Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AdShy reduced its reported average from 7.6 to 2.2 LLM checks per page by making the model a second-stage judge instead of the first filter. Version 0.1.23 capped checks at six, processed strong and visible signals first, delayed off-screen candidates, rejected several predictable false candidates locally, skipped error pages, and backed off on domains that repeatedly produced negative classifications. In a live run, the developer reached 2.2 checks per page across 34 reachable sites out of 40 tested, while recording fewer visible ad iframes and false positives in that run. Those figures come from the developer’s published release-note test, not an independent benchmark, so they show how the optimization worked in that sample rather than guaranteeing the same result on every site.
How did AdShy reduce LLM checks from 7.6 to 2.2 per page?
AdShy is a Chrome Manifest V3 extension that hides labelled advertisements with local rules. It can send an unlabelled, potentially promotional element to TypeSafe’s Jev classifier through OpenRouter when local evidence is inconclusive. In version 0.1.22, that fallback averaged 7.6 model checks per page, and the developer reported that a free allowance of 200 checks per day lasted only about 27 pages.
Version 0.1.23 changed the order and eligibility of those calls. It imposed a hard per-page cap of six checks, gave priority to strong signals and elements already on screen, postponed candidates outside the viewport, avoided requests for HTTP 400-or-higher pages, reduced future work on domains with a sustained run of negative results, and filtered several candidate types before they reached the model. In the reported live run, the average fell to 2.2 checks per page.
The arithmetic difference is 5.4 fewer checks per page. Treating the two published averages as comparable, that is approximately 71% fewer calls ((7.6 − 2.2) ÷ 7.6). The 71% figure is calculated from the reported values; it was not separately published as a benchmark percentage.
Popular on Amazon
- 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 counts as an “AI check per page” in this extension?
An AI check is one candidate evaluation sent to the Jev classifier, rather than a count of all advertisements on a page. A page can contain many elements that local rules identify immediately and therefore never send to the model. Conversely, a page with several ambiguous elements can consume multiple checks, subject to the version 0.1.23 limit.
The metric is an average over a set of pages. It does not mean every page makes exactly 2.2 network requests, and it does not describe tokens, latency, response size, or the number of ads hidden by local rules. Those details were not provided with the published averages, so the figure should be read as a call-volume measure only.
What changed in version 0.1.23?
The release combined six controls. Each one removes a predictable call, postpones an unnecessary call, or limits repeated calls after a domain has demonstrated that its candidates are usually not ads.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteA hard cap of six checks per page
Version 0.1.23 will not make more than six model checks for a page. This is an upper bound, not a target: local rules, pre-filters, page errors, and the domain backoff can all leave the actual count below six. A cap protects the daily quota and makes worst-case usage more predictable when a page contains a large number of ambiguous nodes.
Strong signals and visible blocks go first
The extension evaluates candidates with the strongest evidence first and prioritizes blocks that are already on screen. That ordering lets the user see obvious ads handled promptly while preserving the remaining budget for uncertain items that are currently relevant. It also prevents an off-screen candidate discovered early in page parsing from consuming a slot before a visible candidate is considered.
Lazy evaluation for off-screen candidates
Potential ads outside or far from the viewport wait until the reader scrolls near them. The model is therefore called when the candidate becomes useful, rather than when the document is initially scanned. If the reader never reaches that part of the page, the deferred candidate may never require a check.
Also popular on Amazon
- 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.
Skipping HTTP 400-or-higher pages
The extension makes no model checks for pages whose HTTP status is 400 or higher. This covers client and server error responses, where ad classification is unlikely to improve the page the user is trying to recover. The gate is deliberately based on the response status, not on a guess about whether the page looks broken.
Per-domain backoff after consistent negatives
When a domain produces at least eight “not an ad” answers and no “ad” answers during a 24-hour period, the extension lowers that domain’s per-page cap to two. This is a local confidence policy: a site that repeatedly supplies negative classifications receives fewer future opportunities to spend the model budget. A later positive result or a new time window can change the evidence, so the rule is not a permanent block on that domain.
Filtering candidates before inference
Four categories are removed before they can consume a model call:
- Same-site house promotions, which advertise the publisher’s own content or products.
- Sticky elements that have no other supporting ad signal.
