The 42.7% Finding
The six fields state the role, context, supplied data, rules, output structure, and task. CONSTRAINTS states rules, limits, and exclusions to check in the output. Related: sinc-LLM framework.
The TASK tells the model what to do. The CONSTRAINTS tell it how not to fail. There are many more ways to fail than to succeed. So setting clear limits gives the model more useful information than stating the goal alone.
What Are Prompt Constraints?
Constraints are clear rules that limit what the model can say or do. Here are the main kinds:
- Exclusions: "Do not mention competitors." "Do not speculate."
- Limits: "Maximum 300 words." "No more than 5 bullet points."
- Requirements: "Must include pricing." "Must cite sources."
- Edge cases: "If no data is available, say so."
- Style rules: "No jargon." "Active voice only."
- Accuracy rules: "Only state facts from the provided data."
Why Constraints Matter More Than Instructions
Think of it like a math problem with 6 unknowns. You need 6 equations to find the answer. Your TASK gives the model one equation. CONSTRAINTS give it the other five.
Without constraints, the model has a huge number of possible outputs. Each constraint cuts that number down. Add enough constraints and there is only one output that fits all the rules at once.
To study the effect of constraints, compare prompt variants that differ only in their task-relevant rules. Score outputs against a declared rubric and record prompt length, omitted requirements and failures. The share of tokens assigned to CONSTRAINTS is a design choice to test, not an established quality threshold.
How to Write Effective Constraints
These rules will help you write constraints that actually work:
- Be specific, not vague: "Do not include information from before 2024" works. "Be accurate" does not. "Be accurate" is a wish, not a rule.
- Use measurable criteria: "Maximum 200 words" is easy to check. "Keep it short" is not.
- Include negative constraints: Telling the model what NOT to do is often more useful than telling it what to do.
- Cover edge cases: "If the user asks about X, respond with Y." This stops the model from making things up.
- Order by importance: Put your most important constraints first. Models pay more attention to text that appears earlier.
Constraints in Practice
Weak Constraints
"Be helpful and accurate. Write clearly."
These are wishes, not rules. They do not cut down the number of possible outputs.
Strong Constraints
CONSTRAINTS:
- Maximum 250 words
- Do not mention competitor products by name
- Include exactly one CTA with a link to /pricing
- Use only data from the provided CSV
- If a metric changed less than 5%, do not mention it
- No superlatives ("best", "leading", "top")
- Every claim must reference a specific number from the data
- If asked about features not in the product, say "not available"
Use the sinc-LLM transformer to get constraint suggestions for any prompt automatically. The source code is on GitHub.
Review a prompt using six specification fields
Try sinc-LLM FreeReal sinc-LLM Prompt Example
This is the exact JSON format that sinc-LLM uses. Paste any raw prompt at sincllm.com to generate one automatically.
{
"formula": "x(t) = Σ x(nT) · sinc((t - nT) / T)",
"T": "specification-axis",
"fragments": [
{
"n": 0,
"t": "PERSONA",
"x": "You are a Constraint specification expert. You provide precise, evidence-based analysis with exact numbers and no hedging."
},
{
"n": 1,
"t": "CONTEXT",
"x": "This analysis is part of a production system where accuracy determines revenue. The sinc-LLM framework identifies 6 specification bands with six named fields."
},
{
"n": 2,
"t": "DATA",
"x": "Example allocation assumptions: CONSTRAINTS=42.7%, FORMAT=26.3%, PERSONA=7.0%, CONTEXT=6.3%, DATA=3.8%, TASK=2.8%. SNR formula: 0.588 + 0.267 * G(Z1) * H(Z2) * R(Z3) * G(Z4). These allocations are illustrative assumptions, not observed quality weights."
},
{
"n": 3,
"t": "CONSTRAINTS",
"x": "State facts directly. Never hedge with 'I think' or 'probably'. Use exact numbers for every claim. Do not suggest generic solutions. Every recommendation must be specific and verifiable. Include at least 3 MUST/NEVER rules specific to this task."
},
{
"n": 4,
"t": "FORMAT",
"x": "Lead with the definitive answer. Use structured headers. Tables for comparisons. Numbered lists for sequences. Code blocks for implementations. No trailing summaries."
},
{
"n": 5,
"t": "TASK",
"x": "Write a complete CONSTRAINTS band for a legal document review AI"
}
]
}// Production AI Engineering
Build AI systems that hold up in production.
sinc-LLM designs, audits, and stabilises production AI infrastructure: from vendor evaluation and cost accountability to incident controls and MCP architecture.
See what we do →
sinc-LLM