Factory pilot: prepare the test
A practical protocol: fixture, four tasks and independent acceptance
Slide contents
1. Factory pilot: prepare the test
A practical protocol: fixture, four tasks and independent acceptance
2. Preparing a run, not reporting results
Practical protocol · no executions completed yet
3. One fixture exercises four changes
Node.js/TypeScript, SQLite, React and a local receiver
4. Prove reproducibility before adding an agent
Proposed fixture commands; implementation is still required
5. Acceptance lives outside worker permissions
Public requirements, independently owned test data
6. Freeze conditions before seeing outcomes
Run specification v1: model, commit, tasks and limits
7. Read the behavioural contract before launch
/missions: goal, clarification, milestones and an agreed plan
8. Repeated requests must not duplicate effects
T1: sequential and concurrent delivery, followed by restart
9. API and UI filters must agree
T2: status and time range, including boundaries and empty results
10. Migration begins with pinned versions
T3: a real breaking change and an official migration guide
11. New requirements extend each system’s output
T4: add event-source filtering to accepted T2
12. A success message starts acceptance
Internal checks and independent verdicts have different owners
13. The log exposes required human help
Illustrative T1 trace; this is not a Factory log
14. Classify assistance by its cause
Requirements, environment, access, code and stopping
15. Stopping keeps the experiment interpretable
Time, budget, blockage and independent acceptance
16. Count every attempt and remaining credit
Illustrative bill: payment, consumption and human minutes
17. The control separates factory benefit from agents
A normal agent session with matched tasks and acceptance
18. Results need real execution evidence
Preparation ends after fixture and entitlement verification