated after commitment

Well recognised conceptually; operational controls remain vulnerable to retrospective completion.

9. Management explanation remains anchor

Substantially addressed, with some risk of overcorrecting by delaying useful management input.

10. Performative challenge

Well diagnosed, but likely to reappear through required alternatives, review questions and challenge records.

11. Cognitive overload

High in the full system; insufficient maximum-scaffolding limits.

12. Different experience levels

Conceptually addressed, but novice and expert patterns require more substantial differentiation.

13. Time pressure

Not yet compatible at scale without micro-patterns and strict proportionality gates.

14. Methodology and regulation

Unresolved; crosswalk and regulatory testing are prerequisites.

15. Coaching and development

Strong potential; coaching architecture is one of the strongest parts of the system.

16. Intended thinking observable

Partly. Observable traces are identified, but remain ambiguous and gameable.

17. Measures distinguish adoption/capability

No, not yet. This is a critical gap.

18. Contradictions between documents

Several, primarily flexibility versus control density, and minimum externalisation versus elaborate structures.

Required corrections before the system advances

The following are release conditions rather than stylistic improvements:

1.        Resolve the authority and status of all controlled definitions.

2.        Create a deployment architecture separating designer controls from auditor-facing content.

3.        Create a proportionality gate with maximum permitted scaffolding.

4.        Map CLEAR against audit methodology and documentation requirements.

5.        Define when multiple-hypothesis reasoning is and is not required.

6.        Develop micro-patterns suitable for real time pressure.

7.        Build distinct novice, developing-practitioner and expert applications.

8.        Replace polished model-answer dependence with process-visible exemplars.

9.        Establish a measurement protocol that tests independent performance, transfer and judgement quality.

10.  Run pilots designed to detect harm as well as benefit, including documentation growth, time displacement, AI substitution, over-testing and reviewer gaming.

Final red-team conclusion

The conceptual core should be preserved. The system’s most valuable contributions are:

·      grounding numbers in real business activity;

·      distinguishing procedure completion from assurance;

·      generating alternatives before premature commitment;

·      seeking discriminating rather than merely compatible evidence;

·      treating metacognition as continuous regulation;

·      making evidence-led updating visible;

·      using coaching to improve the auditor’s model rather than replace it.

The current design system is not yet safe for broad implementation.

Its dominant failure mode would be a highly sophisticated form of performative compliance: auditors completing structures about purpose, alternatives, evidence, updating and reflection without those structures materially changing how they understood the entity, selected evidence or formed the judgement.

The next control objective should therefore be:

Reduce the visible machinery while preserving the cognitive mechanism.

Section D — Worked Pattern Library (revised following red-team review)

Status: proposed pattern library, version 0.1 revised (control date 16 July 2026), pre-dating the Guide v1.0. This is the final version of the library, restructured in response to the red-team review in Section C. It is reproduced in full; only formatting has been cleaned. It contains 12 patterns together with the prohibited uses, diagnostic gate, proportionality gate and support levels 0–3, counterfactual-thinking gate, micro-patterns, anti-patterns and evaluation protocol.

The revision below restructures the Pattern Library around deployment control and proportionality rather than simply amending individual examples. It retains the worked audit situations but prevents the library itself becoming a source of mandatory fields, model answers or universal hypothesis structures. The principal changes respond directly to the red team’s finding that the design system was conceptually strong but operationally vulnerable to implementation inversion.

CLEAR Pattern Library v0.1

Revised following independent red-team review

Project: CLEAR Design and Implementation Guide
Workstream: Pattern Library
Status: Proposed pattern library for approval and testing
Version: 0.1 revised
Control date: 16 July 2026

Controlling project materials

·      Project Control Record v0.1

·      CLEAR Conceptual Canon v0.1 revised

·      CLEAR Cognitive Architecture v0.1 revised

·      CLEAR Language and Design Grammar v0.1 revised

·      CLEAR Application Standards v0.1 revised

·      Controlled CLEAR description

·      Source materials on the four forms of thinking

·      Metacognition and counterfactual-thinking source materials

·      Independent Red-Team Review — CLEAR Design System v0.1

1. Purpose

The CLEAR Pattern Library provides designer-facing examples of how language, structure, timing, interaction and defaults can be used to improve particular forms of audit thinking.

It is intended for people designing or reviewing:

·      methodology;

·      task briefings;

·      working papers;

·      audit-platform interactions;

·      AI support;

·      coaching;

·      review;

·      training;

·      communications;

·      implementation pilots.

The library is not:

·      an auditor-facing manual;

·      an audit-programme supplement;

·      a standard set of CLEAR questions;

·      a collection of required working-paper structures;

·      a set of model audit answers;

·      a requirement that each pattern be used in full;

·      evidence that the pattern improves audit quality.

The core design objective is:

Select the least intrusive intervention capable of improving a defined thinking problem at a particular audit moment.

The original library contained cognitively rich, integrated examples. The red-team review found that this richness could be translated into long task cards, hypothesis tables, mandatory fields, standard reviewer questions and AI-assisted completion.

This revision therefore shifts the library from:

complete pattern to be implemented

towards:

design resource from which the smallest useful element may be selected.

2. Evidential and authority status

2.1 Controlled source terms

The following remain controlled source terms:

·      CLEAR;

·      Clarify;

·      Look Beyond;

·      Evidence;

·      Ask;

·      Record and Reflect;

·      Purpose-led;

·      Adaptive;

·      Connected;

·      Narratable;

·      Disconnected Thinking.

The detailed definitions and relationships remain subject to the approval position in the revised Conceptual Canon.

2.2 Pattern status

Every pattern in this library is:

·      an illustrative design proposition;

·      a hypothesis about how a cognitive intervention may work;

·      subject to audit-methodology alignment;

·      subject to technical and regulatory review;

·      subject to realistic auditor testing.

A pattern should not be described as proven to improve audit quality until appropriate evidence exists.

2.3 Worked responses

The responses are not approved audit wording.

They demonstrate:

·      what the user noticed;

·      what changed in their understanding;

·      what evidence decision followed;

·      where uncertainty remained.

They deliberately include:

·      brief working notes;

·      oral responses;

·      incomplete but improving reasoning;

·      alternative valid approaches;

·      concise action decisions.

A polished final paragraph is not treated as the primary evidence of strong thinking.

The red-team review specifically found that the earlier strong exemplars were sufficiently polished to become reusable model answers.

3. Deployment architecture

The following layers must remain separate.

3.1 Designer assurance

This library contains:

·      cognitive mechanisms;

·      unhelpful defaults;

·      response-quality contrasts;

·      anti-boilerplate tests;

·      pilot designs.

These are design-assurance materials.

They should not automatically appear in audit files.

3.2 Optional pattern

A pattern is a possible way of addressing a defined problem.

