Adaptive PNG filtering backfired, inflating 7 of 8 test animations by up to 2.4x

A developer adding adaptive per-scanline filtering to an APNG encoder discovered the feature made seven out of eight test animations larger, with one cicada-molting clip ballooning to 2.41 times its original size. Testing all five PNG filters applied uniformly across every scanline showed that each fixed filter also performed worse than no filtering on the same seven files. The root cause lies in a conflict between how the minimum-sum-of-absolute-differences heuristic selects filters and how DEFLATE compression actually rewards byte repetition. Because the heuristic optimises for low byte magnitude rather than compressibility, it routinely picks filters that break up long runs of identical bytes, hurting DEFLATE. The author concluded that untested optimisation assumptions are effectively guesses, and that a single controlled experiment — forcing one filter constant — revealed what three correlational metrics had failed to show.

Nigerian Teen Built a Ranked Game Engine on Borrowed Laptops and a Phone

Kehinde Owolabi, an 18-year-old developer from Lagos, Nigeria, built a JavaScript game engine called Limn Engine that scored 94 out of 100 and ranked third among 2D JavaScript game engines on SitePoint. He developed the engine using limited hardware, including a Chromebook, a borrowed ThinkPad, and a Toshiba laptop, without access to consistent internet or dedicated equipment. Owolabi's interest in technology began with mobile animation app FlipaClip and a BASIC tutorial in a school textbook, which he studied before he even owned a device. He later taught himself Python using his sister's phone after falling behind in physics class. Limn Engine runs at 60 frames per second on low-end devices such as a Tecno Pop 4 with just 1GB of RAM.

Bytesmith offers plug-and-play JVM compiler toolkit for Android and desktop apps

Bytesmith is a Kotlin and Java compiler toolkit built for JVM and Android applications, designed to simplify adding compilation capabilities without building a full pipeline from scratch. It provides a unified API supporting Kotlin, Java, and mixed-language compilation, along with filesystem abstraction, custom classpaths, compiler plugins, and diagnostics. A key feature is its filesystem abstraction layer, which lets the same compilation code work across traditional JVM file paths and Android's Storage Access Framework URIs. Java compilation is handled via ECJ integration, removing the need for developers to implement it themselves. Available via Gradle, Bytesmith requires only a one-time filesystem configuration before handling all underlying compiler orchestration.

LeetCode 31: How to Find the Next Permutation in O(n) Time

LeetCode problem 31 asks developers to rearrange an array into its next lexicographically greater permutation. The approach involves scanning from right to left to find the first element smaller than its right neighbor, then finding the first element greater than it to swap. After the swap, the suffix following the swapped index is reversed to produce the smallest possible arrangement. A corner case is handled when the array is in fully descending order, in which case the entire array is simply reversed. The solution runs in O(n) time with O(1) auxiliary space.

ToxicPanda 2.0 Android Malware Uses VPN and ADB to Steal Banking Credentials

Zimperium zLabs published research on August 19, 2026, detailing ToxicPanda 2.0, an advanced Android banking malware distributed via AWS-hosted storage. The malware tricks users into granting VPN permissions, which it then uses to block Google Play and Google Play Services network traffic. It subsequently exploits Android's Accessibility Service to autonomously navigate device settings, enable Wireless Debugging, and pair with the local ADB daemon to gain shell-level privileges without rooting the device. Once active, the payload deploys fake overlay screens to harvest banking credentials, device PINs, and other sensitive data, while maintaining a persistent encrypted WebSocket connection to a command-and-control server. The malware supports 167 remote commands, enabling attackers to take screenshots, intercept SMS messages, simulate user input, and install additional APKs on compromised devices.

MP Teachers Cross Flooded River on Tubes to Meet E-Attendance Deadline

Teachers in Bhind district, Madhya Pradesh, are crossing the swollen Baisali river on inflated tubes to reach their schools on time. The risky daily crossings are driven by the state government's mandatory e-attendance application, which requires teachers to mark their presence digitally within a set timeframe. Heavy monsoon rains have made the river difficult to cross through conventional means, leaving the tube crossings as the only viable option. The situation highlights the challenges faced by educators in flood-prone rural areas while trying to comply with digital accountability measures.

How to Build a HIPAA-Compliant Healthcare Document Pipeline on .NET and Azure

A technical guide published on DEV Community outlines how to architect a production-grade healthcare document processing pipeline using .NET and Azure services. The pipeline is designed to handle tens of thousands of monthly hospital documents—including discharge summaries, lab reports, and imaging PDFs—converting them into FHIR-compliant, structured data. Key design choices covered include using Azure Document Intelligence for OCR, Azure Cognitive Search for HIPAA-compliant vector search, and Azure Container Apps for deterministic scaling over serverless options. The guide warns that poor orchestration, aggressive batching, and unversioned embeddings are common production pitfalls that can compromise both compliance and performance. According to the author, proper orchestration—not the AI model itself—represents the largest cost driver, but the right service mix can cut latency by 30–50% while keeping infrastructure costs under 10% of raw compute spend.

Harbhajan Singh Responds After Police Confirm Punjab Drug Video Is Authentic

Former Indian cricketer Harbhajan Singh took a dig at critics after Punjab Police confirmed the authenticity of a drug-related video purportedly shot in Punjab. Singh had earlier shared a post by BJP leader Ravneet Singh Bittu on X referencing the video. His response appeared aimed at those who had questioned or disputed the video's origins. Singh sarcastically remarked that 'paid trolls' would next claim the Jagraon bridge was located in Rajasthan, implying attempts to deflect blame from Punjab.

