Value scenario discovery
Invoke /value-scenario-discovery <phase-or-contract-reference> to propose
scenarios, never to run them. It is the define step of the independently usable
VLBV method.
Emit Starting value-scenario-discovery — <phase> first and
Completed value-scenario-discovery — <phase> (proposal) last. Take phase
identity and totals from actual context; never invent a counter.
Resolve context
Read the governing repository/host policy, the current adopted contract (if any) and any owner calibration already reviewed in this phase. Use the latest authorized supersession, not a remembered command. A missing critical identity or a policy conflict is stated explicitly. A tool being available is not permission to run it.
Reuse the owner's useful operation, baseline (including UNKNOWNs), supported environments, constraints, exclusions and decision owner from Assess. Do not ask again for approval already given. If no examples exist, draft a few for owner review before dependent generation: one ordinary useful operation, one available but irrelevant test (deferred, with reason), and one rare mandatory constraint (integrity, authorization or an irreversible effect). They stay proposals until reviewed.
Produce the finite frontier
Start from the ordinary operation the component exists to deliver. Include mandatory integrity, safety, permission and irreversible-effect constraints even when rare. For each candidate record:
| Field | Content |
|---|---|
| Candidate ID | Local candidate ID; an adopted scenario ID (^[A-Z]+-[0-9]+$) only if one already exists |
| Actor and outcome | Who needs the behavior and the observable value |
| Basis | Source requirement, or an explicitly labelled assumption |
| Reason now | Why it matters to the current phase |
| Likelihood | Sourced, qualitative, or UNKNOWN |
| Mandatory constraints | Kept separate from ordinary usefulness |
| Proposed observation | Real production entry, oracle source, plausible false positive |
| Disposition | Selected, rejected or deferred, with reason |
| Next decision | What observing it would let the owner decide |
Use the contract's finite budget where one exists; otherwise propose a small frontier and its stopping criterion before expanding it. Do not invent probabilities or weighted scores. Stop when the frontier supports the current decision. An imagined alternative does not become a blocking requirement. When the model or context changes materially, review unfamiliar candidates against the saved examples rather than resetting reviewed scope.
Save and hand off
Write the proposal to the requested phase artifact, normally
.kbd-orchestrator/phases/<phase>/scenario-candidates.md, listing the reviewed
examples, selected and deferred candidates with rationale, assumptions, UNKNOWNs
and the specific pending owner decision. Never label unreviewed output adopted.
Assess uses the proposal to refine usefulness. Only owner-adopted cases enter
adopted-contract.json scenarioMatrix during Spec/Plan (id, ownerTask,
productionEntry, oracle, negativeControl, fixtures, evidence) through the
existing adoption path. Bind acceptance-bearing calibration documents through
that contract's adoption or amendment; create no second acceptance ledger and no
extra schema fields.
Discovery defines scenarios before generation. It does not author features or
steps, import generated code, install dependencies, or run dry-runs, builds,
lint, typechecks, probes or runners. Executable authoring is
bdd-cucumber-js author-only (see skills/bdd-cucumber-js/references/author-only.md);
execution is value-scenario-verify at the separately authorized complete
value-phase boundary. No automatic task, phase, adoption, publication or release
follows a proposal.