Rule reference · Clarity

Negative-only constraints

  • negative_only_constraints
  • info in this example
  • Clarity
  • stage · Instruction quality

Fires when: 3+ prohibitions and no positive requirement.

Prohibitions narrow the space of answers without pointing to a good one. The model avoids the banned behaviors and fills the gap with guesses.

Keep the prohibitions that matter, and add the positive behavior next to each one: instead of refunds, offer a replacement; instead of dates, say when the customer will hear back.

Example that triggers it

Reply to the customer message below. Do not mention refunds. Never promise a delivery date. Don't use technical jargon.

<message>
{{message}}
</message>

info 3 prohibitions and no positive requirement.

Why it matters. A list of "don't"s says what to avoid but not what good output looks like.

Fix. For each prohibition, add what the model should do instead.

Quality 84/100 · 32 tokens · 3 findings

Fixed version

Reply to the customer message below in under 120 words.

- Offer a replacement instead of a refund.
- Say that our team will confirm the delivery date by email within 2 business days.
- Use plain language that a non-technical reader understands.

<message>
{{message}}
</message>

Quality 90/100 (+6) · 60 tokens · 1 findings

Resolved by the fix

Scores, token counts (GPT-4.1 tokenizer), and findings on this page are computed by the engine when the site is built.

How it affects the score

Each finding subtracts a fixed penalty from 100: critical 30, warning 10, info 3. This rule counts against the Clarity dimension. See the scoring model for the full formula.

All 20 rules →