An open source Claude skill for measuring what actually matters in AI-native delivery.
By Harveer Singh | Author, * When Data Moves* | Founder, Rizz Wireless
Every team building with AI is measuring the wrong things.
Token usage. Lines of code. Story points. Velocity. Hours burned. These are burned metrics β they measure effort, not value. A model that hallucinates in 100 tokens is not better than one that solves the problem in 10,000. 500 lines of code that don't ship earn nothing. 40 story points closed on work that never reaches production delivers zero business value.
The only measure that matters: did we earn it?
Value is only EARNED when a tangible, verifiable business outcome is achieved. Until that point, all effort is only BURNED.
This is the Earned vs Burned Framework β originally built at Deloitte in 2010 for a Sysco Foods data factory processing 1M+ SKUs (a human-AI hybrid workflow that predated the term). Now generalised for AI-native software delivery.
| Level | Gate | Earned | What it means |
|---|---|---|---|
| 0 | Not Started | 0.00 | Backlog. Zero burn, zero earn. |
| 1 | In Progress | 0.00 | Effort accumulating. Still zero earned. |
| 2 | Dev Complete | 0.25 | Code written and unit-tested. Partial. |
| 3 | QA Passed | 0.60 | Tested and accepted. Meaningful but not in prod. |
| 4 | Deployed to Prod | 0.85 | Running in production. Close β but not confirmed. |
| 5 | Outcome Verified | 1.00 | |
| KPI moved. User confirmed. Revenue impacted. The only full earn. |
| Metric | Formula | Target |
|---|---|---|
| Earn Rate % | L5 outcomes Γ· Total stories | 70%+ |
| Earned / Hours | Total Earned Γ· Total Hours | >0.10 |
| Earned per AI Token | ||
| Total Earned Γ· Total Tokens | Trending up | |
| Outcome Verification % | L5 verified Γ· L4+L5 deployed | Trending up |
Earned per AI Token is the metric the industry doesn't have yet. It replaces token volume entirely.
Install this skill into Claude and it can:
Pull tasks from any project toolβ Linear, Asana, GitHub Issues, Jira, Azure DevOps, or a pasted/CSV list** Score each taskagainst the 5-level Outcome Hierarchy Calculate all metrics**β Earn Rate, E/H Ratio, Earned per Token, Outcome Verification %, Team Effectiveness** Generate an Earned Value Report**β one page, three numbers, replaces your velocity report** Coach your team**β ends every report with the question that changes delivery culture:what is the verifiable outcome that will confirm this work is done?
Works for FTE teams, outsourced/offshored delivery, AI-agent workflows, or any mix.
- Download
earned-vs-burned.skill
from theReleasespage - In Claude Desktop or Cowork: Settings β Capabilities β Install Skill
- Drop the
.skill
file in
git clone https://github.com/[your-org]/earned-vs-burned-skill
Once installed, just talk to Claude naturally:
"Score my Linear sprint against the Earned vs Burned framework"
"Here are my Jira tasks β how much value did we actually earn this sprint?"
"We burned 400 hours and 2M tokens this month. What did we earn?"
"I'm an outsourcing vendor β help me prove our value to the client using outcome metrics"
"Pull our GitHub issues and give me an Earned Value Report"
earned-vs-burned/
βββ SKILL.md # Core skill instructions for Claude
βββ README.md # This file
βββ references/
β βββ framework.md # Full framework, origin story, hierarchy detail
β βββ metrics.md # "So what" interpretation guide for each metric
β βββ integrations.md # How to pull from Linear, Asana, GitHub, Jira, ADO
βββ assets/
βββ tracker_template.csv # Blank tracker to fill in and upload
This is an open framework. Contributions welcome:
New integrationsβ Monday.com, Shortcut, Notion, ClickUp** Example sprints**β Real anonymised data showing the framework in action** Translations**β The framework in other languages** Locale variants**β Adaptations for specific industries (healthcare, finance, gov)
Open a PR or issue. The framework is IP of Harveer Singh; the skill implementation is MIT licensed.
Framework IP: Β© Harveer Singh. The Earned vs Burned Framework, Outcome Hierarchy, and associated concepts are original intellectual property of Harveer Singh (established 2010, Sysco/Deloitte; formalised 2026).
Skill code: MIT License. Use, fork, adapt, and build on the implementation freely. Attribution appreciated.
Harveer Singh is a former Fortune 500 Chief Data Officer (Truist Bank, Western Union), founder of Rizz Wireless, and author of * When Data Moves*.
He built the original Earned vs Burned model at Deloitte in 2010 β before "AI-native" was a phrase anyone used β and has spent 25 years proving that data and technology programs exist to deliver outcomes, not activity.