SShortSingh.
Back to feed

15 Automated Checks All Passed — Yet the Live Demo Was Running the Wrong Version

0
·1 views

A developer's release script cleared all 15 automated checks for version 1.8.0, yet the publicly visible demo was still displaying the previous version, 1.7.0. The checks covered version consistency, changelog updates, site availability, and image loading, but none monitored what the demo app itself was actually running. The root cause was that the deploy script published the marketing site wholesale, while the demo was a separately compiled build that nothing had rebuilt or verified. A fix was added to query Flutter's auto-generated version.json file directly from the live URL and compare it against the expected release version. However, a subtler follow-up bug revealed that manually updating the demo's destination folder was undone by the next deploy, which faithfully re-synced the stale source — illustrating that verifying an action was taken is not the same as verifying the deployed artifact.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

Google Launches Antigravity 2.0 Product Page

Google has published a product page for something called Antigravity 2.0 at antigravity.google. The listing appeared on Hacker News, drawing minimal engagement with only 4 points and no comments. The sparse article provides no technical details or description of what the product or service entails. It remains unclear whether this is a real product launch, an internal experiment, or an April Fools'-style joke by Google.

0
ProgrammingDEV Community ·

React SSG site grows from 3 to 664 clicks after SEO audit uncovers hidden bugs

A developer running devtools.abect.com, a browser-based React 19 static site offering 49 developer tools, grew organic clicks from 3 to 664 and impressions from 1,600 to 21,400 over a three-month period. Key changes included creating one dedicated page per tool with 5,000–12,000 characters of real content, prerendering full HTML at build time, and adding structured JSON-LD schema on every page. Unexpectedly, text conversion tools like JSX-to-HTML drove the majority of traffic, while the image converters that originally motivated the project contributed very little. The author attributed the text tools' success to lower competition for narrow developer search queries compared to the saturated image conversion market. A subsequent audit of the actual shipped HTML — rather than the source code — revealed bugs that had gone unnoticed for months, underscoring the importance of inspecting build output directly.

0
ProgrammingDEV Community ·

Enterprise AI in 2026: Free Cloud Tiers Gone, Agent Reliability Now Top Priority

By 2026, the once-simple AI stack of a single LLM API, vector database, and frontend has given way to complex, multi-layered architectures demanding cost discipline and system resilience. Cloud providers have ended subsidized free tiers as infrastructure costs for training and serving large language models outpaced their ability to absorb them indefinitely. Engineering teams are now shifting from flexible pay-as-you-go pricing to reserved capacity models, with cost optimization treated as a core architectural requirement rather than an afterthought. AI agents, while promising autonomous planning and execution, have proven fragile in production due to non-determinism, making agent reliability the industry's most pressing technical challenge. As a result, software engineers must now build deterministic safeguards and cost-aware routing into systems from the earliest stages of development.

0
ProgrammingDEV Community ·

Agentic vs AI-Powered Testing: A Developer Breaks Down the Real Difference

A developer writing for DEV Community argues that 'AI-powered testing' and 'agentic testing' are frequently conflated despite being distinct concepts. AI-powered testing broadly covers using AI to generate test code, analyse failures, heal broken selectors, or explain whether a bug stems from the product or the test itself. Agentic testing, by contrast, involves describing a user journey or goal in natural language and letting an agent autonomously decide how to execute it within a live application. This approach is particularly suited to end-to-end testing, where the focus is on real user workflows such as sign-in, checkout, or account creation, rather than precise unit-level assertions. The author notes that while self-healing automation was once unreliable, modern browser agents and language models have made limited autonomous recovery increasingly practical.

15 Automated Checks All Passed — Yet the Live Demo Was Running the Wrong Version · ShortSingh