China recalls nearly 3 million Teslas over hidden door handle safety concern

Chinese authorities have ordered the recall of nearly 3 million Tesla vehicles over concerns related to hidden door handles. Several Chinese automakers, including XPeng, Xiaomi, and Geely, are also affected by the recall. The issue raises safety concerns, as concealed door handles may pose risks to passengers, particularly in emergencies. The recall represents one of the largest automotive safety actions in China involving both foreign and domestic manufacturers.

Rishabh Pant Retires Hurt After Blow to Left Arm in Colombo Test

Rishabh Pant was forced to retire hurt on Day 1 of the second Test between India and Sri Lanka in Colombo on Sunday. The wicketkeeper-batter was struck on his left arm by a delivery during India's innings. His participation on Day 2 came into question following the injury. India's head coach addressed the media to provide an update on Pant's availability for the remainder of the match.

52 hospitalised after chlorine gas leak at Nagpur's Godhani water plant

A chlorine gas leak at the Godhani water treatment plant in Nagpur sent 52 people to hospital with respiratory complaints. Twelve of those affected were discharged after receiving treatment. The leak was traced to a faulty, outdated chlorine tank at the facility. Emergency teams from OCW, NDRF, and NMC were swiftly deployed to contain the situation. Authorities have launched an investigation into the cause of the incident.

Nvidia discusses Perplexity investment at $30B

7 Signs Your AI App Is Over-Engineered and How to Fix It

A widely shared piece on DEV Community warns that most AI applications fail not due to poor model selection but because unnecessary complexity is added before core problems are clearly defined. Common over-engineering pitfalls include reflexively setting up vector databases before confirming a retrieval problem exists, and building multi-agent systems for tasks a single well-structured prompt could handle. Fine-tuning models on proprietary factual data is flagged as another costly mistake, since models memorize facts unreliably and the approach is often confused with retrieval. The article argues that simpler alternatives — keyword search, linear prompt chains, and in-context document injection — frequently outperform elaborate architectures. Developers are advised to build the most basic working version first and add complexity only when a specific, measurable limitation demands it.

A plot twist awaits Paramount-Warner Bros merger

Larameter: New Laravel Package to Manage Credits, Plans and Usage Quotas

A developer has released Larameter, an open-source Laravel package designed to handle credit metering, subscription plans, and usage quotas within web applications. The package addresses a common problem where billing logic, usage tracking, and plan limits drift out of sync as apps grow more complex. Larameter supports multiple time windows — such as session, weekly, and monthly caps — derived from a single plan-level credit figure, so updating one number automatically adjusts all limits. It works by adding a trait to any billable model, such as a user or organisation, without requiring extra database columns. The package can be installed via Composer and is modelled on metering systems seen in platforms like Claude and OpenAI.

Chinese robots outpace Usain Bolt's 100m world record at robot games

Chinese humanoid robot Tiangong Ultra completed a 100-meter sprint in 9.39 seconds, beating Usain Bolt's human world record. A rival robot named Lightning also clocked a time faster than Bolt's benchmark. The achievement marks a significant leap in performance since the inaugural robot games held last year. China has designated humanoid robotics as a strategic industry, signaling strong government backing for the sector. Global participation in the robot games has grown considerably in the latest edition.

Developer Auto-Generates 50+ SEO Pages at Build Time for Static Calculator Site

A developer behind TextTimeTools, a free static site offering speaking and reading time calculators, built a script-based system to automatically generate over 50 keyword-targeted landing pages at build time. The site runs on Cloudflare Pages with no backend or CMS, yet targets long-tail search queries such as 'how many words is a 5 minute speech' and 'how long to read 1000 words' through individually generated HTML pages. A Node.js build script reads a list of keyword targets, computes accurate answers from a shared speed table, and outputs complete pages with structured FAQ schema markup for each query. The build pipeline runs four sequential scripts covering page generation, site assembly, and automatic sitemap creation based on actual output files rather than a manually maintained URL list. The developer credits this automated approach as the single highest-impact strategy for growing the site's organic search traffic, eliminating the need to hand-write or manually maintain individual pages.

Skipping Cheap AI Preview Steps Can Cut Costly Errors in Generative Video Pipelines

Software platform Scenematic uses a three-tier cost model in its generative video pipeline, where a quick sketch costs $0.002, a full render costs $0.50, and a flawed scene that compounds through the pipeline costs roughly $15.50. The system normally runs low-step sketch previews to optimize parameters before committing to expensive full renders. However, an out-of-distribution detector built into the pipeline identifies prompts that fall too far outside the training corpus, flagging them to bypass the sketch step entirely. This is because unfamiliar prompts cause the surrogate scorer to return meaningless noise, turning a cheap check into a misleading signal that drives costly downstream errors. The detector uses cosine similarity against a reference corpus of 25 prompts to measure uncertainty, and logs every decision to a database table for ongoing calibration.

How to Extract HEX Colors From Any Image Using JavaScript and Canvas API

Designers and developers often need to identify exact colors from images like logos, screenshots, or UI designs. Using the browser's built-in Canvas API, this can be done entirely client-side without uploading files to a server. The process involves loading an image, drawing it onto a canvas, and reading pixel data using the getImageData() method. This method returns RGBA values for any given pixel, which can then be converted into HEX or RGB color formats. The approach requires no external libraries and works directly within modern web browsers.

Hormuz ship crossings dip below 20