- Iframes with no source or with a known embed pattern.
- Text links located in site navigation menus.
These are heuristics, not proof that every excluded element is harmless. Their purpose is to avoid spending an expensive, quota-limited check on patterns that are weak or routinely non-advertising.
What did the reported live run actually measure?
The developer tested 40 sites, of which 34 were reachable. The following results were reported for the live run comparing the earlier and newer behavior:
| Measure | Version 0.1.22 | Version 0.1.23 | How to read it |
|---|---|---|---|
| Average LLM checks per page | 7.6 | 2.2 | Developer-reported averages; the derived reduction is about 71%. |
| Visible ad iframes | 8 | 0 | Observed in this run, not a guarantee for all sites. |
| Manually recorded false positives | 1 | 0 | Manual count accompanying the same run. |
| Reachable sites | 34 of 40 tested | Six tested sites were not reachable during the run. | |
The table shows an association between the optimization and the reported outcomes, not a controlled accuracy study. The source does not publish the complete page corpus, a randomization procedure, confidence intervals, the exact model prompt version, token accounting, or an independent audit of missed ads and false positives. “Zero” in the table means zero observed in this manual run, not zero on the entire web.
Also popular on Amazon
- 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.
Did the reduction lower the extension’s estimated cost?
Yes, according to the developer’s estimates. Version 0.1.22 was reported at approximately $0.14–$0.18 per 1,000 page views, while version 0.1.23 was reported at approximately $0.04 per 1,000 page views. The assumptions behind those estimates are not fully described, so they should be used as directional figures rather than a universal price quote.
Those numbers are usage estimates tied to the reported call volume. A different classifier, prompt, token length, routing arrangement, or traffic mix could produce a different cost even with the same number of checks. The article does not provide enough token or billing detail to recalculate the estimates independently.
Why is selective inference more effective than sending every candidate to a model?
A browser extension can often make cheap, deterministic decisions from the page structure before it needs semantic judgment. Labelled ad markers, obvious screen placement, navigation context, iframe attributes, and the page’s HTTP status are all available without an LLM call. Sending only the ambiguous remainder to a classifier concentrates model capacity where a rule is least certain.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →The approach also aligns work with user attention. A candidate that is below the fold has no immediate visual impact, so delaying its evaluation avoids paying for an answer the reader may never need. Domain backoff adds a feedback loop: repeated negative classifications lower the future budget on that site instead of treating every page as a fresh, equally uncertain problem.
How can an extension implement the same call-saving pattern?
The following sequence captures the behavior described for version 0.1.23. Exact code and data structures are implementation choices, but the order matters because each earlier filter protects the later budget.
- Read the response status. If the page returned HTTP 400 or higher, stop classification for that page.
- Apply deterministic local rules. Hide elements carrying clear ad labels or other strong, non-semantic signals without contacting the model.
- Build a candidate queue. Record only ambiguous elements that survive local filtering, and mark whether each is visible or near the viewport.
- Remove weak structural candidates. Exclude same-site house promotions, sticky elements without another signal, srcless or known-embed iframes, and text links inside site menus.
- Order the queue. Put strong remaining signals and on-screen blocks before candidates that are off-screen.
- Apply the domain budget. Use a normal maximum of six checks unless the domain has reached at least eight negative answers and zero positive answers in the preceding 24 hours; in that case, use a maximum of two.
- Classify visible candidates. Send candidates to Jev through OpenRouter until the active cap is reached, recording each positive or negative answer for the domain history.
- Defer the rest. Keep off-screen candidates pending and trigger evaluation only when scrolling brings them near the viewport, while still respecting the page and domain caps.
This design should expose counters for page checks, deferred candidates, local exclusions, and domain-backoff decisions. Without those counters, a lower average can be difficult to diagnose: it may reflect fewer candidates, more local matches, unreachable pages, or actual improvements in prioritization.
Also popular on Amazon
- 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
What implementation pitfall did the release fix?
The release notes describe a CSS specificity conflict. A site’s utility classes using !important could override the extension’s hiding style. The fix used an ID-weight selector in the extension stylesheet, increasing specificity so the intended ad-hiding rule could win in that situation.
