Sonal Dinusha scores century vs India, becomes third Sri Lankan to achieve feat

Sri Lankan batter Sonal Dinusha scored a century against India during the ongoing Test series between the two sides. He reached the milestone on the first ball of the third session, flicking Mohammed Siraj to mid-wicket for two runs. Dinusha had been unbeaten on 98 at the tea break before completing his hundred shortly after play resumed. The knock marked another impressive performance from the batter, who has been in strong form throughout the two-match Test series. With this century, Dinusha became only the third Sri Lankan player to achieve this particular feat against India.

ICC bans Pakistan-born ODI cricketer Asif Khan 15 months for steroid use

The International Cricket Council has handed a 15-month ban to Pakistan-born ODI cricketer Asif Khan following a doping violation. A sample collected from Khan on September 8, 2025, tested positive for a metabolite of mesterolone. Mesterolone is a prohibited anabolic steroid commonly used to boost testosterone levels. The ban was imposed in accordance with the ICC's anti-doping regulations.

Nepal power hit, may seek India help

Nepal disaster renews climate change risk debate

How Badge-Swipe Bonuses Turned Office Attendance Into a Game Employees Win

Companies including Google and Amazon have begun tying employee bonuses, performance reviews, and even termination decisions to office badge-swipe data, according to internal memos and a 2024 ResumeTemplates survey of 713 business leaders. The survey found that 18% of companies dock bonuses based on badge data, while 60% use it as their primary attendance-monitoring method. In response, workers have adopted 'coffee badging' — briefly swiping in, buying a coffee, and leaving — a practice that Owl Labs' 2025 report found 43% of hybrid workers were already doing. Data from Occuspace shows this behavior has inflated office occupancy figures by 15–20%, meaning compensation decisions are being made on inaccurate attendance records. Labor strategist John Frehse of Ankura warns that coffee badging signals deeper workplace dysfunction, and that companies are responding not by rethinking the metric but by adding more surveillance tools.

Armenia Signs $2 Billion in Defence Deals with India Since 2020

Armenia has entered into defence contracts with India totalling nearly $2 billion since 2020, signalling a significant and growing bilateral security partnership. The deals reflect India's expanding role as a major arms supplier beyond its traditional markets. For India, the relationship with Armenia offers strategic value in a region bordered by China and Pakistan-aligned nations. The partnership underscores the mutual interest both countries have in deepening military and diplomatic ties.

Windows NTFS Permissions Are More Complex Than Linux's Simple Permission Model

A developer completing cybersecurity training on HackTheBox found Windows permissions significantly more confusing than Linux's after finishing the Windows Fundamentals module. While Linux uses a straightforward read/write/execute model displayed via the 'ls -l' command, Windows relies on the 'icacls' tool and Access Control Lists (ACLs) made up of multiple Access Control Entries. Windows distinguishes between two ACL types: DACLs, which govern who can access a resource and how permissions propagate, and SACLs, which log access attempts. Basic Windows permissions include six categories such as Full Access, Read/Execute, and Modify, though the GUI and command-line tool display these differently, adding to the confusion. The repeated entries seen in icacls output are not duplicates but separate lines conveying permissions and inheritance rules for subfolders respectively.

Benchmark of 5 Managed Graph Databases Shows No Single Winner Across All Workloads

A developer benchmarked five managed graph databases — CognoDB Cloud, Neo4j AuraDB, Memgraph Cloud, FalkorDB Cloud, and ArangoDB Oasis — using the same dataset of 27,770 physics papers and 352,807 citation edges on free-tier cloud instances. Memgraph led on traversal queries and point lookups, completing 1-hop queries at a median 69ms, while Neo4j AuraDB finished second on those same tests. However, Neo4j reversed the standings on full-graph aggregation queries, outperforming Memgraph and leaving CognoDB and ArangoDB far behind at 1.8 and 4 seconds respectively. Under concurrent load testing with 10 and 40 simultaneous clients, four platforms scaled throughput roughly fourfold as expected, but ArangoDB's throughput barely moved, suggesting possible architectural or free-tier resource constraints. The results highlight that short-traversal performance and full-scan aggregation workloads stress graph databases in fundamentally different ways, making single-query benchmarks insufficient for real-world database selection.

Vector RAG Outperforms Classic RAG in Speed, Scale, and Cost Efficiency

Vector RAG (Retrieval-Augmented Generation) is emerging as the preferred retrieval method for production-grade large language model applications, according to a technical analysis published on DEV Community. Unlike classic RAG systems that rely on BM25 or TF-IDF text search, Vector RAG uses dense embeddings and vector indexes to cut query latency from 200–400 ms down to 30–80 ms. Approximate nearest neighbour algorithms used in libraries like FAISS, Milvus, and Pinecone reduce retrieval time by 5–10 times while maintaining recall above 95 percent. The approach also handles semantic similarity better, meaning misspellings, synonyms, and partial matches are less likely to cause hallucinations compared to classic retrieval methods. Vector indexes can be horizontally sharded and scaled with managed services, allowing teams to handle traffic spikes without rebuilding the entire index.

How to Build a Wait-Free Queue in Rust Using Atomics and Ring Buffers

A technical deep-dive on DEV Community walks through building a wait-free queue in Rust, targeting systems where thread blocking is unacceptable. The article distinguishes wait-free algorithms from lock-free and obstruction-free approaches, explaining the stronger progress guarantees each offers. It covers how atomic operations and memory ordering serve as synchronization primitives, and how a bounded ring buffer with sequence numbers resolves producer-consumer ownership conflicts. The guide is aimed at developers working on low-latency pipelines, game engines, network runtimes, or other high-throughput systems. Rust's ownership model and type system are highlighted as useful tools, though the author stresses that understanding the underlying algorithm remains essential.

