What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The larger cloud model produced the clearest positive signal, but none of the three setups earned a prompt promotion. A local Qwen3 4B model finished where it started at 60.0%; the 24B cloud model produced a 64.0% to 68.0% held-out artifact change that failed the statistical gate; and the role-separated runner generated zero proposals. The result is not that model size is irrelevant. It is that useful execution, promising edits and trustworthy promotion are three different outcomes.
What the experiment actually compared
The AgentSelfEdit v0.3.0 field test compared three versions of the same self-edit loop. Each loop used a model to execute tasks, inspect failures, propose a prompt change and evaluate whether that change deserved promotion.
As an Amazon Associate I earn from qualifying purchases.
- Local setup: Qwen3-4B-Instruct-2507-4bit.
- Cloud setup: mistralai/mistral-small-3.2-24b-instruct.
- Role-separated setup: an executor, analyzer and judge that could be assigned different models. In this run, the executor was qwen/qwen3-30b-a3b-instruct-2507 and the analyzer was mistralai/mistral-small-3.2-24b-instruct.
The evaluation used a 40-task promotion corpus and a 25-task held-out set. A candidate needed both a p-value below 0.05 and a minimum effect size of at least 5% to pass the promotion gate. Those requirements matter: a higher final score in one artifact is not, by itself, evidence that the system should ship a new prompt.
Results at a glance
| Setup | Baseline and final result | Iterations | Proposal or promotion outcome | What the run showed |
|---|---|---|---|---|
| Local Qwen3 4B | 60.0% to 60.0% | 5 | 0 promotions | Stable operation, but mostly null or cancelling edits |
| Cloud Mistral 24B | 64.0% baseline; 68.0% final held-out artifact | 8 | 0 promotions | First clean positive movement, below the statistical gate |
| Role-separated Qwen 30B executor plus Mistral analyzer | 60.0% baseline; no candidate artifact was produced | 3 | 0 proposals | Runner and baseline were sane, but candidate generation never began |
These are different failure modes. The local loop searched without finding net improvement. The cloud loop found a potentially useful direction but could not establish it with the required evidence. The role-separated loop stopped before there was an edit to compare.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
What happened with the local 4B model
The local Qwen3-4B-Instruct-2507-4bit began at 60.0% accuracy and ended at 60.0% after five iterations. It promoted nothing.
Its iteration trace explains why the unchanged score is informative rather than empty. Iterations 1, 2 and 4 recorded zero improvements and zero regressions. Iterations 3 and 5 each recorded one improvement and one regression. In other words, the loop occasionally moved an individual task in the right direction, but the gains were exactly cancelled by losses elsewhere.
Why the local model was still useful
The local model was described as cheap and stable, making it useful for executor or baseline work. A model does not need to improve the prompt to be operationally valuable. Predictable execution can provide a repeatable starting point, keep an experiment affordable and make failures easier to reproduce.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why it was weak as an analyzer
The analyzer repeatedly returned null edits or edits that cancelled one another. That behavior limits the search space: the system spends iterations confirming the current prompt or making changes too small to produce a net result. The 60.0% final score therefore should not be read as proof that the prompt was already optimal. It shows only that this analyzer, under this loop and corpus, did not find a change that survived evaluation.
What the 24B cloud model changed—and what it did not
The cloud run used mistralai/mistral-small-3.2-24b-instruct. Its baseline was 64.0%, and the reported final held-out artifact reached 68.0%. Across eight iterations, it produced the strongest apparent movement in the experiment, but it still had zero promotions.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
The best A/B iteration
The best recorded comparison improved two tasks and regressed none. Its mean delta was 0.05, its effect size was 0.0625 and its p-value was 0.79. The effect-size figure cleared the stated minimum of 5%, but the p-value did not come close to the required threshold of 0.05.
That combination is the central lesson. A change can look directionally good and still fail the evidence standard. The 64.0% to 68.0% artifact movement is therefore an evaluation result, not a deployed improvement. No prompt was promoted, so production behavior remained based on the existing prompt rather than the candidate.
Why “better model” did not mean “trusted edit”
The larger model generated the first clean positive signal, which suggests that model capability affected the quality of analysis or proposal generation. But trust requires more than a clean-looking direction. The candidate had to clear both the effect-size floor and the statistical gate on the designated comparison. It did not.
Calling the run a success because its final artifact was 68.0% would collapse three separate stages into one: generating a candidate, observing a score and approving a change. The experiment kept those stages separate, which is why the headline result is “positive signal without promotion,” not “the prompt improved by four points.”
Why role separation produced zero proposals
The role-separated configuration used a qwen/qwen3-30b-a3b-instruct-2507 executor and the 24B Mistral model as analyzer. It ran for three iterations, with a held-out sample of five and a promotion sample of ten.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
The baseline was sane at 60.0%, and the run directories and analysis.json files existed. However, prompt-b.md and ab-comparison.json never appeared. The run did not reach candidate generation or comparison, so its outcome was zero proposals rather than bad proposals.