It may be:

·      used;

·      adapted;

·      reduced;

·      combined with another pattern;

·      rejected.

3.3 Auditor-facing intervention

The auditor should normally see only:

·      one focused prompt;

·      one structural contrast;

·      one coaching question;

·      one review challenge;

·      one short response field;

·      or no additional CLEAR intervention.

3.4 Existing methodology

Where methodology already requires the relevant analysis or documentation, CLEAR should improve that requirement rather than add a parallel form.

3.5 Formal audit record

Only material reasoning required to understand and support the audit judgement should enter the formal record.

The complete pattern-design rationale does not belong in the audit file.

4. Prohibited uses of the library

The library must not be used to create:

·      five mandatory CLEAR fields;

·      universal 14-part audit forms;

·      standard hypothesis tables for every matter;

·      fixed numbers of alternatives;

·      mandatory “most likely” and “most concerning” fields;

·      routine expected-evidence documentation;

·      mandatory metacognitive declarations;

·      standard reviewer scripts applied without diagnosis;

·      automatic insertion of strong responses into templates;

·      AI-generated CLEAR narratives entered directly into the audit file;

·      performance measures based on completed fields or CLEAR language;

·      novice tasks requiring unsupported free hypothesis generation;

·      expert tasks requiring completion of junior scaffolding.

The original 14-part pattern format remains a designer analysis format, not an auditor response format.

5. Diagnostic gate before selecting a pattern

A pattern should not be selected merely because a weakness can be called Disconnected Thinking.

5.1 Identify the specific break

Possible thinking breaks include:

·      purpose–procedure disconnection;

·      number–reality disconnection;

·      evidence–claim disconnection;

·      evidence–source disconnection;

·      finding–implication disconnection;

·      evidence–updating disconnection;

·      workstream–judgement disconnection;

·      reasoning–conclusion disconnection;

·      specialist-work–engagement-judgement disconnection.

5.2 Identify non-cognitive causes

Determine whether the problem is principally:

·      lack of technical knowledge;

·      lack of entity or industry knowledge;

·      lack of time or staffing;

·      lack of authority;

·      defective methodology;

·      unavailable or unreliable data;

·      weak supervision;

·      commercial or hierarchical pressure;

·      governance or incentive failure;

·      ethical or conduct failure.

A prompt should not substitute for:

·      teaching;

·      resources;

·      authority;

·      evidence;

·      specialist support;

·      methodology correction;

·      ethical action.

5.3 Pattern-selection statement

Before approving a pattern, the designer should be able to state:

This pattern is intended for [user] at [audit moment] to address [specific thinking break]. The problem is not primarily a knowledge, capacity, authority, methodology, data or conduct failure because [reason].

6. Proportionality gate

6.1 Support Level 0 — No additional scaffold

Use where:

·      the task is tightly defined;

·      the criterion is clear;

·      explanatory ambiguity is low;

·      methodology already supports the required thinking;

·      another prompt would add little value.

6.2 Support Level 1 — Micro-intervention

Use one focused question or structural change.

This should be the default form of CLEAR implementation.

Examples:

·      What must this procedure establish?

·      What does this evidence not establish?

·      What real event does this number represent?

·      What changed because of the exception?

·      What fact requires escalation?

6.3 Support Level 2 — Guided reasoning support

Use where:

·      ambiguity is meaningful but bounded;

·      the auditor needs support constructing or testing a model;

·      management or prior-year anchoring is influential;

·      evidence sources must be compared;

·      a significant but contained judgement is involved.

Only the relevant CLEAR functions should be supported.

6.4 Support Level 3 — Integrated judgement architecture

Reserve for significant matters involving several of:

·      multiple genuinely plausible underlying realities;

·      linked and sensitive assumptions;

·      significant management judgement;

·      high consequence of premature closure;

·      distributed or contradictory evidence;

·      material residual uncertainty;

·      reporting consequences.

6.5 Maximum-scaffolding rule

Each pattern must state:

·      its intended support level;

·      the smallest effective form;

·      the maximum structure justified.

A pattern should be reduced where:

·      a single question would produce the same benefit;

·      methodology already captures the reasoning;

·      documentation burden exceeds likely benefit;

·      completion would be retrospective under realistic time pressure;

·      it displaces higher-value evidence gathering or supervision.

7. Counterfactual-thinking gate

Counterfactual thinking—used here in its controlled sense of parallel diagnostic hypothesis thinking—is selective rather than universal.

7.1 Use explicit alternatives where

·      the same observed condition could arise from materially different realities;

·      those realities would require different audit responses;

·      management interpretation is influential;

·      premature closure would have significant consequences;

·      evidence could meaningfully distinguish between the explanations.

7.2 Do not require alternatives where

·      a defined fact is being verified;

·      a clear technical criterion is being applied;

·      a specified control is being tested;

·      a known risk has a prescribed response;

·      different explanations would not change work or conclusion.

7.3 Minimum factual base

Before formal hypothesis generation, establish:

·      the observed condition;

·      relevant entity facts;

·      the process or transaction;

·      known technical and commercial constraints.

“Generate alternatives before commitment” must not become “brainstorm before understanding the facts”.

7.4 Plausibility and consequence

A concerning explanation should remain live only where both:

·      its plausibility on current facts; and

·      its consequence if missed

justify further audit attention.

7.5 Stopping rule

Deprioritise or remove a hypothesis where:

·      evidence materially weakens its factual basis;

·      it no longer changes the audit response;

·      further testing would be disproportionate;

·      the remaining risk is already addressed by required work.

8. Management-information rule

Patterns should distinguish:

1. Management’s factual account
   What management says happened, who acted and what process operated.

2. Management’s causal interpretation
   Why management says the condition occurred.

3. Management’s accounting conclusion
   The accounting or reporting implication management adopts.

The control is against unexamined adoption, not against hearing management early.

Management information may be needed to establish the factual field before alternatives can be generated intelligently.

The appropriate information order depends on:

·      the audit moment;

·      available entity knowledge;

·      anchoring risk;

·      user expertise;

·      the nature of the judgement.

9. Grounding control

Grounding must distinguish:

Element

Question

Observed condition

What has actually been observed?

Inferred connection

What causal interpretation is proposed?

Predicted trace

What else should be present or absent?

Actual evidence

What relevant evidence was found?

Unresolved inference

What remains assumed or unobserved?

A response fails the grounding test where it is:

·      vivid;

·      commercially coherent;

·      easy to narrate;

but its material causal connections remain unsupported.

Grounding does not replace:

·      technical analysis;

·      completeness testing;

·      source-reliability assessment;

·      sampling;

·      control testing;

·      contradictory evidence;

·      uncertainty over unobserved events.

The red-team review identified plausible storytelling as a material risk, particularly where AI can manufacture realistic causal narratives from sparse facts.

