Verify the fixture. Line up a ratifier.
You read the same fixture the engineers read, but from the specification side rather than the diff. Preparation confirms the repository runs and that you have someone who can ratify your work.
Bring a ratifier, or the course runs at half strength
Bring a ratifier, or the course runs at half strength
Stage 6 hands your specification to someone who can say whether it is buildable and evaluable. You cannot play that role for yourself, and a second product manager is a weaker substitute than an engineer.
- Best
- Run this alongside course 301. The engineers ratify the specs and receive the escalation at stage 8, which is the point of pairing the two courses.
- Workable
- One engineer attends stages 6 and 8 only. Roughly an hour of their time, at two fixed points.
- Weakest
- Pairs ratify each other. The specs still improve, but nobody in the room can settle a feasibility objection.
- Stage 9
- Needs a facilitator-supplied outcome signal the room has not seen. Round 0 cannot be reused, because by then you have read its answer.
Three steps, shared with 301
Verify the fixture
The same sibling repository course 301 uses, at the same verified revision. You need it running to read the specs and open an evidence bundle, not to change code.
cd ../payments-validation-fixturegit rev-parse --short HEADnpm run verify
- specs/product/payments.md
- The requirement you will diagnose in stage 1 and rewrite across stages 2 to 5.
- specs/tech/payments.md
- What the engineers assumed you had already decided.
- corpus/classes.ts
- F1 through F8. Stage 1 asks which of them a specification can cause.
Produce one evidence bundle
Stage 7 reads a bundle rather than generating one. Produce it once during preparation so instructional time is spent reading it.
export PATH="$PWD/bin:$PATH"workbench validate --no-gatesworkbench report
- The bundle
- JSON written to bundles/. Stage 7 works from this file.
- notExamined
- The list of classes no check looked at. Stage 7 reads it before the findings.
Read the workbench reference
Read it once before the course. You are not learning to run the checks. You are learning what each kind of evidence can support, so that stage 3 can attach an evaluation path to every criterion.
- Eight failure classes
- Which are mechanized, which need judgment and which one is reserved for a person.
- Six evidence types
- Counterexample, proof, statistical, corroborated opinion, single opinion and silence.
- The misreadings
- Each type has a characteristic misreading. Stages 3 and 7 both turn on knowing them.
The referenceThe workbench is shared with course 301. Both audiences read the same eight failure classes and the same six evidence types: open the workbench.
Ready to start when
- The fixture is at the expected revision and npm run verify passes.
- One evidence bundle exists in bundles/ and workbench report names the unexamined classes.
- You have read the workbench reference and can name what a statistical result does not support.
- A ratifier is committed for stage 6, and a facilitator has an unfamiliar signal ready for stage 9.