Conversational pattern catalog¶
These five patterns cover recurring cycle blockers—not five screens or prompts. They can guide a meeting, brief, tracked dashboard or governed assistant. None requires AI.
| ID | Pattern | Moment | Minimum output |
|---|---|---|---|
CDP-001 |
Decision contract | Frame | Decision, owner, alternatives, date and expected action |
CDP-002 |
Evidence boundary | Ground | Facts, inferences, unknowns and fitness for purpose |
CDP-003 |
Option challenge | Deliberate | Comparable options, trade-offs and change conditions |
CDP-004 |
Decision–action handoff | Commit | Authorized choice, execution, guardrails and escalation |
CDP-005 |
Expectation–result review | Learn | Deviation, calibrated attribution and next-cycle adjustment |
CDP-001 — Decision contract¶
Use when: conversation begins with “analyze sales” or “what should we do?” without a choice. Convert the topic to who chooses what, among which alternatives and by when; complete PDAMR; state affected people, costs and reversibility; include no action when material; confirm with the rights holder.
Exit: a minimum Decision Brief. Stop: without a real choice or owner, reclassify as information exploration. Risk: false legitimacy if a facilitator assigns ownership without authority; bureaucracy at second order. Alternative: operational rule or decision table for stable recurring choices. PULSE: Business First, Decision First, PDAMR, Focus.
CDP-002 — Evidence boundary¶
Use when: figures, definitions, filters or explanations conflict. Restore population, period, unit, filters and metric; identify source, owner, lineage and permission; separate fact, calculation, inference, hypothesis, recommendation and unknown; assess fitness for purpose; expose counterevidence and sensitivity; answer, abstain or escalate with a verifiable reason.
Exit: bounded evidence whose conclusion does not exceed its sources. Stop: unreconciled material definitions, missing permissions or inadequate quality. Risk: over-governance and citation-as-authority. Alternative: controlled report for regulated recurring calculations. PULSE: Trust First, Context First, Evidence Before Authority, Perceive.
CDP-003 — Option challenge¶
Use when: one dominant recommendation, false dichotomy or unstructured idea list appears. Make criteria and defensible weights explicit; create materially different alternatives including no action; state mechanism, fragile assumption, cost, delay, risk and reversibility; formulate the strongest counterargument; seek a potentially superior alternative; state what changes the ranking.
Exit: auditable comparison with provisional option and calibrated confidence. Stop: apply a preapproved emergency rule when delay would increase immediate harm. Risk: option theater and numerical criteria hiding political or distributional judgment. Alternative: facilitated workshop, formal decision analysis or multidisciplinary review. PULSE: Generate options, Anticipate, Evidence Before Authority.
CDP-004 — Decision–action handoff¶
Use when: recommendation or preference exists but commitment and subsequent action are unclear. Separate recommend, decide, approve, execute, stop and answer; record chosen and rejected alternatives; translate choice into accountable action and timing; define AI permissions, review, guardrails, stop, reversal and escalation; freeze expectation and review date; require positive confirmation.
Exit: authorized decision and traceable action plan, or explicit abstention/escalation. Stop: without decision right, approval or risk-proportional reversal, only prepare the handoff. Risk: authority laundering, concentrated power and traces used to suppress dissent. Alternative: deterministic transactional workflow. PULSE: Decide and Act, Governance Enables Speed, Human-in-Control.
CDP-005 — Expectation–result review¶
Use when: a decision horizon ends. Open but do not overwrite the prior snapshot; record outcomes, process measures, guardrails, unintended effects and execution quality; describe deviation before explaining; examine counterfactual, confounding and sample size; separate ex-ante quality, execution fidelity and result; decide what to hold, scale, stop, reverse or investigate.
Exit: learning that changes perception, criterion, action or control. Stop: without a prior snapshot, label the review exploratory and prohibit strong expectation-accuracy or effect claims. Risk: institutionalizing noise and defensive forecasting under punitive culture. Alternative: causal evaluation, independent audit or aggregation across cycles. PULSE: Learn, Value Must Be Observable, Decision Measurement Record.
What would validate or refute the catalog¶
The conclusion changes if longitudinal use shows no improvement in completeness, traceability or action fidelity; materially higher latency without lower error; gaming, conformity pressure or improper delegation; missing decision classes; or better performance as non-conversational rules, forms or workflows. Until then the catalog remains candidate-synthesis.