10. Expected-evidence control

Expected-evidence structures can reduce hindsight fitting but can become a new confirmation frame.

Where they are used:

·      label predictions as provisional;

·      permit unexpected observations;

·      permit revision of the original model;

·      allow new hypotheses;

·      ask what was observed that no current account predicted;

·      do not treat absence of an expected trace as automatic proof of misstatement.

Expected-evidence recording should not be required for routine work.

11. Response-quality framework

The earlier categories of high-quality, partial and superficial response are retained, but their interpretation is revised.

11.1 Strong response

A strong response may be:

·      a concise oral explanation;

·      a short working note;

·      a changed evidence request;

·      a revised sample;

·      an escalation decision;

·      a justified no-change decision;

·      a clear acknowledgement of uncertainty.

It need not be polished.

11.2 Developing response

A developing response shows some intended thinking but may contain:

·      incomplete causal reconstruction;

·      an evidence gap;

·      an unsupported inference;

·      excessive reliance on management;

·      limited integration;

·      weak calibration.

It should indicate what coaching or additional knowledge is required.

11.3 Superficial response

A superficial response can be produced without performing the intended mental operation.

Common features include:

·      copied methodology;

·      generic alternatives;

·      management explanation repeated as conclusion;

·      evidence listed without function;

·      “no exceptions” closure;

·      polished prose unsupported by entity-specific evidence.

11.4 Alternative valid strong responses

Patterns should permit more than one strong response.

Different auditors may reasonably:

·      prioritise different evidence;

·      seek support at different points;

·      reach the same conclusion through different reasoning;

·      decide that no formal hypothesis structure is necessary;

·      conclude that the existing methodology is sufficient.

12. Revised pattern-design record

The following structure is for designers, not auditors.

1.        Audit moment and user

2.        Specific thinking break

3.        Non-cognitive causes excluded or addressed

4.        Support level

5.        Minimum intervention

6.        Maximum justified scaffold

7.        Intended attention and mental operation

8.        Weak existing design

9.        Cognitive mechanism

10.  Epistemic controls

11.  Process-visible exemplars

12.  Reviewer or coach diagnostic

13.  Boilerplate, workload and retrospective-completion risks

14.  Pilot and evaluation method

Not every design record needs equal detail. Level 1 micro-patterns should normally be documented compactly.

13. Pattern coverage map

Pattern

Audit moment

Support level

Principal operation

1

Revenue analytical review

2

Explain a movement without adopting one cause

2

Receivables task allocation

1

Distinguish what cash evidence establishes

3

Inventory planning

2

Connect count planning with valuation risk

4

Impairment evaluation

3

Test linked assumptions against operating reality

5

Provision coaching

2

Separate components and conflicting evidence

6

Going-concern partner challenge

3

Test critical dependencies and decision thresholds

7

Review-note drafting

1

Identify the precise evidential gap

8

AI-supported inquiry

1–2

Preserve provenance and auditor ownership

9

Integrated training simulation

Learning

Develop prediction, updating and integration

10

Routine control testing

0–1

Avoid unnecessary hypothesis architecture

11

Sampling exception

1

Connect exception to population implication

12

Expert anomaly challenge

1

Interrupt over-familiar pattern completion

Patterns 10–12 are added to reduce the previous over-concentration on high-judgement narrative cases. The red-team review concluded that the earlier nine patterns were insufficient to establish general audit realism.

Pattern 1 — Revenue movement during planning analytical review

1. Audit moment and user

User: Audit senior, supported by manager
Moment: Planning analytical review
Decision: Whether the revenue movement changes risk assessment or planned work

2. Specific thinking break

·      movement–cause disconnection;

·      management interpretation becoming the default account;

·      analysis completed without affecting the audit response.

3. Non-cognitive factors

Before using the pattern, confirm the team has:

·      sufficient revenue-process understanding;

·      product and customer data;

·      relevant accounting knowledge;

·      access to management and operational information.

Where these are absent, provide context or data rather than asking for better hypotheses.

4. Support level

Level 2 — Guided reasoning support

Counterfactual entry conditions are met because:

·      several realities could produce the same movement;

·      different realities require different audit work;

·      management’s explanation is influential.

5. Minimum intervention

Revenue increased by 18%, mainly in November and December, while receivables rose faster and margin fell. What facts explain the movement, and what part remains unexplained?

This may be sufficient for an experienced senior.

6. Maximum justified scaffold

For a significant or unusual movement, the maximum structure is:

·      observed facts;

·      management factual account and interpretation;

·      two or three materially different explanations;

·      evidence likely to distinguish them;

·      effect on audit work.

Do not require:

·      a fixed number of hypotheses;

·      a full CLEAR table;

·      formal recording of every discarded possibility.

7. Intended thinking

The auditor should:

·      reconstruct the movement through customers, volumes, prices, terms and timing;

·      separate observed facts from causal inference;

·      determine whether genuine trading, commercial changes or recognition issues remain plausible;

·      select evidence that could change the audit response.

8. Weak design

Obtain management’s explanation for the movement, corroborate it and conclude whether the movement is reasonable.

9. Cognitive mechanism

The intervention:

·      presents the pattern before the conclusion;

·      makes related movements visible;

·      asks for an unresolved component;

·      avoids making formal hypotheses mandatory;

·      connects analysis to planned work.

10. Epistemic controls

Separate:

·      observed movement;

·      management factual information;

·      management causal interpretation;

·      auditor inference;

·      actual evidence;

·      unresolved movement.

Expected evidence remains provisional. The auditor should also ask:

What did we observe that none of the current explanations predicted?

11. Process-visible exemplars

Strong contemporaneous note

£19.3m increase; £9.6m in Nov/Dec.
Customer count +3%; list prices +5%: not enough to explain increase.
Management identifies four new customers and discount campaign.
Need customer/product bridge and December dispatch terms.
Receivables +31% and margin down: extended terms or pull-forward still possible.
Change: separate high-growth customers and post-year dispatch population.

This is strong because it changes the work and keeps an unresolved question visible.

Alternative strong oral response

“The new customers probably explain part of it, but the timing and receivables pattern mean I am not yet treating growth as the complete explanation. I want to see the customer bridge and dispatch pattern before deciding whether this is mainly a commercial or recognition issue.”

Developing response

“The increase appears linked to new customers. Receivables also increased, so I will increase cut-off testing.”

The response notices a relationship but does not yet know what evidence distinguishes extended terms, delayed cash and premature recognition.

Superficial response

“Management explained the increase through new customers and demand. The movement is reasonable.”

12. Reviewer or coach diagnostic

·      Which part of the increase is evidenced and which is inferred?

·      What did management provide as fact versus interpretation?

·      Which evidence would change the planned work?

·      Is a concerning explanation still plausible on current facts?

·      What would allow it to be deprioritised?

