Critical Reasoning And Argument

Critical Reasoning And Argument

This is the engine's universal thinking layer. It prevents the engine from merely collecting information and arranging it into plausible prose: every serious output must expose its question, terms, evidence, assumptions, inference chain, countercase, limits, and action implications.

Use When

Use this skill for all forms of writing and reporting that contain a claim or conclusion: academic essays, theses, dissertations, literature reviews, systematic reviews, intelligence products, market analyses, history essays, business reports, proposals, board briefs, due diligence, policy analysis, legal argument, technical recommendations, and executive communication.

Do Not Use When

• The task is purely formatting, file movement, or mechanical conversion.
• The output contains no claim, recommendation, interpretation, evaluation, forecast, or decision support.
• A domain skill already handles a narrow factual extraction and no conclusion is being drawn.

Reasoning Intake Guidance

• The user's real question, decision, audience, and output type.
• Available sources, data, notes, or draft text.
• Known constraints, scope boundaries, definitions, and country or domain context.
• Any required confidence, citation, or evidence standard.

Non-negotiable rule

No conclusion ships until it has passed the reasoning protocol in references/reasoning-protocol.md.

For research work, this skill operates after source-evaluation: source quality determines what can be used as evidence; this skill determines whether the evidence supports the conclusion. For intelligence or forward-looking work, run analytic-tradecraft after this skill to add probability language, hypothesis handling, and uncertainty discipline.

Reasoning Method Detail

1. Frame the real question. Convert the prompt into a precise question the output must answer. If the prompt asks for "a report", infer the decision, controversy, interpretation, or action the report is meant to support.
2. Define scope and terms. Identify key terms, boundaries, jurisdiction, time period, population, market, method, or discipline. Flag ambiguous or loaded language.
3. Separate the materials. Distinguish facts, claims, assumptions, values, interpretations, inferences, recommendations, and unknowns.
4. Map the argument. Build the chain: main conclusion -> intermediate conclusions -> premises/evidence -> warrants/linking assumptions. Use the table in references/reasoning-protocol.md.
5. Test evidence quality. Apply source-evaluation; then ask whether the available evidence is relevant, sufficient, representative, current, and independent.
6. Run the countercase. State the strongest plausible objection, alternative explanation, rival hypothesis, or historiographical/market/technical counter-reading.
7. Audit inference quality. Test causal claims, generalisations, analogies, statistics, definitions, conditionals, necessity/sufficiency, and certainty calibration.
8. Check coherence and fallacies. Look for contradiction, circular reasoning, equivocation, straw-manning, false alternatives, hasty generalisation, post hoc causation, appeal to authority, appeal to majority, and unsupported value leaps.
9. Choose the action form. Route the tested reasoning into the right structure and tone using references/audience-action-forms.md.
10. Run the serious-analysis overlay. Use references/essential-questions-mental-models-strategic-logic.md when the output must show deeper thought, business sense, feasibility, strategic judgment, or problem-solving logic.
11. Ship with limits. State what the conclusion means, what it does not mean, confidence limits, operational implications, and next actions.

Quality Standards

Every output using this skill must be:

• Convincing: the reader can see why the conclusion follows from the evidence.
• Authentic: the tone fits the institution, audience, discipline, and stakes; no generic filler or artificial certainty.
• Realistic: recommendations name constraints, tradeoffs, risks, time, cost, incentives, and implementation friction.
• Actionable: the reader knows what decision, interpretation, next step, or research move follows.
• Proportionate: claim strength matches evidence strength; uncertainty is not hidden.

References

Domain Routing

Anti-Patterns

• Data dump with no conclusion or decision logic.
• Confident conclusion with no visible evidence-to-inference chain.
• Source summary pretending to be analysis.
• Annotated bibliography pretending to be a literature review.
• Recommendation without owner, timeline, cost, constraint, or risk.
• History narrative with chronology but no causation or interpretive argument.
• Market report with trends but no buyer, competitor, margin, regulation, or execution implication.
• Intelligence judgment that conflates source reliability with probability of the event.
• Academic argument that cites many sources but does not show the gap, method logic, or contribution.
• Counterargument dismissed by tone instead of defeated by evidence and reasoning.
• "It is clear that" language where evidence is partial, contested, stale, or indirect.

Reasoning Deliverable Detail

• A precise core question and scope.
• Argument map for each load-bearing conclusion.
• Evidence, assumptions, warrants, countercase, limits, and implications.
• Calibrated conclusion or recommendation.
• Revision notes where the reasoning does not yet support the intended output.

Universal Ship Gate

• Core question stated.
• Key terms and scope defined.
• Facts, assumptions, values, inferences, conclusions, and recommendations separated.
• Argument map completed for every load-bearing conclusion.
• Source quality checked through source-evaluation.
• Evidence relevance, sufficiency, representativeness, independence, and currency checked.
• Strongest objection or alternative explanation addressed.
• Essential questions, mental models, feasibility, and strategic-fit checks applied where stakes require them.
• Causal, analogical, statistical, generalisation, and definition claims audited where present.
• Fallacy/coherence audit completed.
• Certainty calibrated to evidence strength.
• Output form selected for audience and action.
• Limits, risks, implications, and next actions stated.

Companion Skills

Inputs

Capability Contract

Critique defaults to read-only. Rewriting, publishing, operational decisions, or certifying correctness requires explicit authority.

Degraded Mode

With incomplete evidence, return a qualified argument map and unresolved questions; do not convert plausibility into proof.

Decision Rules

Reasoning Scenario

A recommendation supported only by correlation is qualified until the causal warrant is established.

Companion Skills

• source-evaluation - mandatory for source-backed research; establishes evidence discipline.
• analytic-tradecraft - intelligence, uncertainty, forecasts, contested evidence, and adversarial settings.
• academic-writing - academic form, originality, citation, synthesis, register, and source-away composition.
• academic-reporting-standards - thesis/dissertation examination standards and study-reporting checklists.
• report-and-proposal-craft - business reports, proposals, white papers, and formal research reports.
• business-writing - short-form professional prose.
• executive-communication - executive summaries, board briefs, and top-down communication.
• research-design - research questions, methods, and contribution logic.
• research-techniques - source mining, gap analysis, synthesis, and verification workflows.
• systems-process-requirements - systems, processes, workflows, requirements, scope, interfaces, data architecture, and design-system descriptions.

Workflow

1. State the conclusion, audience, decision, and standard of proof.
2. Map claims to evidence, warrants, assumptions, countercases, and implications.
3. Stop when a load-bearing claim lacks a warrant or supporting evidence.
4. Recover by sourcing, qualifying, restructuring, or removing the failed claim before resuming.
5. Test the strongest countercase and calibrate the final conclusion.

Outputs

Evidence Produced

Worked Example

A recommendation based only on correlation is qualified until a defensible causal warrant or explicitly limited decision rationale is supplied.