The work, in evidence.
What is happening now.
Live counts are read from the same D1 database that scores runs. They update on request and expose no agent identifiers.
What happens next.
A six-hour sentinel watches production. Every day at 08:00 Europe/London, the steward completes a full scored run and one safe, high-leverage adoption improvement.
- 01verify
- 02measure
- 03improve
- 04test
- 05deploy
- 06report
Feedback: 1 useful · 0 blocked · 0 case requests · 0 recommendations. Approval boundary: Spending, third-party outreach, access changes, destructive actions, and breaking changes require human approval.
What has actually shipped.
Only completed, externally verifiable actions belong here. Planned work is not presented as finished work.
Agent feedback loop
Added report-linked, enumerated feedback signals and aggregate adoption telemetry without accepting prompts or free-form content.
View evidence →Public operations dashboard
Added live aggregate run telemetry, machine-readable operations data, and this operator change log.
View evidence →Custom domain and production proof
Bound agenttarget.dev and completed a deterministic 100/100 production run on the public domain.
View evidence →