13. Workload and boilerplate risk

Reduce the design where:

·      the movement is fully explained by reliable disaggregated data;

·      standard revenue work already addresses the residual risk;

·      the user is generating generic fraud alternatives.

Test under the time normally available for planning analytics.

14. Pilot method

Compare:

·      existing explanation-and-corroboration prompt;

·      micro-intervention;

·      guided structure.

Assess:

·      quality of disaggregation;

·      evidence selected;

·      changes to risk and work;

·      time used;

·      independent performance on a changed case.

Pattern 2 — Trade receivables during task allocation

1. Audit moment and user

User: Associate
Moment: Allocation of subsequent-cash work
Decision: What the evidence says about selected balances and whether escalation is needed

2. Specific thinking break

·      procedure–purpose disconnection;

·      cash receipt treated as universal proof;

·      exception recorded without customer-level interpretation.

3. Non-cognitive factors

The associate needs:

·      a basic model of sale, invoice, due date, dispute, credit and payment;

·      clear criteria for escalation;

·      context on what subsequent cash can and cannot establish.

Do not ask the associate to generate a broad hypothesis field.

4. Support level

Level 1 — Micro-intervention

5. Minimum intervention

For each selected receipt, state what it establishes and what remains unresolved.

6. Maximum justified scaffold

A compact table may be used:

Receipt or event

What it establishes

What remains unresolved

Escalation

No additional CLEAR fields are needed.

7. Intended thinking

The associate should distinguish:

·      existence;

·      settlement;

·      timing;

·      dispute;

·      credit note;

·      recoverability;

·      possible revenue implications.

8. Weak design

Agree the selected balance to post-year-end cash and record any exception.

9. Cognitive mechanism

The establishes/unresolved contrast interrupts the assumption that:

cash received = balance fully valid and recoverable.

10. Epistemic controls

The associate should record only observable events:

·      amount received;

·      date;

·      invoice allocation;

·      dispute;

·      credit;

·      unpaid remainder.

Interpretation beyond their knowledge should be escalated rather than guessed.

11. Process-visible exemplars

Strong working note

Invoice balance £420k.
£370k received against invoice.
£35k credit raised after year end; £15k still disputed.
Establishes settlement of £370k.
Does not establish validity/recoverability of £50k.
Escalate: assess credit note against revenue and customer-specific allowance.

Developing response

£370k received; £50k difference requires follow-up.

Superficial response

£370k agreed to bank. Exception £50k.

12. Reviewer or coach diagnostic

What does the cash evidence not prove?

Then:

What real customer event explains the remaining balance?

13. Workload and boilerplate risk

The intervention should add seconds, not a separate paper.

Reject a design that requires a customer narrative for every routine settled item.

14. Pilot method

Measure whether the micro-prompt improves:

·      identification of disputes and credits;

·      appropriate escalation;

·      links to revenue and loss allowance;

·      task time;

·      quality after the prompt is removed.

Pattern 3 — Inventory during planning

1. Audit moment and user

Users: Manager, senior and inventory lead
Moment: Count and valuation planning
Decision: Which locations, products and observations matter

2. Specific thinking break

·      count logistics disconnected from valuation risk;

·      management’s stock-build explanation adopted without testing;

·      inventory treated as a quantity rather than physical goods exposed to demand and ageing.

3. Non-cognitive factors

The team needs:

·      inventory data by age, category and location;

·      understanding of the purchasing and sales cycle;

·      knowledge of count and NRV requirements.

4. Support level

Level 2 — Guided reasoning support

5. Minimum intervention

If the stock build is planned and recoverable, what should we see in the warehouse and in post-year-end sales? What would look different if demand had deteriorated?

6. Maximum justified scaffold

At most:

·      observed pattern;

·      management factual account;

·      two materially different realities;

·      expected physical or data differences;

·      effect on count and valuation work.

7. Intended thinking

Connect:

·      product;

·      age;

·      location;

·      physical condition;

·      movement;

·      demand;

·      markdown;

·      saleability.

8. Weak design

Discuss the increase, count locations and NRV approach.

9. Cognitive mechanism

The contrast between planned build and demand deterioration directs attention to:

·      different warehouse conditions;

·      different sales patterns;

·      different site and product selections.

10. Epistemic controls

Distinguish:

·      inventory observed physically;

·      ageing recorded in systems;

·      management explanation;

·      inferred demand;

·      actual sales and margins;

·      goods or channels not included in the evidence.

11. Process-visible exemplars

Strong planning decision

“We do not need a full hypothesis map. Two realities matter: recent relaunch stock versus older demand-failed stock. We will split the population by range and age, attend the overflow warehouse, and ask observers to record condition and apparent movement for the older seasonal lines.”

Strong count-team cue

“For the selected older lines, note condition, packaging, segregation and evidence of movement. Do not conclude on NRV at the count.”

Developing response

“Slow-moving risk is higher, so we will increase NRV samples.”

Superficial response

“The increase relates to the relaunch. Standard count and NRV procedures will be performed.”

12. Reviewer or coach diagnostic

·      What observation would distinguish planned build from failed demand?

·      Which location is most diagnostic?

·      What did count planning change?

·      What unexpected warehouse observation would create a new line of inquiry?

13. Workload and boilerplate risk

Do not require all count teams to complete a narrative business-reality section.

Use targeted cues only for relevant products and locations.

14. Pilot method

Compare existing and targeted count instructions.

Assess:

·      site and product selection;

·      observations raised;

·      effect on valuation work;

·      observer burden;

·      findings outside the predicted pattern.

Pattern 4 — Impairment evidence evaluation

1. Audit moment and user

User: Manager, partner and valuation specialist
Moment: Evaluation of a significant impairment model
Decision: Whether recoverable amount and disclosures are supportable

2. Specific thinking break

·      model–operating-reality disconnection;

·      technical precision masking unsupported assumptions;

·      evidence from common management sources treated as independent.

3. Non-cognitive factors

The team needs:

·      accounting-estimate expertise;

·      valuation support;

·      entity and market knowledge;

·      access to operating evidence;

·      clear ownership between audit team and specialist.

4. Support level

Level 3 — Integrated judgement architecture

5. Minimum intervention

For expert discussion:

What real operating events carry the headroom, and which evidence source is least independent of management’s forecast?

6. Maximum justified scaffold

For a significant judgement, a limited integrated structure may include:

·      critical assumptions;

·      real conditions required;

·      observed evidence;

·      source origin;

·      linked downside;

·      decision threshold.

Do not require a separate narrative for every assumption.

7. Intended thinking

Translate:

·      growth;

·      margin;

·      savings;

·      discount rate;

·      terminal value

into:

·      contracts;

·      pricing;

·      demand;

·      utilisation;

·      cost actions;

·      timing;

·      execution capability.

8. Weak design

Recalculate the model, agree assumptions to the approved forecast, assess the discount rate and perform sensitivities.

