Interaction and elicitation

PUBLISHED

Obtaining reliable results through deliberate interaction.

Statements

—

Version

v1.29

Last reviewed

30.08.2026

In scope

Task framing and decomposition

Provision of context, and judgement about what context is required

Specification of constraints, format and success conditions

Iterative refinement against a defined objective

Role, persona and perspective techniques

Elicitation of reasoning, and its limits as evidence

Diagnosis of failure within a response

Recognition that the system is the wrong instrument for the task

Construction and maintenance of reusable interaction assets

Out of scope

Excluded	Where it lives
Programmatic and API interaction	`D4`
Systematic evaluation of output quality	`D6`
Injection and adversarial input	`D8`

Boundary notes

Where this domain abuts another, and how the line is drawn

Against D4 Automation and Integration

D3 covers a person interacting with a system. D4 begins where a system interacts with a system.

Against D6 Evaluation and Assurance

D3 covers judging a single response in context. D6 covers judging a system's output systematically and with evidence.

Against D8 Security

D3 techniques create the surface D8 defends. Cross-reference deliberately rather than duplicating.

Statements by level

Every identifier is a permanent address. Indicators are normative; they state what would be observed in a person who meets the statement.

L1 Aware

4 statements

D3.L1.01Knowledge

Describe what an instruction to an AI system needs to contain to be usable.

Indicators

Names task, context, format and constraint as the parts of an instruction

Recognises an instruction that omits the material the task needs

States that the system cannot ask for what it was not given

D3.L1.02Knowledge

State that the wording of an instruction changes the output.

Indicators

Identifies wording that produced a different result

Recognises that a small change may alter output substantially

Avoids treating a single phrasing as authoritative

D3.L1.03Judgement

Recognise when an output has answered a different question from the one asked.

Indicators

Compares the output against the request rather than reading it alone

Identifies a substituted, narrowed or widened question

Re-asks rather than accepting the near answer

D3.L1.04Knowledge

State that a system will answer rather than decline when a request is unclear.

Indicators

Recognises an assumption the system made and did not state

Supplies the missing detail rather than relying on inference

Treats a confident answer to an unclear request as a warning

L2 Applied

6 statements

D3.L2.01Skill

Write an instruction stating task, context, format and constraints.

Indicators

States the output form required, not only the topic

States the constraints that must not be broken

Provides the context needed rather than assuming it

D3.L2.02Skill

Supply the material a system needs rather than relying on its recall.

Indicators

Identifies what the system cannot know and supplies it

Supplies material in a usable form

Cites the material supplied in the record of the work

D3.L2.03Skill

Revise an instruction in response to a specific defect in the output.

Indicators

Names the defect before changing the instruction

Changes one thing at a time and observes the effect

Stops when the defect is resolved rather than continuing to tune

D3.L2.04Skill

Use examples to establish the required form of an output.

Indicators

Selects examples that show the form rather than the content

Includes an example of what to avoid where the risk warrants it

Checks that the examples do not narrow the task unintentionally

D3.L2.05Judgement

Recognise and interrupt an unproductive exchange rather than continuing it.

Indicators

Identifies repetition, drift or accumulated error in a session

Restarts with a corrected instruction rather than patching further

Records what did not work

D3.L2.06Skill

Record an instruction that worked so that it can be reused and reviewed.

Indicators

Records the instruction, the system and the date together

States the task the instruction is fit for

Stores it where colleagues doing the same task will find it

L3 Proficient

6 statements

D3.L3.01Skill

Design reusable instruction patterns for a recurring task.

Indicators

Separates the stable instruction from the case-specific input

States the scope the pattern is fit for

Provides for the pattern being used by someone who did not write it

D3.L3.02Skill

Decompose a task into steps a system can perform reliably.

Indicators

Identifies the step at which reliability breaks down

Places verification between steps where consequence warrants it

Justifies the decomposition rather than assuming smaller is better

D3.L3.03Judgement

Diagnose whether a failure lies in the instruction, the material or the system.

Indicators

Isolates one variable at a time

States the evidence for the conclusion reached

Avoids attributing to the model what the instruction caused

D3.L3.04Skill

Design instructions that fail safely when input falls outside their scope.

Indicators

States what the instruction must do with out-of-scope input

Prefers refusal or escalation to a best attempt

Tests the out-of-scope path, not only the intended one

D3.L3.05Skill

Evaluate a change to an instruction against a case set rather than by impression.

Indicators

Re-runs an existing case set after the change

Reports the effect on cases that previously passed

Records the version of the instruction that was evaluated

D3.L3.06Judgement

Advise others on interaction practice, giving reasons rather than recipes.

Indicators

Explains why a technique works and where it stops working

Corrects practice that succeeds for the wrong reason

Declines to prescribe where judgement is required

L4 Advanced

3 statements

D3.L4.01Practice

Establish organisational practice for instruction design, reuse and review.

Indicators

Defines where instructions are shared assets and where they are personal

Sets a review obligation tied to model and process change

Provides for practice that develops faster than the policy

D3.L4.02Judgement

Define where interaction is standardised and where practitioner judgement governs.

Indicators

States the criterion for standardising rather than standardising by default

Identifies where a standard instruction would cause harm

Records the decision so it can be challenged

D3.L4.03Practice

Establish how instruction assets are owned, versioned and retired.

Indicators

Assigns named ownership to each shared asset

Versions assets so that a result can be traced to an instruction

Retires assets rather than leaving them in circulation

Editorial notes

Known gaps, open questions and contested points

Published because the record is more useful than the appearance of completeness.

The template domain. Its shape, statement density and indicator style govern every domain drafted after it.

All eight

Domains index

Filtered view

Statements in D3

Proficiency

Levels

This page displays version 1.29 · last reviewed 30.08.2026