On-chain audit reveals 0.002 BNB discrepancy due to missing gas grant in treasury receipt

·1 views

A treasury's buyback receipt showed a perfect balance, but an on-chain audit revealed a 0.002 BNB discrepancy between two consecutive transactions. The difference was a gas grant sent to an escrow wallet, which the receipt builder had failed to record. The team confirmed the error and fixed the reporting tool. The incident highlights the importance of mechanical, reproducible checks that anyone with a node can perform. The audit process emphasized verifying raw data, avoiding assumptions about decimals, and requiring specific evidence for all on-chain claims.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Study finds fewer billing surprises but more credit complaints in photo apps

A review analysis compared consumer complaints between subscription-based and credit-based photo editing apps. The study examined 189 reviews for subscription apps and 106 reviews for apps using a credit system. While unexpected billing complaints were lower in credit-based apps, new complaints emerged about credit systems, such as purchase issues and unclear pricing. The author notes the analysis is preliminary and based on a limited sample of apps that use a mixed subscription-credit model.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Viral AI gummy bear test highlights flaws in relying on benchmarks or anecdotes

A recent viral social media test compared AI models Claude Opus 5.5 and Gemini 4 Argon by prompting them to simulate gummy bear physics, revealing a stark difference in their outputs. While entertaining, the test was an informal, single-prompt exercise and not a controlled evaluation of overall model capability. The author argues such tests can reveal specific behavioral nuances that aggregate benchmarks miss, but cautions against letting viral anecdotes or leaderboards dictate model choice. The only reliable evaluation for a user's specific needs is testing the models directly on their own workload and data, a point underscored by widely varying cost reports for the same gummy bear task.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How artificial intelligence identifies objects in images using computer vision

Computer vision enables AI to recognize objects in images by analyzing pixel patterns rather than relying on manual programmer descriptions. Systems use models trained on labeled datasets to learn distinguishing features, allowing them to identify objects in new, unseen images. Deep learning approaches, particularly convolutional neural networks, help extract meaningful patterns despite variations in lighting, backgrounds, and angles. The training process involves optimizing model parameters through repeated comparisons between predictions and actual labels. Modern computer vision applications like Google Lens demonstrate this technology's practical implementation in everyday tools.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Infinimesh 3.4 IoT platform adds built-in alarm and rule system

Infinimesh, an open-source Apache-2.0 IoT platform written in Go, has released version 3.4. The update integrates rules and alarm features directly into its core, eliminating the need for separate monitoring systems. A developer demonstrates setting up the platform and creating a functional alarm rule in about ten minutes. The example involves configuring a simulated industrial device to trigger an alert when a sensor reading exceeds a set threshold.

0 comments Read more at DEV Community