This matters because call reduction and visual enforcement are separate problems. A classifier can correctly identify an element, yet the element may remain visible if page CSS overrides the extension’s rule. Testing should therefore check both the number of model calls and whether an identified element is actually hidden.
Is the 2.2-check result reproducible across the web?
It is a documented result from a 34-reachable-site live run, not a reproducibility claim for all websites. Six of the 40 tested sites were unreachable, and the published account does not specify how pages were selected, how many pages each site contributed, whether the two versions saw identical content, or whether model responses were sampled under identical prompts and routing conditions.
For a stronger comparison, an evaluator would need the same URL set, identical page states, the same Jev model and prompt, fixed extension builds, and complete logs of candidates, answers, tokens, and skipped pages. It should also measure missed ads, not only visible iframes and manually noticed false positives. Until those details are available, 2.2 is best treated as an encouraging engineering benchmark with clear scope limits.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What was available to users when the change was reported?
At the time of the report, version 0.1.23 was still under Chrome Web Store review, while the store continued to serve version 0.1.22. That means a reader could encounter the older 7.6-check behavior even after reading about the optimization. Store approval and propagation are separate from the code change itself, so the installed version should be checked before comparing local results with the published figures.
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 minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11The extension’s model-assisted path is optional; its local rules remain the first line of processing. Users and developers should also account for the reported 200-check daily free quota when estimating how many pages an older build can process.
Also popular on Amazon
- 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.
What should developers take away from the AdShy change?
- Use deterministic, inexpensive rules before probabilistic inference.
- Spend the model budget on visible or high-value ambiguity first.
- Do not classify work that the user may never see.
- Stop on page states where classification has little practical value, such as HTTP error responses.
- Learn from domain-level outcomes, but make backoff reversible and time-bounded.
- Measure call volume, user-visible blocking, false positives, and CSS enforcement separately.
- Publish the test scope and evidence limits alongside any percentage improvement.
AdShy’s result illustrates a general systems principle: reducing unnecessary inference can improve cost and responsiveness without replacing the classifier. The strength of the result lies in the layered policy—local rules, prioritization, lazy work, and feedback—not in any single six-call limit.
FAQ
Does a six-check cap mean every page gets six model decisions?
No. Six is only the maximum for a normal page in version 0.1.23. Local matches, pre-filtered candidates, HTTP error gating, and the two-check domain backoff can all reduce the number further, while deferred candidates may be evaluated later only if the reader scrolls to them.
Can the reported dollar figures be used as a quote for another extension?
No. They are approximate estimates attached to this implementation’s reported traffic and model usage. Because the publication does not provide complete token accounting or billing assumptions, another extension should calculate its own cost from its provider rates, prompt sizes, routing, and observed call mix.
Recommended Free Tools
What does “zero visible ad iframes” prove?
It shows that none were visible in the reported run after the change. It does not establish that every ad format was detected, that hidden or personalized ads were absent, or that the same outcome will occur on sites outside the tested sample.
Why might the Chrome Web Store still show the old behavior?
The newer build was reported as being in review while the store served version 0.1.22. Review and publication can therefore lag behind a developer’s release notes; the installed extension version is the relevant one when checking whether the six-check cap and other changes are present.
Frequently Asked Questions
Does a six-check cap mean every page gets six model decisions?
No. Six is only the maximum for a normal page in version 0.1.23. Local matches, pre-filtered candidates, HTTP error gating, and the two-check domain backoff can all reduce the number further, while deferred candidates may be evaluated later only if the reader scrolls to them.
Can the reported dollar figures be used as a quote for another extension?
No. They are approximate estimates attached to this implementation’s reported traffic and model usage. Because the publication does not provide complete token accounting or billing assumptions, another extension should calculate its own cost from its provider rates, prompt sizes, routing, and observed call mix.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →What does “zero visible ad iframes” prove?
It shows that none were visible in the reported run after the change. It does not establish that every ad format was detected, that hidden or personalized ads were absent, or that the same outcome will occur on sites outside the tested sample.
Why might the Chrome Web Store still show the old behavior?
The newer build was reported as being in review while the store served version 0.1.22. Review and publication can therefore lag behind a developer’s release notes; the installed extension version is the relevant one when checking whether the six-check cap and other changes are present.
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.
