GPT-4 is the model I reach for when a problem genuinely requires multi-step reasoning — financial modeling, legal interpretation, complex debugging chains. The model's strength isn't just knowledge breadth; it's the ability to hold multiple reasoning threads simultaneously and synthesize them. But this strength is also a weakness: GPT-4 will reason about everything, including things you didn't ask it to. The sinc template's job is to focus that reasoning on the problem you actually have.
This example targets a financial analysis task where GPT-4's reasoning depth is essential. Note how the CONTEXT band gives the model full situational grounding so it doesn't reason about the wrong scenario:
{
"formula": "x(t) = Σ x(nT) · sinc((t - nT) / T)",
"T": "specification-axis",
"fragments": [
{
"n": 0,
"t": "PERSONA",
"x": "You are a CFO-level financial strategist who thinks in systems and second-order effects. You're not a banker — you're an operator. You've run P&L for businesses from $2M to $200M ARR. You communicate with precision and never hedge when the data is clear."
},
{
"n": 1,
"t": "CONTEXT",
"x": "A B2B SaaS company is considering two growth paths: (A) raise $8M Series A and hire aggressively, or (B) stay bootstrapped, grow at 80% YoY, and preserve optionality. Current state: $1.2M ARR, $40K MRR net new, 18% churn, 14-month runway, 8 employees."
},
{
"n": 2,
"t": "DATA",
"x": "Path A: $8M raise at $24M valuation (33% dilution). Hire 12 people in 6 months. Target: $4M ARR in 18 months. Path B: Current burn rate $85K/month. At 80% YoY: $2.16M ARR in 12 months, $3.88M in 24 months. Market: competitor raised $40M last quarter, growing faster."
},
{
"n": 3,
"t": "CONSTRAINTS",
"x": "Do not give me both sides without a recommendation. Pick a path and defend it. Show the second-order effects, not just the first-order math. Do not assume the fundraising closes — model the cost of failure. Avoid VC-speak. Speak as an operator, not an advisor."
},
{
"n": 4,
"t": "FORMAT",
"x": "Recommendation in the first sentence. Then 3 reasons, each with a specific number or concrete risk/outcome. Then 1 paragraph on the key decision variable that would change your recommendation. Max 350 words."
},
{
"n": 5,
"t": "TASK",
"x": "Make a clear, defensible recommendation on which growth path to take, with quantified reasoning."
}
]
}GPT-4's biggest strength is reasoning given complete situational context. When you give it a thin prompt, it fills in the gaps with assumptions — and those assumptions may not match your reality. The sinc template forces you to externalize all situational context into the CONTEXT and DATA bands before asking the model to reason. This turns GPT-4 from a general reasoner into a specific one.
In my testing, GPT-4 prompts with full CONTEXT and DATA bands produced recommendations with 3-4x more specific, verifiable claims than thin prompts. The model wasn't smarter — it had better inputs to reason from.
GPT-4-specific tip: GPT-4o responds well to "think step by step" embedded in the TASK band for complex multi-step problems. For simpler tasks, skip it — GPT-4 will over-reason. Save the chain-of-thought trigger for genuinely complex derivations.
Should I raise a Series A or stay bootstrapped? I have $1.2M ARR, 18% churn, 14 months runway. Competitor just raised $40M.
PERSONA: CFO-level operator, not advisor.
CONSTRAINTS: Pick a path. Show second-order effects. Model cost of fundraising failure.
FORMAT: Recommendation in first sentence. 3 numbered reasons with specific numbers. 350 words max.
The raw prompt produces a "here are the pros and cons of each path" non-answer. The sinc structure forces GPT-4 to commit to a recommendation and defend it with specific reasoning — which is what you actually need to make the decision.
Try AI Transform — Decompose Your Prompt Free