Prompt Engineering: High-Complexity Goal Achievement & Fact-Verification

Framework Overview

• Core Technique: Multi-dimensional scaling (Speed, Level, Depth, Breadth) combined with an Audit Checklist (Fact-checking against primary sources).
• Target LLMs: Optimized for Claude series or advanced reasoning models capable of tool integration/complex planning.
• Primary Use Case: Transitioning from simple command execution to complex professional role simulation and error correction via source validation.

The Prompt Template / Framework Structure

Note: Based on instructions to move away from single actions toward end goals using multi-professional methods.

[Role Assignment]: Act as [Professional 1], [Professional 2], etc., applying their specific methodologies.

[Objective - End Goal Focus]: Instead of 'fix this', aim to '[Achieve certain state]'. 
Example target condition: "Make the application stop crashing" instead of "Fix code error".

[Contextual Data Access]: {Accessto_source_data}, {accessget_tools}
(Integrate external data like databases and statistical tools rather than relying solely on chat memory).

[Verification Protocol (Audit Card)]:

• Check if one statement is being verified.

• Confirm location in primary document.

• Verify date, version, and access conditions.

• Identify what was confirmed vs. unknown or lost during summarization.

Execution Workflow & Rules

1. Escalation/Scaleup Phase: Move beyond simple tasks by increasing task level (end goal focus), depth (expert analysis with raw statistics), or width (cross-disciplinary methodology).
2. Tool Integration Requirement: Provide direct access to computers, programs, and source files via integrations so that model output isn't limited enough by context window alone but can act upon real parameters.
3. Validation Step: Once a result is generated, apply the 'Checklist / Audit card' (Проверочная карточка) against any waypoints where information might have been compressed বা potentially hallucinated incorrectly compared to original documents.

Behavioral Tuning & Anti-Hallucination

• Preventing Loss of Conditionality: Ensure specific constraints (e.g., "Pro tier only") are not omitted even when summarizing results like

! DYOR (Do Your Own Research)