gate-19: duplicate scenario titles collapse to one slug — one test credits several scenarios, and dangling anchors are never reported

[rubenvdlinde]

Summary

admin-settings/spec.md declares the same four scenario titles twice, and gate-19 slugifies a
scenario from its heading text alone. Two structurally different scenarios therefore collapse onto
one <spec>::<slug> ref, and one e2e test marks both covered.

Evidence

openspec/specs/admin-settings/spec.md — the hide-slogan checkbox requirement and the
show-menu-labels checkbox requirement each declare:

line	scenario
128 / 156	#### Scenario: Checkbox reflects enabled state
134 / 161	#### Scenario: Checkbox reflects disabled state
140 / 167	#### Scenario: Checkbox label text and accessibility
146 / 172	#### Scenario: Checkbox change triggers API call

check_e2e_coverage.py::_slugify derives the slug from the heading text only — there is no
de-duplicating suffix and no parent-requirement qualifier. Both headings yield
admin-settings::checkbox-reflects-enabled-state.

In run_gate() coverage is tested with if s['ref'] not in covered_refs, so a single reference
satisfies every scenario sharing that ref. The show-menu-labels scenarios have been counted as
covered by the hide-slogan tests.

Two failure modes, both silent

1. Over-crediting. N duplicate-titled scenarios are discharged by 1 test.
2. Dangling anchors are not reported. The test author had assumed a -1 suffix and wrote
   ...#checkbox-reflects-enabled-state-1 and ...#checkbox-label-text-and-accessibility-1.
   Neither slug exists. The gate neither credited them nor warned — a typo'd anchor and a correct
   one are indistinguishable in its output. Found by diffing refs collected from tests/e2e/**
   against slugs parsed from openspec/specs/*/spec.md; the gate will never tell you.

Suggested fix

Qualify a Format-A slug with its enclosing ### requirement, as Format B already does
(<parent-req-slug>-scenario-<n>). That makes duplicate titles under different requirements
distinct and is consistent with the existing Format-B convention.

Separately, gate-19 should report anchors that resolve to no scenario rather than ignoring
them. That is a few lines in collect_covered_refs / run_gate and would have surfaced both
defects immediately.

Impact on counts

Fixing the collision will make several apps' gate-19 numbers go up, because scenarios currently
credited by a same-titled sibling will become individually uncovered. That is the honest number.

Found while raising nldesign's gate-19 coverage. The two dangling anchors in
tests/e2e/spec-coverage/admin-settings.spec.ts are corrected in ConductionNL/nldesign (they now
point at the real slug); the collision itself cannot be fixed from the test side.

[rubenvdlinde]

Concrete instance, with the feature pair it actually confuses — nldesign.

Measured on ConductionNL/nldesign@101869f with the canonical checker: 786 scenario entries, 781 distinct refs — 5 colliding slugs.

Found by accident, which is the point: I computed dangling anchors as covered_refs - all_refs using a set(), and my positive control printed 781 where --mode report says 786. Without that control the 5 would have been invisible.

admin-settings::checkbox-reflects-enabled-state      2 scenarios   anchored by a test: YES
admin-settings::checkbox-label-text-and-accessibility 2 scenarios  anchored by a test: YES
token-set-apply-dialog::admin-unchecks-a-row          2 scenarios  anchored by a test: YES
admin-settings::checkbox-change-triggers-api-call     2 scenarios  anchored: no
admin-settings::checkbox-reflects-disabled-state      2 scenarios  anchored: no

The collisions are across two different requirements for two different features:

### Requirement: Hide Slogan Checkbox          (line 125)
#### Scenario: Checkbox reflects enabled state (line 128)
#### Scenario: Checkbox change triggers API call (line 146)

### Requirement: Show Menu Labels Checkbox     (line 153)
#### Scenario: Checkbox reflects enabled state (line 156)
#### Scenario: Checkbox change triggers API call (line 172)

The slug is derived from the scenario title alone and is not namespaced by the enclosing requirement, so hide_slogan and show_menu_labels — two independent appconfig-gated features — share one addressable ref.

Two consequences, and the second is worse than the accounting error:

1. One test credits both. admin-settings::checkbox-reflects-enabled-state is anchored, so a single test written against the hide-slogan checkbox marks the menu-labels scenario covered too. Three scenarios here are credited by a test that never touches them.
2. A distinguishing anchor cannot be written. There is no ref that names the menu-labels scenario specifically. This is not "the author should have tagged more carefully" — the address space does not contain the thing to point at. Fixing it by renaming the scenario titles changes every existing anchor.

Since parse_spec_scenarios() already tracks current_req_slug for Format B, the fix looks available: namespace the Format A slug by its enclosing requirement the same way, or fall back to a -2 suffix on collision and report the collision rather than silently merging. Either way, a run that produces fewer distinct refs than scenarios should say so — the count discrepancy is already computable inside the checker.

⚠️ Note for anyone re-measuring: --mode report counts entries (786) while any set-based analysis counts refs (781). Two numbers that both look like "the number of scenarios" and are not the same number.