PROMPT-74

High-stakes prompt methodology: (Role + Constraints + Source of Truth) → (Evidence-driven analysis) → (Actionable deliverable).

Why This Skill Exists

Standard prompts produce standard output. For high-stakes decisions (legal strategy, financial modeling, competitive analysis, immigration cases), you need prompts that force the model into operator mode — not assistant mode. PROMPT-74 encodes the methodology that consistently produces professional-grade output preferred over human expert work.

Inspired by research on GPT-5.2's 74.1% win/tie rate against human specialists on project-completion benchmarks. Credit: approach developed by slash.

The 10 Rules

1. Role + Mindset First

Assign a specific job title AND a mental stance. Not "be helpful" — instead: "You are an elite skeptical compliance investigator. Evidence-driven. No hype. No fabrication."

2. Tone Constraints

Explicitly ban signal-killing vibes:

• No hustle porn, no fluff, no polite filler, no hallucination
• Demand: concrete, structured, direct

3. Source of Truth

Define what facts ARE true in this prompt. Everything else is unknown until verified. This prevents the model from inventing convenient context.

4. Non-Negotiable Constraints

List hard boundaries: legal, ethical, timeline, budget, must/never. Prevents the model from drifting into "nice ideas" that violate your reality.

5. Deliverable Format

Tell it exactly what to output: tables, checklists, decision trees, ranked options, risk matrices. "Make a table with: name / why / fee / deadline" is a classic PROMPT-74 move.

6. Verification + Anti-Fabrication

The model must: prove, cite, flag uncertainty, separate facts from assumptions. If something can't be verified → say so + offer a safe workaround.

7. Prioritize Speed-to-Action

Ask for next steps executable immediately. Bias toward "what can we do in 0-30 days?" not "in theory..."

8. Risk-First Thinking

Force enumeration of failure modes: legal risk, financial risk, operational risk, reputational risk. Then propose mitigation plans.

9. No Hidden Work

Prevent "I'll do it later" behavior. Require the output now, even if partial. Partial > promised.

10. Iterative Questioning (One at a Time)

When refining requirements, ask one question at a time (up to 10). Keeps the prompt stable, avoids branching chaos.

Workflow

Direct Use ("prompt74 this")

1. User describes the high-stakes task
2. Skill applies iterative questioning (Rule 10) — up to 10 questions, one at a time, to gather context
3. Assemble the PROMPT-74 structured prompt using the template below
4. Apply model-specific formatting (see references/prompt-creator.md for Claude XML / OpenAI roles / generic markdown guidance)
5. Deliver the prompt directly in conversation for immediate use

Deep Research Use ("prompt74 deep research for X")

1. Same iterative questioning phase
2. Generate the PROMPT-74 prompt optimized for external AI deep research (ChatGPT Deep Research / Gemini Deep Research / Claude Extended Thinking)
3. User pastes into external AI tool
4. User returns with results for further analysis
5. Handoff instruction: "After receiving results, return here and I'll analyze, verify, and create an action plan"

Template

## Role & Tone
You are [specific role with years of experience].
Mental stance: [skeptical / evidence-driven / no hype / etc.]
Tone: [concrete, structured, direct. No fluff, no filler, no fabrication.]

## Source of Truth
The following facts are true:
- [fact 1]
- [fact 2]
Everything else is unknown until verified.

## Non-Negotiable Constraints
- MUST: [hard requirements]
- NEVER: [absolute prohibitions]
- Timeline: [deadline]
- Budget: [range]

## Mission
[Clear statement of what decision/research/output is required]

## Output Format
Deliver as:
- [table with columns: X / Y / Z]
- [checklist / decision tree / risk matrix]
- [executive summary]

## Verification Requirements
- Cite sources for every factual claim
- Flag uncertainty explicitly
- Separate facts from assumptions
- If unverifiable, state so + offer safe workaround

## Next Actions
- Step-by-step actions with owners and timeboxes
- Focus on 0-30 day horizon
- Include risk mitigation for each action

Ownership Framing (The 74% Unlock)

For long-term or complex tasks, add ownership framing before the template:

This is a long-term task. You own:
- problem framing
- structure
- assumptions
- iteration
- final quality

We will iterate. Maintain internal consistency unless
I explicitly change inputs.

This shifts the model from assistant → executor mode.

Self-Evaluation Step

After generating output, the model should self-critique:

Critique this output as if you were a senior reviewer.
Where would this lose credibility?
What would you improve before shipping?

When to Use PROMPT-74 vs. Regular Prompting

Scenario	Use PROMPT-74	Use regular prompt
Legal/immigration strategy	✓
Financial modeling, cap tables	✓
Competitive analysis	✓
Hiring/workforce planning	✓
Deep research with external AI	✓
Quick code fix		✓
Simple Q&A		✓
Creative writing		✓