Prompt Lab Engine

🔍 SECTION 1: AiGanak® Gap Analysis & Root Cause Protocol

• Deterministic Alignment Score: 25%
• Instructional Weight Deconstruction:
  • • Attention Dispersion: The model acknowledged deployment latency data but completely ignored discount vectors and tone requirements.
  • • Context Bias: Overfocused on apology text string, underweighted “professional yet warm” parameters and missing field structures.
  • • Competing Priorities: The tracking update promise and promo discount code competed directly with the simple text apology; the executing model defaulted to the lowest-token-effort execution path.
• Root Cause Engineering Diagnostics:
  • • Semantic Drift: Deviated significantly from multi-constraint baseline target.
  • • Format Regression: Completely dropped required transactional elements (discount validation rules).
  • • Instructional Dissipation: The system instruction block was buried too deep inside un-tagged strings, resulting in model under-attendance.
  • • Boundary Breach: Failed to maintain professional warmth boundary, defaulting to dry prose.
  • • Context Contamination: Carried only the baseline apology context forward into the completion array.

📊 SECTION 2: Quantitative Performance Metrics (Estimated)

• Logic Depth Expansion:
  • • Introduced gated attention nodes to enforce constraint weighting (apology, discount, tone).
  • • Deployed explicit instruction segmentation to prevent drift (semantic anchors per rule).
  • • Embedded boundary-violation detectors to reject outputs missing mandatory elements.

⚡ SECTION 3: The Optimized Production Architecture (The Fix)

<prompt>
<system_intent>
  Provide a sympathetic, professional customer support reply for a late order.
</system_intent>
<rules>
  <rule id="R1">Start with a warm apology acknowledging frustration.</rule>
  <rule id="R2">Offer discount code SORRY10 for 10% off next order.</rule>
  <rule id="R3">Promise immediate tracking update.</rule>
  <rule id="R4">Use professional yet warm tone; avoid slang.</rule>
</rules>
<variables>
  <user_issue>{{order_delay_details}}</user_issue>
</variables>

<chain_of_thought>
<!-- INTERNAL_CoT:
1. Parse user emotion → angry, frustrated.
2. Map to apology template.
3. Insert discount code per R2.
4. Append tracking commitment per R3.
5. Apply tone filter per R4.
-->
</chain_of_thought>
<response_template>
<![CDATA[
I completely understand your frustration, and I sincerely apologize for the delay. Your business means a lot to us.
As a gesture of goodwill, please use code SORRY10 for 10% off your next order.
I’m tracking your shipment right now and will ensure it reaches you ASAP.
]]>
</response_template>
<negative_constraints>
  <must_not>Omit discount code</must_not>
  <must_not>Omit tracking promise</must_not>
  <must_not>Use casual/unprofessional language</must_not>
</negative_constraints>
</prompt>