Unity vs Valve

AppLovin and Unity Legal Dispute over Data Leakage

AppLovin and Unity, two mobile advertising giants, are embroiled in a legal dispute over the definition of 'data leakage' in ad transactions. The conflict, formalized through court filings in San Francisco, centers on where one company's right to observe an ad transaction ends and the other's proprietary data begins. The outcome could have far-reaching implications for data handling and competitive practices in mobile advertising. The dispute underscores the growing importance of data ownership and transparency in the ad tech ecosystem.

• AppLovin and Unity are in a legal dispute over 'data leakage' in ad transactions.
• Court filings were made in San Francisco.
• The dispute centers on data ownership and observation rights in mobile advertising.

Unity Launches Official Plugin for Grok Build

Unity Technologies has made its official Unity Plugin available for Grok Build, marking the third coding agent to receive native Unity engineering guidance. The plugin provides Grok Build with over 30 skills covering UI Toolkit, uGUI, 2D/tilemaps, URP and Shader Graph, audio, navigation, physics, in-app purchases, multiplayer, web, and localization. This move follows Unity's recent launch of its CLI and official plugins for other coding agents like Claude Code and Codex. Unity emphasizes supporting developers across coding agents, rather than dictating tool choice. The plugin is maintained from a single repository, ensuring consistent updates across agents, and is compatible with Unity 6 and above. Unity plans to publish developer usage data of its CLI and plugins later this month.

• Unity released an official plugin for Grok Build, making it the third coding agent with native Unity support.
• The plugin provides over 30 skills including UI Toolkit, 2D, URP, Shader Graph, audio, navigation, physics, and more.
• The Grok Build plugin is compatible with Unity 6 and above and installs via a terminal command.

From attention to action: why the state of play matters for marketers

New blog post added: 'From attention to action: why the state of play matters for marketers' with link to unity.com/blog/mobile-gaming-attention-economy-driving-marketing-performance.

OpenAI GPT-6 Astra Agent Beats Portal, Costs $570

A developer known as Cozyblaze used OpenAI's GPT-6 Astra model to autonomously complete the video game Portal, a task involving 3D spatial reasoning, puzzle solving, and multi-step planning. The AI agent was connected to a modified version of the game via the Model Context Protocol. Over 24 hours, the agent processed screenshots and positional data, pausing to plan each action. The experiment required over 3,300 tool calls and incurred API costs exceeding $570. The developer noted this aligns with OpenAI's initial vision of a single agent mastering many games, although it is not a formal benchmark. Potential reliance on publicly available solutions or training data is acknowledged.

• OpenAI released GPT-6 Astra on September 3, 2026.
• Developer Cozyblaze used GPT-6 Astra to autonomously play Valve's Portal.
• The AI agent used screenshots and positional data via Model Context Protocol.

Developer Reports 55,000 Steam Refunds

German developer Mateo Covic (alias Zoroarts) reported that more than 55,000 players requested refunds for his short cooperative game Paddle Paddle Paddle on Valve's Steam store, producing an unusually high refund rate (~21%) despite roughly 90% positive user reviews. The refunds exploit Steam's automated refund rule that allows returns for games owned fewer than 14 days with under two hours playtime — a window that disadvantages deliberately short titles. Covic called on Valve to change the system. The game's publisher, Assemble Entertainment, and sales of about 270,000 units mean the project remains commercially successful despite the high refund volume. Valve has not yet issued an official response.

• Developer Mateo Covic (alias Zoroarts) recorded over 55,000 refunds for his game Paddle Paddle Paddle.
• The refunds represent an approximate refund rate of 21% while user reviews on Steam remain about 90% positive.
• Steam's refund policy grants refunds automatically if a game has been owned for less than 14 days and played under two hours.

Guide: How to Handle Very Large Datasets

This technical guide explains practical steps for working with datasets that exceed spreadsheet limits. It defines 'large' (e.g., >1,048,576 rows or multi-GB files), recommends moving data into a database (SQLite) rather than a spreadsheet, and details real-world quirks: zipped downloads, slow imports and database growth, the performance benefits of creating indexes (including expression indexes), sampling queries during development, aggregating early, and checking data types. It notes SQLite handles many multi-GB files but suggests DuckDB or pandas with chunksize if you outgrow SQLite.

• A dataset is 'large' for spreadsheets when it exceeds about one million rows; Excel and Google Sheets have a hard row limit of 1,048,576 rows.
• The guide's running example is a public Steam reviews file (recommendations.csv) of about 2 GB and 41 million rows.
• Author recommends moving large CSVs into a database (SQLite via DB Browser) because SQLite has no practical row limit for this use case.