Newly closed Results windows use method v2. A metric can report a rise or fall only with full coverage of both windows. Citation comparisons use prompts active on the same platform in both windows. Answer-fetch comparisons count only sensor-covered days, and the test allows for variation in daily counts. Citation comparisons also allow for variation between prompts and need at least three prompt and platform pairs with answers in each window.
Previously closed windows keep their v1 results. The ledger shows Method changed when its rows mix methods.
API and MCP
The Results ledger adds items[].method, either v1 or v2 for closed Results windows with evidence and null while open or for Agent Picks rows. Agent Picks keeps agentPicks.method. Timeline outcome events carry measuredAction.method. Both the REST API and devtune_get_outcomes_ledger expose the same method and evidence.