9. Cognitive mechanism

The design identifies which real event carries the conclusion and whether apparently different evidence shares the same management origin.

10. Epistemic controls

For each critical dependency distinguish:

·      observed operating evidence;

·      management forecast;

·      inferred causal effect;

·      independent or external evidence;

·      unresolved assumption.

Expected evidence should not constrain attention to unexpected indicators.

11. Process-visible exemplars

Strong manager note

Headroom mainly depends on margin recovery and £6m savings.
Contract renewal evidence remains non-binding.
Savings plan exists; only £0.8m realised and milestones delayed.
Specialist rate work addresses discount rate, not cash-flow achievability.
Need integrated downside, not separate 1% sensitivities.

Strong partner challenge

“Which single operating dependency would move us across the impairment threshold, and what evidence do we have that does not originate from management’s plan?”

Developing response

“The assumptions appear optimistic and headroom is sensitive to margin.”

Superficial response

“The model is mathematically correct, the rate is in range and the forecast is board approved.”

12. Reviewer or coach diagnostic

·      Which operating event does most work in the conclusion?

·      Which evidence sources share the same origin?

·      What changed because prior forecasts were missed?

·      What combined scenario crosses the decision threshold?

·      When can the downside be stopped or deprioritised?

13. Workload and boilerplate risk

Apply only to critical dependencies.

Do not decompose every model line into a separate business-reality field.

14. Pilot method

Use blinded scoring of:

·      evidence selection;

·      source-dependency detection;

·      threshold judgement;

·      effect on conclusion;

·      performance without the scaffold;

·      time displacement.

Pattern 5 — Provision coaching during execution

1. Audit moment and user

Users: Manager coaching senior
Moment: Mid-execution on a complex project provision
Decision: What remains uncertain and what action is required

2. Specific thinking break

·      evidence collected but not integrated;

·      several uncertainties combined in one provision;

·      detailed schedules creating false comfort.

3. Non-cognitive factors

Determine whether the senior lacks:

·      legal knowledge;

·      engineering knowledge;

·      contract knowledge;

·      authority to seek specialist support.

Teaching or specialist involvement may be required.

4. Support level

Level 2 — Guided coaching

5. Minimum intervention

Explain the provision without using the provision schedule. Which component can you explain least well?

6. Maximum justified scaffold

The coach may separate:

·      known cost;

·      uncertain cost;

·      obligation;

·      recovery;

·      legal outcome;

·      timing.

No formal CLEAR form is required.

7. Intended thinking

The senior should:

·      construct the project reality;

·      identify distinct uncertainties;

·      recognise conflicting evidence;

·      identify the specific knowledge or evidence gap;

·      decide what action follows.

8. Weak design

Add more detail on the assumptions and legal evidence. Challenge management.

9. Cognitive mechanism

Eliciting the current model before giving the answer exposes:

·      what is understood;

·      what is merely repeated;

·      where teaching versus challenge is needed.

10. Epistemic controls

The coach should separate:

·      fact;

·      management estimate;

·      legal interpretation;

·      engineering interpretation;

·      auditor inference.

Metacognition and adaptation should not become separate boxes.

11. Process-visible exemplars

Strong coaching dialogue

Manager: What part of the provision is most uncertain?

Senior: The remediation scope. I originally focused on the legal penalty because the correspondence was more visible.

Manager: What changed?

Senior: Engineering emails suggest a third zone may be affected. I need the latest testing and to separate remediation cost from the penalty estimate.

Manager: What support do you need?

Senior: Engineering-specialist input before I can assess the range.

Developing response

“There is uncertainty over remediation and penalties, so more evidence is needed.”

Superficial response

“The provision agrees to the supporting reports and is within a reasonable range.”

12. Reviewer or coach diagnostic

·      Can the senior explain the matter independently?

·      Is the gap cognitive or technical?

·      What changed after coaching?

·      Did the senior retain ownership?

·      Was unnecessary documentation avoided?

13. Workload and boilerplate risk

Document the resulting material work or judgement—not the full coaching dialogue—unless methodology requires otherwise.

14. Pilot method

Assess:

·      quality of next action;

·      independent explanation;

·      appropriate help-seeking;

·      later transfer;

·      manager time;

·      whether review notes decrease.

Pattern 6 — Going-concern partner challenge

1. Audit moment and user

User: Engagement partner
Moment: Late-stage significant-judgement meeting
Decision: Going-concern basis, material uncertainty, disclosure and reporting

2. Specific thinking break

·      forecast–event disconnection;

·      separate sensitivities failing to capture linked outcomes;

·      completion and reporting pressure encouraging closure.

3. Non-cognitive factors

The team needs:

·      going-concern technical expertise;

·      current financing and covenant evidence;

·      realistic cash-flow data;

·      authority to reopen work;

·      access to consultation.

4. Support level

Level 3 — Integrated judgement architecture

5. Minimum intervention

For an experienced partner:

What is the weakest dependency in the conclusion, and what fact would move the reporting decision?

6. Maximum justified scaffold

A compact dependency map may be used for:

·      event;

·      deadline;

·      evidence status;

·      cash consequence;

·      decision threshold.

Do not require a full CLEAR narrative.

7. Intended thinking

The partner should:

·      convert the forecast into dated events;

·      assess which events are controllable;

·      identify linked adverse outcomes;

·      distinguish lender discussion from commitment;

·      test the reporting threshold.

8. Weak design

Review forecasts, sensitivities, lender correspondence and disclosure. Confirm whether material uncertainty exists.

9. Cognitive mechanism

The weakest-dependency and changed-fact questions interrupt broad comfort and direct attention to:

·      fragility;

·      timing;

·      threshold;

·      consequence.

10. Epistemic controls

Distinguish:

·      signed agreement;

·      lender discussion;

·      management intention;

·      controllable mitigation;

·      contingent action;

·      inferred probability.

A concerning scenario should remain active only where its plausibility and consequence justify it.

11. Process-visible exemplars

Strong partner discussion note

Facility renewal needed by March. No commitment.
Lender requests revised covenants and security.
Contract renewal and covenant compliance are linked.
Combined delayed savings/lower-margin renewal leaves <£1m liquidity.
Reporting decision changes if committed finance not obtained before signing.

Alternative strong challenge

“I do not need another sensitivity. Show me the first month in which the downside becomes irreversible and which mitigating action is actually executable before then.”

Developing response

“The base case has headroom but refinancing remains uncertain.”

Superficial response

“Management expects renewal and sensitivities show positive cash.”

12. Reviewer or coach diagnostic

·      What dependency carries the conclusion?

·      What is evidenced versus intended?

·      Which adverse events are correlated?

·      What fact changes the reporting decision?

·      Is the conclusion being retained because of evidence or completion pressure?

13. Workload and boilerplate risk

Expert application should remain compact.

