A new Claude skill dropped on Hacker News this weekend that tackles one of the most persistent problems in modern software delivery: measuring whether AI-augmented teams are actually creating value or just burning resources. The Earned vs Burned Framework, built by former Fortune 500 Chief Data Officer Harveer Singh, replaces traditional velocity tracking with a outcome hierarchy that only credits teams when measurable business results hit production.

The Problem With Current Delivery Metrics

Singh argues that most teams building with AI are measuring the wrong things entirely. Token usage, lines of code written, story points closed—these are all "burned metrics" that measure effort rather than value delivery. His framework flips this completely: value is only earned when a tangible, verifiable business outcome is achieved. Until that point, every hour and token spent is simply burned. The framework originated at a manufacturing and logistics data factory processing over one million SKUs using human-AI hybrid workflows—work that predated the term "AI-native delivery" but demonstrated the same measurement challenges teams face today. Singh has now generalized this approach for modern software delivery with his Claude skill implementation.

The Five-Level Outcome Hierarchy

The framework uses a simple progression from zero to full earn. Level 0 (Not Started) earns nothing. Level 1 (In Progress) still earns nothing—effort alone doesn't count. Level 2 (Dev Complete) earns 25% credit once code is written and unit-tested. Level 3 (QA Passed) reaches 60% but isn't production-ready yet. Level 4 (Deployed to Prod) sits at 85%, tantalizingly close but not confirmed. Only Level 5—Outcome Verified, where KPIs move, users confirm value, and revenue impacts are measurable—fully earns the work. This strict definition forces teams to ask uncomfortable questions about whether their "done" work actually delivered anything.

Key Metrics That Actually Matter

The skill calculates four core metrics: Earn Rate % (L5 outcomes divided by total stories, target 70%+), Earned per Hours ratio (total earned divided by hours, targeting above 0.10), and crucially, Earned per AI Token—a metric Singh notes the industry doesn't have yet that replaces raw token volume tracking entirely. The fourth metric tracks Outcome Verification percentage to ensure teams aren't getting credit for production deployments without confirmed results.

Integration and Installation

The skill works with Linear, Asana, GitHub Issues, Jira, and Azure DevOps out of the box. Teams can paste task lists directly or upload CSV files. Once installed in Claude Desktop or Cowork via the .skill file from the Releases page, users interact naturally: "Score my Linear sprint against the Earned vs Burned framework" or "We burned 400 hours and 2M tokens this month—what did we earn?" The skill generates a one-page Earned Value Report with three key numbers that Singh says should replace traditional velocity reports entirely. It also ends every report by coaching teams to ask the question that changes delivery culture: what is the verifiable outcome that will confirm this work is done?

Open Source and Industry Applications

The skill implementation carries MIT license, allowing free use, forking, adaptation, and building. The framework intellectual property remains with Singh, but contributions are welcome for new integrations (Monday.com, Shortcut, Notion, ClickUp), example sprints with anonymized data, translations, and industry-specific adaptations for healthcare, finance, and government.

Key Takeaways

  • Traditional delivery metrics like story points and velocity measure effort, not value—Singh calls these "burned" metrics
  • The framework only credits teams at Level 5 (Outcome Verified) where business KPIs actually move
  • Earned per AI Token replaces token volume as the efficiency metric for AI-augmented teams
  • Works across FTE teams, outsourced delivery, and pure AI-agent workflows

The Bottom Line

This is the kind of measurement framework the industry desperately needs right now. Too many teams are celebrating "shipping faster with AI" while ignoring whether that speed translates to actual business value. Singh's Earned vs Burned framework won't fix bad product decisions, but it will make invisible waste impossible to ignore—which might be exactly what your sprint retrospectives have been missing.