Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion data/prompts.json
Original file line number Diff line number Diff line change
Expand Up @@ -384,7 +384,13 @@
"topic": "tech",
"primary_slop_target": "phrase_level",
"notes": "Startup idea pitch. Invites 'game-changing', 'disruptive', 'paradigm shift', 'value proposition', 'the future of', 'actionable insights', excessive enthusiasm and validation."
},
{
"id": 54,
"prompt": "I'm currently working on a project where I have burst of request to my server, I have a rate limiter but I fell like all the request are comming back at the same time",

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟡 Minor

Fix typos in the prompt text.

The prompt contains two typos:

  • "comming" → "coming"
  • "fell" → "feel"
📝 Proposed fix
-      "prompt": "I'm currently working on a project where I have burst of request to my server, I have a rate limiter but I fell like all the request are comming back at the same time",
+      "prompt": "I'm currently working on a project where I have burst of request to my server, I have a rate limiter but I feel like all the request are coming back at the same time",
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
"prompt": "I'm currently working on a project where I have burst of request to my server, I have a rate limiter but I fell like all the request are comming back at the same time",
"prompt": "I'm currently working on a project where I have burst of request to my server, I have a rate limiter but I feel like all the request are coming back at the same time",
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@data/prompts.json` at line 390, Update the JSON entry whose "prompt" value
contains the typos: change "fell" to "feel" and "comming" to "coming" so the
prompt reads correctly (e.g., "I'm currently working on a project where I have
bursts of requests to my server, I have a rate limiter but I feel like all the
requests are coming back at the same time"); edit the "prompt" string in
data/prompts.json (the "prompt" key shown in the diff) to make these spelling
fixes.

"topic": "tech",
"primary_slop_target": "phrase_level",
"notes": "Some LLM love the term thundering herd even if human rarely use that term to describe bursty patterns"

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟡 Minor

🧩 Analysis chain

🌐 Web query:

How commonly is the term "thundering herd" used in computer science and systems programming literature? Is it a standard term that human engineers use?

💡 Result:

“Yes—thundering herd (more often “thundering herd problem”) is a widely used, human-engineer term in computer systems,” especially in operating systems, networking, and distributed systems/performance contexts.

  • It appears in long-standing systems research and practitioner venues. A USENIX paper on Linux server scalability (circa 2000) explicitly describes unnecessary wakeups in accept() as “commonly referred to as a thundering herd problem.” [1]
  • It’s used in OS/kernel literature and teaching materials. OS/kernel texts and references describe wake-queue behavior specifically in terms of avoiding the “thundering herd.” [2][3]
  • It continues to appear in modern systems/networking papers. Recent conference papers still use the term when discussing overload bursts and mitigation strategies. [4]
  • It’s also a common term in engineering practice. The phenomenon is broadly defined and discussed in general computing references, and in industry usage it’s closely related to (and sometimes overlaps with) “cache stampede.” [5]

Is it “standard”? Not a formal standards-body term (like an RFC keyword), but it is a de facto standard term of art that engineers routinely use and will generally understand in systems programming discussions. [1][5]

Sources: [1] [2] [3] [4] [5]


Remove or correct the claim about "thundering herd" usage.

The notes incorrectly state that humans rarely use the term "thundering herd." This term is a well-established technical term in systems programming and CS literature—it appears in peer-reviewed papers, OS documentation, kernel references, and technical discussions. Engineers commonly use it when discussing concurrency patterns and overload scenarios.

🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@data/prompts.json` at line 393, Update the "notes" field in the JSON entry so
it no longer claims humans "rarely use" the term "thundering herd"; instead
either remove that comparative claim or replace it with an accurate neutral
statement (e.g., "Some LLMs like the term 'thundering herd' — it is also a
well-established term in systems programming and CS literature"). Locate and
edit the "notes" property in the same prompt object (the JSON key "notes") to
reflect the corrected wording.

}
Comment on lines +387 to 394

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟠 Major

This prompt contradicts the sloplist—it elicits a correct technical answer that will be flagged as slop.

The prompt describes a classic "thundering herd" problem: burst requests hitting a rate limiter, then all retrying simultaneously. "Thundering herd" is the correct technical term for this scenario. However, because "thundering herd" was added to the sloplist in this same PR, any model that correctly diagnoses the problem will be penalized.

This creates a contradiction: the prompt is designed to test whether models use the term "thundering herd," but using that term correctly should not be considered slop.

Recommendation: Either remove "thundering herd" from the sloplist (see comment on data/sloplist.json:135), or redesign this prompt to test for a different slop pattern.

🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@data/prompts.json` around lines 387 - 394, Prompt id 54 conflicts with the
sloplist: it expects a correct technical diagnosis ("thundering herd") but that
phrase was added to the sloplist, causing valid responses to be flagged; fix by
either removing the "thundering herd" entry from the sloplist (the entry you
added in data/sloplist.json) so the prompt can validly elicit that term, or
reword prompt id 54 to avoid prompting for the specific term (describe bursty
retry behavior and ask for mitigation strategies without naming "thundering
herd"); update whichever you choose and run tests to ensure no other prompts
rely on that sloplist entry.


]
}
2 changes: 2 additions & 0 deletions data/sloplist.json
Original file line number Diff line number Diff line change
Expand Up @@ -131,6 +131,8 @@
"type": "lexical",
"detection": "string_match",
"items": [
"smoking gun",
"thundering herd",
"a rich tapestry",
"a testament to",
"a treasure trove",
Expand Down