Sandbox family-grouping exhibit: the A3 deferred receipt

[gunbai-bot]

What

The A3 deferred receipt (named in #7589's body): the sandbox gains a family-grouping exhibit — one declared family rendered by the real grouping path (band_entries → workspace_entries_ul, the same fold and renderer production uses) at the three declared falsifier widths (wide 1680 / medium 1024 / narrow 480).

Each frame shows:

• the family claim stated once in the group head, with the derived "3 nodes" count;
• member rows carrying subjects only (dag_compile_clean_gate, affected_set_falsifier, regen_floor_skip_witness) — the subject swap runs through family_subject_row in presentation, not a display trick;
• an undeclared neighbour row keeping its own full headline outside the group;
• all rows routine (OpenLifecycle + SchedulingReady + NoActiveWork), so the frame also exhibits the A3 chip collapse — zero open chips, no hollow activity.

Constructed views rather than the live closing-contract family, for the same reason the A2 fixture constructs its specimen (sandbox_work_row_fixture_note): the exhibit must stay byte-stable across roadmap edits so the three width frames compare composition, not data drift.

Witness (by execution)

witness_family_fixture_claim_grouped_with_subjects — green. Pins through the serialized page: claim present, "3 nodes" present, both subjects present, family-group/family-rows classes present, neighbour headline present, and the discriminating negative: the member's full task headline (Declare the closing contract for dag_compile_clean_gate) is absent from the whole page — reverting family_subject_row to pass the headline through turns this red.

Neighbouring sandbox witnesses re-run green: witness_sandbox_gallery_and_live_render, witness_sandbox_renders_register_depths, witness_sandbox_carries_auditioner_control, witness_auditioner_blocks_emitted_with_derived_inks, witness_auditioner_missing_swatch_is_loud, witness_gallery_covers_every_wire_label, witness_gallery_drop_is_located_red.

No style changes — roadmap_css untouched, so the design-register parity digest is unmoved.

Screenshot receipt (served locally via gunbc serve, full sandbox page): the three frames render the group head once each with subject-only member rows and the ungrouped neighbour below.

🤖 Generated with Claude Code

[gunbai-bot]

Addressed review 46754 in 14b325d: the containment-only pins are replaced by three layers — (1) exact list equality on the typed band_entries output (exactly one BandFamily: claim exact, member count exactly 3, member headlines exactly the three subjects in order; then exactly one BandRow with the neighbour's full headline), which catches claim duplication, count drift, and the neighbour absorbed into the family at the structure the renderer consumes; (2) exact occurrence counts on one serialized frame — family-claim/family-count/family-group/family-rows each exactly once, the count span exactly '>3 nodes<' (so '13 nodes' cannot pass a containment check), the member's full headline exactly zero (the RED control), and the chip collapse now asserted: zero 'status status-open' and zero activity markup; (3) the page contains all three width frames verbatim, transferring the frame pins to the page. All three new fns green by execution alongside the full existing roster. Note: the current CI red on this PR is the main-side extdeps scope gate racing #7573 (session_dashboard.dag unenrolled) — routed to the owning lane; not this diff. — sent from sleek-tern-275