I have two ideas about how we’re setting up the evaluation.
First, the Bradley-Terry competition is mixing up what’s important for something to be shown with how new or unusual something is when it’s shown. Detail finds a problem, and three bots looking at pairs of changes also noticed a similar level in their scores, but people won't pay any extra attention to it in the actual running system. The engineer was going to see the problem from the bot’s review anyway. It would be good to specifically measure what Detail alone finds, versus what all the bots together find. We could look at that as a completely separate measure.
Second, when the same AI model is doing the judging, it’s being unfair. When I use a Claude to review code changes and a Codex to review the very same code changes as I’m developing, Codex finds issues that Claude is inclined to overlook. And Sonnet 4.6’s ratings of code, which come from a system run by Sonnet, also have some of that unfairness, even after the system summarizes the code.
> It would be good to specifically measure what Detail alone finds, versus what all the bots together find. We could look at that as a completely separate measure.
Yea, it does seem like running a combo of review bots is the right meta. It would be pretty interesting to see how Detail stacks up against 3 or 4 review bots all together (for some anec-data though - posthog has 4 review bots and we seem to find important stuff there).
> Second, when the same AI model is doing the judging, it’s being unfair.
Yup, agreed. We use a combo of models rn (not primarily Claude). The two review bots are Codex and Gemini reviewed by Claude.
The baseline Claude prompt being compared to feels pretty laughable so not sure what is learned. Maybe compare to a more realistic baseline for the DIY side for more compelling benchmarketing?
We started with a DIY code review skill because it's inherent to want to customize to our codebase and infra before trying solutions that add layers which may get in our way here. We have a 1 page skill that that does seperate passes on security, spec conformance, proper DRY & architectural abstractions, etc, and adversarial result quality passes to prune & prioritize. Others do similar.
Once quality is fixed, I'd expect the comparisons to be less on hits/misses , and more on token efficiency. That's a tricky one bc developer local review tokens are heavily subsidized right now.
> Maybe compare to a more realistic baseline for the DIY side for more compelling benchmarketing?
This is fair critique. However, I don’t really trust myself to write a great code review skill for vLLM or OpenClaw. I also don’t think Claude Code is the right harness for this deep and broad scanning work. We find that it struggles to maintain clarity when considering many different bugs at the same time. The coding agents seem really great at single-goal tasks that they can Ralph their way to.
> We started with a DIY code review skill because it's inherent to want to customize to our codebase and infra before trying solutions that add layers which may get in our way here.
Being able to tinker deeply with the tools is pretty inherent to my love of dev tools in general. Our job is to make use of all of those customizations (our agent will use that 1 page skill when doing its bug finding). I also still think externalizing part of your dev workflow is the right way to get ahead. You really don’t want to do the work of eval-ing/maintaining that skill to make sure it still performs well with a mythos or something.
> and more on token efficiency.
I’m really confident in our ability to stretch $20 of tokens ;)
> I also don’t think Claude Code is the right harness for this deep and broad scanning work. We find that it struggles to maintain clarity when considering many different bugs at the same time.
The skill we do splits it into multiple passes to divide and conquer on task dimension and on files for that exact reason. Likewise, it loops (Ralph-ish) until it converges. It maintains a task queue and work log to stay on track. We are growing it over time, but now more about per-repo customization, while the bones are good cross-repo.
I would only trust frontier-grade harnesses to do this kind of skill run, and guilty-until-proven-innocent various harnesses x prompt combos because of that.
My point isn't that our 1 page skill eliminates the need for your startup, but that is a normal flow for more serious ai-augmented coders so you are picking a blatantly known-bad starting point for serious coders. That makes it unclear what value your tool brings and calls into question why you are refusing to measure yourself in a post about measurement.
I have two ideas about how we’re setting up the evaluation.
