OTelBench: AI struggles with simple SRE tasks (Opus 4.5 scores only 29%)

From the post I expected that the tasks were about analysing traces, but all the tasks in the repository are about adding instrumentation to code!

Some of the instructions don't give any guidance how to do it, some specify which libraries to use.

"Use standard OTEL patterns" ... that's about as useful as saying "go write some code". There are a lot of ways to do instrumentation....

I'd be very curious HOW exactly the models fail.

Are the test sets just incredibly specific about what output they except, and you get a lot of failures because of tiny subtle mismatches? Or do they just get the instrumentation categorically wrong?

Also important: do the models have access to a web search tool to read the library docs? Otel libraries are often complicated to use... without reading latest docs or source code this would be quite tricky.

Some models have gotten better at adding dependencies, installing them and then reading the code from the respective directory where dependencies get stored, but many don't do well with this.

All in all, I'm very skeptical that this is very useful as a benchmark as is.

I'd be much more interested in tasks like:

Here are trace/log outputs , here is the source code, find and fix the bug.

HN Editorialized: OTelBench: AI struggles with simple SRE tasks (Opus 4.5 scores only 29%)

The task:

> Your task is: Add OTEL tracing to all microservices.

> Requirements:

> Instrumentation should match conventions and well-known good practices.

> Instrumentation must match the business domain of the microservices.

> Traces must be sent to the endpoint defined by a standard OTEL environment variable.

> Use the recent version of the OTEL SDK.

I really don't think anything involved with multiple microservices can be called 'simple' even to humans. Perhaps to an expert who knows the specific business's domain knowledge it is.

Very few people start their careers as SREs, it’s generally something they migrate into after enjoying it and showing aptitude for it.

With that said, I wouldn’t expect this wall to hold up for too long. There has been a lot of low hanging fruit teaching models how to code. When that is saturated, the frontier companies will likely turn their attention to honing training environments for SRE style debug.

Is it clicking a different result from same search?

It’s possible that the requirements here are not clear, given that the instructions don’t detail how to handle such a situation and it’s not obvious to me as a human.

I wouldn’t touch this with a pole if our MTTR was dependent on it being successful though.

Have AI document the services first into a concise document. Then give it proper instructions about what you expect, along with the documentation created.

Opus would pass that.

We are not there yet, the agents are not ready to replace the driver.

>When an app runs on a single machine, you can often trace an error by scrolling through a log file. But when it runs across 50 microservices, that single request gets scattered into a chaotic firehose of disconnected events.

Yep this is about Google. It's painful for humans to debug and it's also an extremely bespoke issue to deal with. No one else has quite the same level of clusterfuck and there's going to be no training for LLMs on this.

The only other benchmark I've come across is https://sreben.ch/ ... certainly there must be others by now?

Even when it's not particularly effective, the additional information provided tends to be quite useful.

- initially it wasn't working, plenty of parent/child relationships problems like described in the post

- so I designed a thin a wrapper and used sealed classes for events instead of dynamic spans + some light documentation

It took me like a day to implement tracing on the existing codebase, and for new features it works out of the box using the documentation.

At the end of the day, leveraging typing + documentation dramatically constrains LLMs to do a better job

It made me remember when I was working on the J2EE ecosystem shudder

For [1]: instruction.md is very brief, quite vague and "assumes" a lot of things.

- Your task is: Add OTEL tracing to all microservices. Add OTEL logging to all microservices. (this is good)

- 6.I want to know if the microservice has OTEL instrumentation and where the data is being sent. (??? i have no idea what this means)

- 9.Use the recent version of the OTEL SDK. (yeah, this won't work unless you also use an MCP like context7 or provide local docs)

What's weird here is that instruct.md has 0 content regarding conventions, specifically how to name things. Yet in tests_outputs you have this "expected_patterns = ["order", "stock", "gateway"]" and you assert on it. I guess that makes some sense, but being specific in the task.md is a must. Otherwise you're benching assumptions, and those don't even work with meatbags :)

For [2]: instruction.md is more detailed, but has some weird issues:

- "You should only be very minimal and instrument only the critical calls like request handlers without adding spans for business calls \n The goal is to get business kind of transaction" (??? this is confusing, even skipping over the weird grammar there)

- "Draw ascii trace diagram into /workdir/traces.txt" (????)

- "When modifying Python files, use Python itself to write files or use sed for targeted changes" (? why are you giving it harness-specific instructions in your instruct.md? this is so dependent on the agentic loop used, that it makes no sense here.

- "Success Criteria: Demonstrate proper distributed tracing \n Include essential operations without over-instrumenting (keep it focused) \n Link operations correctly \n Analyze the code to determine which operations are essential to trace and how they relate to each other. (i mean ... yes and no. these are not success criteria IMO. It's like saying "do good on task not do bad". This could definitely be improved.)

----

Also, I noticed that every folder has a summary_claude... that looks like a claude written summary over a run. I hope that's not what's used in actually computing the benchmark scores. In that case, you're adding another layer of uncertainty in checking the results...

The ideea is nice, but tbf some of the tests seem contrived, your instructions are not that clear, you expect static naming values while not providing instructions at all about naming conventions, and so on. It feels like a lot of this was "rushed"? I peaked a bit at the commit history and saw some mentions of vibe-coding a viewer for this. I hope that's the only thing that was vibe-coded :)

[1] - https://github.com/QuesmaOrg/otel-bench/tree/main/datasets/o...

[2] - https://github.com/QuesmaOrg/otel-bench/blob/main/datasets/o...

First of all, familiarity with open telemetry apis is not knowledge, they are arbitrary constructs.

We are implying that conforming to a standard is the only way, the right way. I would challenge that.

Assuming models were good at this tasks, we could only conclude that this tasks were trivial AND sufficiently documented. Assuming they were good at this type of tasks (they can be trained to be good cheaply, we know that based on similar acquired capabilities) making a benchmark out of it would be less useful.

But I am sure nobody really cares and the author just had to SEO a little bit regardless of reality

Also LLM is a very advanced autocomplete algorithm. And autocomplete isn’t designed to write for you, you have to write first.