Online edition for BLOGE
0.9.8-RC1· facts verified 2026-09-15 · 中文
Arc 5 Recap — Technical Confidence (Chapters 18–22)
You can now state exactly what your tests, telemetry, wiring, and capacity evidence prove—and where their boundary ends.
The Confidence Stack
| Chapter | Question answered | Evidence produced |
|---|---|---|
| 18 — Testing Your Graphs | Did the chosen behavior occur? | Operator, graph, and DSL checks with controlled time |
| 19 — Observability in Production | Can one real execution be explained? | One incident reconstructed through four projections |
| 20 — Spring and Production Wiring | Who creates and overrides runtime components? | Minimal auto-configuration plus one named override |
| 21 — Scheduling and Complexity in One JVM | How do ready sets and graph complexity shape local work? | Scheduling contract tests plus a local measurement protocol |
| 22 — Distributed Runtime and Capacity Validation | What makes a cluster claim credible? | Routing, lease, store, recovery, and migration boundaries |
The chapters form a ladder. A green graph test does not prove production explainability; a trace does not prove business correctness; one-host throughput does not prove horizontal scale.
What You Should Be Able to Do Now
- Separate operator, graph, DSL, and time-dependent tests by the claim each supports.
- Reconstruct one incident from lifecycle, business, trace, and capacity views.
- Auto-configure a Spring application and override one component with one owner.
- Tune complexity limits and report a reproducible single-JVM measurement from your recorded environment.
- Refuse a distributed-capacity claim until routing, lease, shared-store, recovery, and migration evidence exists.
Common Claim Errors
| Claim | Hidden gap | Better statement |
|---|---|---|
| “The workflow is correct.” | Only terminal output was checked | “This scenario produced the expected node states and output.” |
| “Production incidents are debuggable.” | Logs cannot be joined into one execution | “These four projections share one execution identity.” |
| “Spring manages everything.” | A manual bean silently replaced auto-configuration | “This named override has one explicit owner.” |
| “BLOGE scales horizontally.” | Benchmark ran on one host | “One JVM sustained X under the recorded workload.” |
Try It: Reject an Overclaim
A benchmark reaches 20,000 graph runs per second on one 32-core host. The team writes “BLOGE scales horizontally.” List the missing evidence, then rewrite the claim so it is exactly supported. Separate scheduler throughput from routing, lease ownership, shared-store recovery, and migration.
Where to Go Next
Arc 6 changes the question from “did the runtime execute?” to “did the business promise hold, and what evidence may enter a release decision?” Continue with Chapter 23 — Business Correctness Is Not Test Pass. For production configuration details, use Appendix I.