First, the Bradley-Terry competition is mixing up what’s important for something to be shown with how new or unusual something is when it’s shown. Detail finds a problem, and three bots looking at pairs of changes also noticed a similar level in their scores, but people won't pay any extra attention to it in the actual running system. The engineer was going to see the problem from the bot’s review anyway. It would be good to specifically measure what Detail alone finds, versus what all the bots together find. We could look at that as a completely separate measure.
Second, when the same AI model is doing the judging, it’s being unfair. When I use a Claude to review code changes and a Codex to review the very same code changes as I’m developing, Codex finds issues that Claude is inclined to overlook. And Sonnet 4.6’s ratings of code, which come from a system run by Sonnet, also have some of that unfairness, even after the system summarizes the code.
> It would be good to specifically measure what Detail alone finds, versus what all the bots together find. We could look at that as a completely separate measure.
Yea, it does seem like running a combo of review bots is the right meta. It would be pretty interesting to see how Detail stacks up against 3 or 4 review bots all together (for some anec-data though - posthog has 4 review bots and we seem to find important stuff there).
> Second, when the same AI model is doing the judging, it’s being unfair.
Yup, agreed. We use a combo of models rn (not primarily Claude). The two review bots are Codex and Gemini reviewed by Claude.
The baseline Claude prompt being compared to feels pretty laughable so not sure what is learned. Maybe compare to a more realistic baseline for the DIY side for more compelling benchmarketing?
We started with a DIY code review skill because it's inherent to want to customize to our codebase and infra before trying solutions that add layers which may get in our way here. We have a 1 page skill that that does seperate passes on security, spec conformance, proper DRY & architectural abstractions, etc, and adversarial result quality passes to prune & prioritize. Others do similar.
Once quality is fixed, I'd expect the comparisons to be less on hits/misses , and more on token efficiency. That's a tricky one bc developer local review tokens are heavily subsidized right now.
> Maybe compare to a more realistic baseline for the DIY side for more compelling benchmarketing?
This is fair critique. However, I don’t really trust myself to write a great code review skill for vLLM or OpenClaw. I also don’t think Claude Code is the right harness for this deep and broad scanning work. We find that it struggles to maintain clarity when considering many different bugs at the same time. The coding agents seem really great at single-goal tasks that they can Ralph their way to.
> We started with a DIY code review skill because it's inherent to want to customize to our codebase and infra before trying solutions that add layers which may get in our way here.
Being able to tinker deeply with the tools is pretty inherent to my love of dev tools in general. Our job is to make use of all of those customizations (our agent will use that 1 page skill when doing its bug finding). I also still think externalizing part of your dev workflow is the right way to get ahead. You really don’t want to do the work of eval-ing/maintaining that skill to make sure it still performs well with a mythos or something.
> and more on token efficiency.
I’m really confident in our ability to stretch $20 of tokens ;)
> I also don’t think Claude Code is the right harness for this deep and broad scanning work. We find that it struggles to maintain clarity when considering many different bugs at the same time.
The skill we do splits it into multiple passes to divide and conquer on task dimension and on files for that exact reason. Likewise, it loops (Ralph-ish) until it converges. It maintains a task queue and work log to stay on track. We are growing it over time, but now more about per-repo customization, while the bones are good cross-repo.
I would only trust frontier-grade harnesses to do this kind of skill run, and guilty-until-proven-innocent various harnesses x prompt combos because of that.
My point isn't that our 1 page skill eliminates the need for your startup, but that is a normal flow for more serious ai-augmented coders so you are picking a blatantly known-bad starting point for serious coders. That makes it unclear what value your tool brings and calls into question why you are refusing to measure yourself in a post about measurement.
> That makes it unclear what value your tool brings and calls into question why you are refusing to measure yourself in a post about measurement.
Wait, definitely no refusal! Is there a public repo with a good skill we can measure against?
Author here!
Lmk if there is any qs I can answer about Detail or the post.
Is this something paid post? asking someone with high karma to post and making it to first page of HN news?
Nope, I’m not paying drob to post. Both of us work at Detail.