Do not impose junior-level hypothesis or reflection fields on the partner.

14. Pilot method

Assess:

·      threshold recognition;

·      linked-scenario reasoning;

·      reporting calibration;

·      oral reconstruction;

·      performance under time pressure;

·      quality without the prompt.

Pattern 7 — Diagnostic review note

1. Audit moment and user

User: Manager or partner
Moment: Review of a specific evidence gap
Decision: How to communicate the gap without replacing the preparer’s judgement

2. Specific thinking break

·      review note asks for more detail rather than resolving the real issue;

·      preparer guesses the reviewer’s preferred answer;

·      improved prose substitutes for improved evidence.

3. Non-cognitive factors

Confirm whether the issue arises from:

·      missing knowledge;

·      unclear methodology;

·      inaccessible evidence;

·      inadequate original briefing.

4. Support level

Level 1 — Micro-intervention

5. Minimum intervention

A diagnostic review note should state:

1.        what the work currently establishes;

2.        what it does not establish;

3.        why the gap matters;

4.        the question or decision to resolve.

6. Maximum justified scaffold

One concise note.

Do not add a separate CLEAR review form.

7. Intended thinking

The preparer should understand the precise evidential or reasoning gap and decide how it affects the work or conclusion.

8. Weak design

Please add more evidence and explain why the conclusion is appropriate.

9. Cognitive mechanism

The note identifies the claim rather than inviting generic additional documentation.

10. Epistemic controls

The reviewer should not prescribe the conclusion unless a mandatory technical response applies.

11. Process-visible exemplars

Strong note

The invoice and later cash establish that a transaction was recorded and paid. They do not establish that the contractual acceptance condition was met before year end. Determine when acceptance occurred and update revenue, receivables and related testing as necessary.

Developing response

“January acceptance may affect cut-off. Further work in progress.”

Superficial response

“Additional invoice and bank support attached. Conclusion unchanged.”

12. Reviewer or coach diagnostic

What is different in the work or conclusion because the note was addressed?

13. Workload and boilerplate risk

Clear notes should reduce review cycles.

Track whether diagnostic notes become standard scripts detached from the actual claim.

14. Pilot method

Compare generic and diagnostic notes for:

·      number of review cycles;

·      relevance of evidence;

·      conclusion change;

·      preparer understanding;

·      review time.

Pattern 8 — AI-supported inquiry

1. Audit moment and user

User: Senior or manager
Moment: Early evidence planning or exploratory analysis
Decision: What human investigation should follow

2. Specific thinking break

·      AI-generated fluency mistaken for audit reasoning;

·      management data and explanation blended;

·      source limitations hidden;

·      auditor ownership weakened.

3. Non-cognitive factors

Confirm:

·      approved AI use;

·      data permissions;

·      source availability;

·      methodology requirements;

·      user knowledge sufficient to evaluate output.

4. Support level

Level 1 or 2

AI should normally support one cognitive operation rather than perform a complete CLEAR analysis.

5. Minimum intervention

AI asks:

What decision are you trying to support, and which uploaded sources directly evidence it?

6. Maximum justified scaffold

AI may help:

·      separate facts, explanations and missing data;

·      identify provenance;

·      suggest conditional questions;

·      compare evidence patterns;

·      flag unresolved claims.

It should not draft a final conclusion until the auditor has reconstructed and verified the reasoning.

7. Intended thinking

The auditor should:

·      remain responsible for framing;

·      inspect source provenance;

·      identify missing populations;

·      distinguish suggestion from evidence;

·      decide what human work follows.

8. Weak design

Analyse this issue using CLEAR and draft the audit conclusion.

9. Cognitive mechanism

Elicit-before-generate and visible provenance prevent the AI from silently supplying the mental model.

10. Epistemic controls

Separate:

·      source fact;

·      management explanation;

·      AI inference;

·      auditor inference;

·      missing information;

·      verified evidence.

AI should explicitly state when evidence is compatible with several explanations.

11. Process-visible exemplars

Strong auditor note

AI found 82% sell-through, but dataset covers online channel only.
Store markdowns and wholesale disposals absent.
No conclusion.
Action: obtain full-channel sales and margin data; reconcile population.

Developing response

“AI identified slow-moving risk and more sales data are needed.”

Superficial response

“AI concluded that the provision is reasonable based on post-year-end sales.”

12. Reviewer or coach diagnostic

·      Which reasoning was performed by the auditor?

·      What did the AI actually inspect?

·      What source limitation changed the plan?

·      Can the auditor explain the matter without the AI output?

·      Was generated wording stronger than the evidence?

13. Workload and boilerplate risk

Do not measure success through:

·      output volume;

·      number of hypotheses;

·      narrative completeness;

·      time saved alone.

Monitor whether AI use erodes unaided framing and evidence evaluation.

14. Pilot method

Compare:

1.        unaided work;

2.        AI-first conclusion generation;

3.        structured AI-supported inquiry.

Assess:

·      prompted performance;

·      independent performance;

·      transfer;

·      retention;

·      judgement effect;

·      burden;

·      unsupported confidence.

Pattern 9 — Integrated training simulation

1. Audit moment and user

Users: Seniors and new managers
Moment: Capability development
Decision: Whether participants can transfer CLEAR thinking to a changed case

2. Specific thinking break

·      framework recall mistaken for capability;

·      model-answer imitation;

·      no evidence-led updating;

·      linked judgements considered separately.

3. Non-cognitive factors

Participants require prior technical teaching.

The simulation should not be used to teach technical requirements they have never encountered.

4. Support level

Learning architecture, progressively reduced

5. Minimum intervention

For experienced participants:

·      commit to an initial view;

·      receive staged evidence;

·      state what changed;

·      explain the next audit action.

6. Maximum justified scaffold

Early practice may include:

·      factual reconstruction;

·      guided alternative comparison;

·      expected evidence;

·      evidence-function sorting;

·      updating.

The scaffold must then be withdrawn.

7. Intended thinking

Participants should:

·      distinguish fact, explanation and conclusion;

·      predict evidence;

·      update after contradictory information;

·      integrate linked judgements;

·      recognise when help is required.

8. Weak design

Presentation on CLEAR followed by a model-answer case.

9. Cognitive mechanism

Staged evidence exposes whether participants genuinely update rather than produce a final polished explanation.

10. Epistemic controls

Include cases where:

·      management is ultimately supported;

·      the concerning explanation is deprioritised;

·      unexpected evidence creates a new account;

·      no additional work is the proportionate answer.

11. Process-visible exemplars

Strong learner response

“I initially treated the insurance recovery as likely because legal correspondence existed. The reservation clause means the evidence does not support recognition. I need to assess the provision gross and update liquidity.”

Developing response

“The provision may be higher and this may affect going concern.”

Superficial response

“Management’s provision is reasonable and forecasts remain positive.”