Zero proposals is a systems result
“No proposal” is not interchangeable with “the proposed prompt failed.” In this case, there was no candidate prompt to score and no A/B comparison to test. The useful diagnostic question is where the pipeline stopped: execution and analysis artifacts were present, but the handoff into candidate generation was not completed.
The executor shapes the analyzer’s evidence
The experiment’s mechanism is failure-trace dependent. As author Debashish Ghosal put it, “The analyzer does not work on abstract task labels. It works on failure traces.” An analyzer sees the concrete errors, omissions and patterns produced by the executor. Change the executor, and you change the evidence available for analysis.
That means replacing one model while leaving the rest of the loop unchanged can alter more than execution quality. It can change whether the analyzer sees a repeated, stable failure that can be turned into a prompt rule. In the separated run, the executor’s output did not yield a candidate proposal, even though the analyzer itself was the stronger cloud model used in the single-model run.
What kinds of edits the loop was actually finding
Across the runs, proposal behavior remained narrow. The common directions were urgency-boundary rewrites, local wording clarifications and tight rule changes. Structural prompt revisions, broad task-decomposition changes and genuinely diverse proposal families were uncommon.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallRank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
This matters when interpreting a no-promotion benchmark. A loop that only searches small wording adjustments may be good at polishing a local edge case but poorly equipped to discover a change in task structure. Conversely, a loop that makes broad structural edits could generate more dramatic candidates while also increasing regression risk. The reported runs do not establish that one strategy is universally better; they show that the search behavior was concentrated in a small family of edits.
How to interpret a no-promotion benchmark
A zero-promotion result is not a single verdict. Read it by asking which stage failed and what evidence exists.
- Check execution. Did the runner complete its tasks and produce the expected logs? The role-separated run did produce run directories and
analysis.jsonfiles. - Check proposal generation. Was a candidate prompt written? In the role-separated run,
prompt-b.mdwas absent, so there was nothing to evaluate. - Check the A/B comparison. Did the candidate produce an improvement, regressions or a tie? The local run had balanced one-improvement/one-regression iterations, while the cloud run’s best comparison had two improvements and zero regressions.
- Check the gate. Did the result meet both requirements? The cloud iteration had an effect size of 0.0625 but a
p-value of 0.79, so it failed the gate. - Check deployment status. Was the candidate actually promoted? None of the setups promoted a prompt. A held-out artifact score must not be reported as shipped behavior when promotion never occurred.
This classification prevents a misleading aggregate such as “the agent failed.” The local analyzer searched but did not produce net gain; the cloud analyzer found a promising but statistically untrusted direction; and the separated runner failed to produce a candidate at all.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the comparison says about model size, cost and stability
The local 4B model had the clearest operational advantage in the report: it was cheap and stable. Its limitation was analytical search, not basic usability. The 24B cloud model generated a stronger signal, but the experiment did not establish a trusted improvement and therefore did not justify treating the additional capability as a successful deployment.
The role-separated setup shows why cost or parameter count alone cannot predict the outcome. It combined a larger executor with the same 24B analyzer and still produced zero proposals. Separating roles changes the information flow between components; it is not simply a way to attach a stronger model to a fixed pipeline.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
A practical decision framework for similar experiments
Use a local model when repeatability is the priority
A local 4B-class model can make sense for baseline execution, mechanics and inexpensive iteration. Treat its analyzer output conservatively when it produces mostly null edits or balanced improvements and regressions. Its value may be infrastructure reliability rather than prompt discovery.
Use a larger cloud model to search for signal, not automatic approval
A stronger model is useful when the goal is to broaden analysis and generate cleaner candidate directions. Keep the promotion gate unchanged. A candidate that improves a small comparison but fails the required p-value is evidence for another experiment, not a reason to replace the production prompt.
Separate roles only when the handoff is observable
If executor and analyzer models differ, log the failure traces passed between them and verify that candidate-generation artifacts are created. A role-separated design can expose useful specialization, but it can also change the evidence distribution enough to suppress proposals entirely.
Recommended Free Tools
How to report these results without overstating them
- Report the baseline, final artifact and number of iterations together.
- Distinguish a candidate artifact from a promoted prompt.
- State improvements and regressions, not only the final percentage.
- Include the effect size and
p-value when discussing a gate. - Call out “zero proposals” separately from “proposals that failed evaluation.”
- Describe which model performed execution and which performed analysis in a role-separated run.
The most accurate summary is also the least sensational: the local model was useful but did not improve net accuracy; the larger cloud model produced the first clean positive movement but failed the promotion criterion; and role separation changed the evidence path enough that no proposal was generated.
As Ghosal’s formulation captures it, “Better is not the same thing as good enough to trust.” That distinction is the reason a no-promotion benchmark can still be valuable: it reveals whether the bottleneck is execution, analysis, candidate generation or statistical confidence.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
