test(storyboard): refactor cascade-skip-on-skip to (skip_reason × peer_shape × topology) matrix (#1548)

[bokelley]

Summary

Refactors test/lib/storyboard-cascade-skip-on-skip.test.js from prose-driven scenarios (one named test per scenario, ~2000 lines, 35 tests) to a fixture-table matrix keyed by the three orthogonal dimensions of cascade behavior:

skip_reason     ∈ { missing_tool, missing_test_controller, not_applicable, real_failure }
peer_shape      ∈ { sole_stateful, peer_passes, peer_fails, peer_substitute_declared, peer_substitute_missing }
phase_topology  ∈ { same_phase, downstream_phase }

PR #1545 fixed an asymmetric blind spot — the sole-stateful-step exemption applied to not_applicable (#1146) but not to missing_tool. The parallel nodejs-testing-expert review of #1545 traced the cause to test architecture: prose-driven scenarios hide empty cells in the implicit cube until production hits one. This refactor makes those cells structurally visible.

What's in the new file

• 24 matrix rows — every historical cascade-behavior scenario is now a named matrix row tagged with its (skip_reason, peer_shape, phase_topology) triple. Test names carry the tag prefix so failures point straight at the cube cell.
• 1 control row — the no-skip-occurs sanity check.
• 9 parse-time validation rows — peer_substitutes_for / depends_on malformed-input rejection assertions.
• 1 parity-invariant test — the loop PR fix(storyboard): sole stateful step missing_tool no longer trips cascade #1545 introduced for {missing_tool, not_applicable} × sole_stateful × downstream_phase. Add new exemption-family reasons here when they land.
• 1 coverage-holes assertion — partitions the 40-cell cube into covered / impossible / tracked / empty. Fails CI when a new axis value is introduced and left uncovered.

Total: 36 tests pass (was 35 — added the coverage assertion).

Matrix coverage map

Bucket	Count	Notes
Covered by ≥1 row	11	dedup'd from 24 rows; multiple rows can target the same cell with different fixtures
Structurally impossible	4	real_failure × peer_substitute_* — peer_substitutes_for rescues skips, not failures
Tracked elsewhere or in follow-up	25	mostly missing_test_controller × * (covered in storyboard-controller-seeding.test.js) and real_failure × peer_passes/peer_fails × *_phase (covered in storyboard-runner.test.js)
Empty (uncovered, untracked)	0	the assertion at the bottom fails if this is ever > 0

Reproduce locally: MATRIX_REPORT=1 node --test test/lib/storyboard-cascade-skip-on-skip.test.js prints the full breakdown.

Net line shrink

	Lines
Before	2143
After	1256
Delta	−887 (−41%)

Shrink comes from extracting the repeated fakeAgent setup, storyboardWith[Phases] boilerplate, and 35 near-identical runStoryboard invocations into three small helpers (step, runWith, plus the existing storyboardWith[Phases]). No assertions dropped — every original assert.* call has a 1:1 successor in a row's assert(result) callback.

What's NOT in this PR

• No runner change. This is purely a test refactor. The not_applicable exemption from fix(conformance): sole-stateful-step not_applicable no longer cascades #1146 and the missing_tool extension from fix(storyboard): sole stateful step missing_tool no longer trips cascade #1545 stay green.
• If the matrix expansion exposes additional bugs in the runner, those will be filed as separate issues per the issue's instructions — not fixed in this PR.
• Companion issue Storyboard examples: add hello-seller-adapter-proposal-mode for sole-stateful-step coverage #1549 (per the issue brief) will land as a separate PR.

References

• adcp-client#1548 — this issue
• adcp-client#1545 — bug + fix that surfaced the test-architecture gap
• adcp-client#1146 — original not_applicable exemption
• adcp#4053 — spec coordination for the exemption rule

Test plan

• npm run format:check clean
• node --test test/lib/storyboard-cascade-skip-on-skip.test.js — 36 / 36 pass
• npm test — 8253 / 8256 pass (3 pre-existing skipped, 0 new failures)
• Coverage assertion fires correctly when a row is removed (manually verified the partition by inspecting MATRIX_REPORT=1 output)

Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com