Your own factory: three open approaches
Fabro, Symphony and Gas Town: architecture and practical preparation
Slide contents
1. Your own factory: three open approaches
Fabro, Symphony and Gas Town: architecture and practical preparation
2. Different factory layers are open
A process graph, a stream of issues and coordinated workers
3. A README is not a run
Pin source evidence separately from execution evidence
4. The process becomes a graph
Fabro: plan, approve, implement, simplify
5. Transition rules are visible in diffs
Node work, model class and branching can be read separately
6. Each stage leaves inspectable evidence
Code and execution metadata are captured in checkpoints
7. The tracker drives dispatch
The orchestrator reads state; the agent performs the work
8. Policy lives in WORKFLOW.md
YAML configures execution; Markdown instructs the agent
9. Restart reconstructs scheduler state
Tracker and files survive; orchestrator memory does not
10. Coordination gets explicit roles
Dispatch, worker health and merging have separate owners
11. Code and work have separate state
Git worktrees hold code; Beads and Dolt track work
12. A shared queue checks merged work
Test a batch, then isolate the failing request
13. Compare responsibilities, not names
Unit of work, state, validation and operator
14. Write access is not acceptance
The worker changes the product; the evaluator owns criteria
15. Open code does not erase costs
Orchestration license, models, environment and operator effort
16. One bug probes three approaches
T1: repeated delivery must not repeat the effect
17. Interruption probes real state
Interrupt a known phase, then observe resumption
18. Installation is part of the outcome
A clean environment, a pinned version and an intervention log
19. Choose from the bottleneck
Process, issue flow or parallel-work coordination
20. A factory needs an operator
Openness enables changes and leaves operational responsibility