Rule reference · Consistency
Near-duplicate instructions
Fires when: Two sentences with ≥ 80% word overlap.
Near-duplicates appear when prompts are assembled from pieces or edited by several people. Small wording differences ("the customer's name" and "the customer name") can make the model think they are two separate requirements.
Merge them into one sentence with the most precise wording.
Example that triggers it
Draft a reply to the support ticket below. Include the customer's name, order number, and delivery date in the reply. Include the customer name, order number and delivery date in your reply. Return plain text.
<ticket>
{{ticket}}
</ticket>info Two sentences say nearly the same thing.
Why it matters. Near-duplicates waste tokens, and small wording differences between them can read as conflicting requirements.
Fix. Merge them into one sentence.
Quality 97/100 · 51 tokens · 1 findings
Fixed version
Draft a reply to the support ticket below. Include the customer's name, order number, and delivery date. Return plain text.
<ticket>
{{ticket}}
</ticket>Quality 100/100 (+3) · 34 tokens · 0 findings
Resolved by the fix
near_duplicate_instruction: Two sentences say nearly the same thing.
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 Consistency dimension. See the scoring model for the full formula.
Other consistency rules
- Contradictory instructions: Conflicting formats, lengths, word limits, tools, or languages.