12. Reviewer or coach diagnostic

·      What changed?

·      Which evidence caused the change?

·      Could the participant perform the operation without the scaffold?

·      Can they transfer it to a different account?

13. Workload and boilerplate risk

Do not assess participants through:

·      CLEAR terminology;

·      number of alternatives;

·      narrative length;

·      resemblance to the model response.

14. Pilot method

Assess:

·      prompted performance;

·      unaided performance;

·      transfer to a changed audit area;

·      retention after six weeks;

·      decision quality;

·      time and cognitive burden.

Pattern 10 — Routine control testing

1. Audit moment and user

User: Associate
Moment: Testing whether a specified control operated
Decision: Whether the control was performed as designed for selected items

2. Specific thinking break

·      purpose of the control test not understood;

·      exception cleared mechanically;

·      unnecessary hypothesis generation around a tightly defined task.

3. Non-cognitive factors

The associate needs:

·      clear control definition;

·      evidence criteria;

·      exception and escalation guidance.

4. Support level

Level 0 or 1

Explicit counterfactual thinking is normally unnecessary.

5. Minimum intervention

What does this evidence establish about the control, and what fact would make the item an exception?

6. Maximum justified scaffold

No more than:

·      control criterion;

·      evidence;

·      result;

·      exception implication.

7. Intended thinking

Purpose-led and connected thinking are sufficient:

·      understand the control action;

·      understand the evidence generated;

·      determine whether it occurred.

8. Weak design

Complete CLEAR for each control-test sample.

9. Cognitive mechanism

The intervention connects evidence to the defined control criterion without manufacturing alternative hypotheses.

10. Epistemic controls

Do not infer control operation from:

·      presence of a document alone;

·      retrospective approval;

·      evidence generated outside the control process.

11. Process-visible exemplars

Strong note

Evidence shows review completed before payment release. Reviewer identity and date agree to control design. No exception.

Developing response

Review sign-off present; timing not yet checked.

Superficial response

Document inspected. No exception.

12. Reviewer or coach diagnostic

What feature of the evidence shows that the control operated as designed?

13. Workload and boilerplate risk

A full CLEAR structure would be disproportionate and should be prohibited.

14. Pilot method

Test whether the single purpose/evidence prompt improves exception identification without increasing sample time materially.

Pattern 11 — Sampling exception and population implication

1. Audit moment and user

User: Senior
Moment: Evaluation of a substantive-testing exception
Decision: Whether the exception is isolated, systematic or affects the wider population

2. Specific thinking break

·      exception resolved locally without considering population implication;

·      cause inferred without evidence;

·      additional samples selected mechanically.

3. Non-cognitive factors

The user needs:

·      sampling-methodology knowledge;

·      population information;

·      authority to modify work.

4. Support level

Level 1 — Micro-intervention

5. Minimum intervention

What caused the exception, and what characteristic identifies where else it could occur?

6. Maximum justified scaffold

A short exception record:

·      observed exception;

·      supported cause;

·      relevant population characteristic;

·      changed response.

7. Intended thinking

Connect the individual finding to:

·      process;

·      population;

·      risk;

·      further work.

8. Weak design

Investigate the exception and select additional items.

9. Cognitive mechanism

The population-characteristic question prevents arbitrary extension and requires a supported basis for targeting further work.

10. Epistemic controls

Distinguish:

·      observed error;

·      management explanation;

·      auditor-inferred cause;

·      evidence supporting the cause;

·      population to which it could apply.

11. Process-visible exemplars

Strong note

Exception: invoice dated before dispatch.
Cause not yet established. All similar items were generated by manual month-end journal.
Extend work to full manual-journal population, not random additional invoices.

Developing response

Cut-off exception noted; increase sample.

Superficial response

Exception cleared with management.

12. Reviewer or coach diagnostic

Why does the proposed population follow from the evidence?

13. Workload and boilerplate risk

Do not require a full causal map for immaterial clerical exceptions.

14. Pilot method

Assess:

·      quality of population extension;

·      unnecessary additional work avoided;

·      exception causes supported;

·      transfer to another account.

Pattern 12 — Expert anomaly and decision-threshold challenge

1. Audit moment and user

User: Partner, director or experienced specialist
Moment: Review of a significant judgement that appears coherent and near completion
Decision: Whether expert pattern recognition has closed too quickly

2. Specific thinking break

·      over-familiarity;

·      rapid pattern completion;

·      evidence-source dependency;

·      conclusion retained without testing the decision threshold.

3. Non-cognitive factors

The expert must have:

·      relevant technical knowledge;

·      current entity context;

·      access to the critical evidence;

·      authority to reopen the matter.

4. Support level

Level 1 — Sparse expert interruption

5. Minimum intervention

Select one:

·      What anomaly does the current account explain least well?

·      Which evidence sources are less independent than they appear?

·      What single changed fact would move the decision?

·      What pattern from another engagement might be misleading us here?

·      What would an informed independent partner challenge first?

6. Maximum justified scaffold

One short pre-mortem or decision-threshold discussion.

Experts should not complete integrated novice scaffolds.

7. Intended thinking

Interrupt:

·      familiarity;

·      authority effects;

·      source convergence;

·      unnoticed fragility.

8. Weak design

Complete the CLEAR judgement template and record the strongest alternative.

9. Cognitive mechanism

A sparse perturbation exploits expert schemas while challenging their possible over-compression.

10. Epistemic controls

The challenge should not reward dramatic but remote alternatives.

Any counter-position must be:

·      factually plausible;

·      decision-relevant;

·      capable of being tested.

11. Process-visible exemplars

Strong partner response

“The conclusion is not sensitive to the rate. It is sensitive to whether the customer renewal occurs before the covenant test. That is the decision threshold, and the current evidence is only management correspondence.”

Developing response

“The forecast is sensitive to customer renewal.”

Superficial response

“All key assumptions have been challenged.”

12. Reviewer or coach diagnostic

·      Did the challenge identify a real dependency?

·      Did it change evidence or judgement?

·      Was the alternative plausible?

·      Was the expert’s time used proportionately?

13. Workload and boilerplate risk

Rotate challenge questions.

Do not convert them into a standard partner checklist.

14. Pilot method

Use blinded expert cases and compare:

·      existing review;

·      one targeted perturbation;

·      full structured scaffold.

Assess:

·      anomaly detection;

·      decision-threshold identification;

·      time;

·      acceptance;

·      effect on judgement.

14. Common micro-pattern library

These are candidate Level 1 interventions.

14.1 Purpose cue

What must this procedure establish?

14.2 Procedure-limit cue

What does a clean result still not establish?

14.3 Reality cue

What real event, item or condition does this number represent?

14.4 Epistemic-separation cue

Which part is observed, and which part are you inferring?

14.5 Evidence-function cue

What does this evidence establish, and what does it leave unresolved?