Mentat Offers Runtime LLM Steering for Financial Agents Without Retraining

Y Combinator F24 startup Mentat has introduced a runtime intervention method that modifies token probabilities inside a language model's computation graph during inference, without altering model weights or requiring fine-tuning. Developers send steering directives alongside their API requests, and Mentat intercepts the model's forward pass at specific transformer layers to bias reasoning patterns in real time. The approach targets financial agent use cases where deterministic, auditable outputs are critical, offering an alternative to costly fine-tuning or unreliable prompt engineering. The technique carries an estimated 10–30% per-token latency overhead and introduces additional memory and batch-efficiency trade-offs compared to standard inference. Compliance considerations remain significant, as firms must log exact rule versions per inference call and be prepared to explain steering logic to regulators.

Could UK military sites come under attack? London gets warning from Russia

NCLT Slashes Subhash Chandra's Debt from Rs 22,006 Crore to Rs 6.5 Crore

The National Company Law Tribunal (NCLT) has approved a resolution plan in the personal insolvency proceedings of media mogul Subhash Chandra, dramatically reducing his outstanding debt. His total liability of Rs 22,006 crore has been cut down to just Rs 6.5 crore following the tribunal's order. This marks a significant development in Chandra's personal insolvency case, which had been ongoing before the NCLT. The ruling provides substantial financial relief to the Essel Group founder, effectively resolving the bulk of his personal debt obligations.

How to Handle Algorithm Output Gaps When Porting Computer Vision to New Frameworks

Porting computer vision algorithms from Python OpenCV to other frameworks often produces output inconsistencies due to differences in algorithmic formulations, numerical precision, and hardware-specific optimizations. These variations are not implementation errors but structural trade-offs between platforms, such as differing default kernel sizes or floating-point rounding behaviors. In parallel deployment environments like cloud and manufacturing systems, such discrepancies can erode user trust, especially where regulatory or performance standards apply. Experts recommend defining 'functional equivalence' based on end-user tolerance thresholds rather than pursuing exact numerical replication. A diagnostic approach — isolating subalgorithms, comparing mathematical formulations, and quantifying intermediate deviations — helps engineers make informed trade-offs during the porting process.

India, China, Canada & more: 10 countries still under Trump’s tariff squeeze and why

AI Writes Fluent But Factually Wrong Docs — Here Is How to Fix That

Large language models tasked with generating technical documentation tend to produce well-structured output that contains fabricated details, such as non-existent parameters or incorrect retry logic, because they pattern-match code to similar libraries rather than reading it precisely. The author argues that wrong documentation is more harmful than no documentation, as it discourages developers from consulting the source code. The proposed fix is to stop asking AI to invent documentation and instead ask it to transform artifacts that already encode ground truth, such as type signatures, passing tests, OpenAPI specs, database migrations, and git history. Constraining the model to these inputs — and instructing it to write 'UNKNOWN' for anything not present — significantly reduces hallucination. The author also recommends scoping documentation updates to individual diffs rather than whole-repository passes, making AI-generated suggestions small enough to review meaningfully.

ChatGPT, Claude or Gemini: A Developer's Practical Guide to Choosing in 2026

Benchmark leaderboards rarely reflect real-world developer needs, making hands-on testing a more reliable way to choose between ChatGPT, Claude, and Gemini. The three models differ notably in how they handle long multi-part instructions, with Claude tending to restate constraints, ChatGPT sometimes dropping later items, and Gemini occasionally compressing requests. Their behavior when they lack knowledge also varies significantly — ChatGPT is often reported to fabricate plausible but incorrect API calls, while Claude hedges more openly and Gemini's accuracy shifts depending on whether search grounding is enabled. Safety refusals present another operational difference, as Gemini can block responses at the API layer in a way that may cause code errors, unlike the text-based refusals from ChatGPT and Claude. Developers are advised to evaluate the specific model-tool combination they plan to deploy, since agents like Claude Code, OpenAI Codex, and Gemini CLI each behave differently in agentic workflows.

Tutorial: Build a Fault-Tolerant Market Data Aggregator in Python with SerpApi

A new developer tutorial on DEV Community walks through building a production-grade Python command-line tool that fetches real-time stock, market index, and cryptocurrency data using SerpApi's Google Finance engine. The guide focuses on connection resilience, a common challenge in automated trading bots and financial scraping systems. Developers are shown how to implement custom exception handling and an exponential backoff retry loop to gracefully manage network timeouts and API failures. The tool reads ticker symbols as command-line arguments and outputs formatted price and movement data for assets like AAPL, GOOGL, and BTC-USD. Authentication is handled securely via environment variables, avoiding hardcoded API keys in source files.

Solo Developer Builds AI Contract Analysis Tool That Scores Clause-Level Risk

A solo developer has launched EasyContract, a web tool that analyzes PDF and DOCX contracts by breaking them down clause by clause. The tool assigns risk scores ranging from low to critical, provides plain-language explanations of legal jargon, and suggests rewrites for risky clauses. Built on a stack of Laravel, PostgreSQL, Vue 3, and AWS infrastructure, the platform also supports shareable reports. Users can try the tool without creating an account at easy-contract.app/try. The developer has shared the project on DEV Community and is actively seeking feedback from other developers.

Folio

Discussion | Link