14.6 Unexpected-evidence cue

What did we observe that the current explanation did not predict?

14.7 Updating cue

What should change because of this finding?

A justified response may be:

No change, because…

14.8 Help-seeking cue

Is the gap one of thinking, knowledge, evidence or authority?

14.9 Population cue

What characteristic identifies where else this exception could occur?

14.10 Closure cue

What fact would require the matter to be reopened?

14.11 Expert anomaly cue

What does the current account explain least well?

14.12 Decision-threshold cue

What single changed fact would move the conclusion?

These prompts are optional. They should not be assembled automatically into a sequence.

15. Anti-pattern library

15.1 Full CLEAR completion

Complete each CLEAR section for this task.

Why it fails: Makes framework visibility the objective.

15.2 Generic alternative requirement

List three alternative explanations.

Why it fails: Rewards quantity and can expose knowledge gaps as generic speculation.

15.3 Concern inflation

State the most concerning explanation.

Why it fails: Does not test plausibility or stopping conditions.

15.4 Management-last rule

Generate alternatives before speaking to management.

Why it fails: May prevent necessary factual clarification.

15.5 Expected-evidence tunnel

Record expected evidence and test whether it is present.

Why it fails: Can suppress attention to unexpected evidence.

15.6 Narrative grounding

Tell the business story behind the number.

Why it fails: Rewards vividness without evidential testing.

15.7 Reflection declaration

Explain how you reflected and adapted.

Why it fails: Creates a retrospective account rather than supporting regulation.

15.8 More-evidence review note

Obtain further support and explain why the conclusion remains reasonable.

Why it fails: Presumes the conclusion and encourages accumulation.

15.9 AI-completed reasoning

Apply CLEAR and write the conclusion.

Why it fails: Produces the appearance of thinking without ownership.

15.10 Strong-answer copying

Use the exemplar as a guide for the audit-file response.

Why it fails: Turns cognitive examples into boilerplate.

16. Evaluation protocol

The red-team review found that the system lacked a valid way to distinguish genuine capability from artefact production.

Pattern evaluation should distinguish six outcomes.

16.1 Prompted performance

How well does the user perform while the pattern is available?

16.2 Independent performance

Can the user perform the operation after the pattern is removed?

16.3 Transfer

Can the user apply the operation to a changed:

·      account;

·      sector;

·      workflow moment;

·      evidence pattern?

16.4 Retention

Does the capability remain after time has passed?

16.5 Judgement effect

Does the pattern improve:

·      evidence selection;

·      procedure design;

·      escalation;

·      confidence calibration;

·      conclusion quality?

16.6 Burden and displacement

Measure:

·      additional time;

·      documentation volume;

·      reviewer time;

·      evidence work displaced;

·      reliance on AI;

·      retrospective completion.

16.7 Blinded evaluation

Where feasible, assess outputs without showing evaluators:

·      whether CLEAR was used;

·      which pattern was used;

·      which response contained CLEAR terminology.

Scoring should focus on:

·      audit question;

·      grounding;

·      evidence relevance;

·      epistemic calibration;

·      adaptation;

·      conclusion.

16.8 Harm testing

Test whether a pattern causes:

·      excessive alternatives;

·      suspicion bias;

·      failure to deprioritise weak hypotheses;

·      reduced attention to unexpected evidence;

·      excessive documentation;

·      slower execution without quality benefit;

·      weakened auditor confidence;

·      dependence on scaffolding;

·      reduced expert acceptance;

·      AI-enabled rationalisation.

16.9 Release threshold

A pattern should not be scaled merely because:

·      users like it;

·      files become longer;

·      reviewers report clearer narratives;

·      more alternatives are recorded;

·      CLEAR terminology appears more often.

Scaling requires evidence of improved judgement or evidence behaviour that persists beyond the scaffold and is proportionate to the burden.

17. Methodology and regulatory dependencies

Before a pattern becomes a firm-wide requirement, determine:

·      which existing methodology requirement it improves;

·      whether it duplicates current documentation;

·      what existing wording or field should be removed;

·      whether any new record is required;

·      whether discarded hypotheses should be retained;

·      whether coaching or AI interactions remain outside the audit file;

·      whether the structure changes technical ownership;

·      how the pattern interacts with supervision and review;

·      whether regulatory stakeholders consider the documentation proportionate.

CLEAR should improve the performance of existing audit requirements rather than create a parallel methodology.

18. Coverage limitations

This library now includes patterns for:

·      high-judgement analysis;

·      task allocation;

·      planning;

·      coaching;

·      review;

·      AI;

·      training;

·      routine controls;

·      sampling exceptions;

·      expert challenge.

It remains insufficient to establish general application across:

·      IT audit;

·      group audit;

·      component teams;

·      specialists;

·      service organisations;

·      completeness testing;

·      regulated entities;

·      small entities;

·      data analytics;

·      journal testing;

·      late evidence and clearance pressure;

·      remote and distributed teams.

These areas should be tested before the library is presented as generally validated.

19. Pattern approval checklist

Before approving a pattern, the design owner should confirm:

1.        What specific thinking break is being addressed?

2.        Is the problem actually cognitive?

3.        What knowledge, capacity, authority or methodology issue also exists?

4.        What support level is justified?

5.        What is the smallest intervention likely to work?

6.        What is the maximum acceptable scaffold?

7.        Is counterfactual thinking genuinely needed?

8.        Is there a sufficient factual base?

9.        How is management information treated?

10.  Are observation, inference and evidence separated?

11.  Could expected evidence create anchoring?

12.  Can the response be produced mechanically or by AI?

13.  Could a concise oral or action response demonstrate the thinking?

14.  What should remain outside the audit file?

15.  How will workload and displacement be measured?

16.  How will independent performance and transfer be tested?

17.  What harm would cause the pattern to be withdrawn?

18.  What existing methodology or content will be removed?

20. Closing control statement

The Pattern Library is successful only if it helps designers create fewer, better and more precisely timed interventions.

It should not produce:

·      more visible CLEAR activity;

·      more fields;

·      more alternatives;

·      longer explanations;

·      more review notes;

·      more AI-generated reasoning.

It should help auditors, teams and reviewers:

·      identify the actual question;

·      connect accounting representations to tested business reality;

·      evaluate what evidence establishes and leaves unresolved;

·      remain open where genuine ambiguity exists;

·      update work and confidence when evidence changes;

·      seek knowledge or support where the problem is not solvable by reflection;

·      communicate only the material reasoning needed for execution and challenge.

The governing test is:

Does the selected intervention improve the live audit judgement sufficiently to justify its cognitive, operational and documentation cost—and can the improvement be demonstrated beyond the visible CLEAR artefact?

This revision converts the library from a catalogue of full exemplar architectures into a controlled set of optional patterns, micro-interventions and testable design hypotheses.