feat(repo): showcase-workspace E5 — release gate (8-preset dogfood + doc sweep + umbrella close-out)

[w7-mgfcode]

Sub-issue of #389 (umbrella: showcase workspace — preserve, restore, and replay showcase runs). Release gate — closes only after Foundation (E1 #390) + all Parallel epics (E2 #391, E3 #392, E4 #393) close. All four are CLOSED as of 2026-06-12 and shipped in v0.2.22, so this gate is immediately actionable.

Purpose

Close out umbrella #389 with verified evidence: a full e2e dogfood pass across all 8 ScenarioPreset cards on /showcase (green or documented expected-skip per RUNBOOKS entry 28), the deferred documentation sweep, and the umbrella success-criteria audit — so #389 closes on proof, not on per-epic merges alone.

Sub-tasks

• 8-preset dogfood matrix — run each preset card on a fresh seeded stack (Re-seed first; Reset for holiday_rush per its pinned-window semantics); record green / expected-skip / expected-fail (sparse) outcomes vs. RUNBOOKS entry 28; file issues for any deviation
• Workspace-mode dogfood — one preservation=keep run per representative preset (demo_minimal, showcase_rich); verify list/load/replay + workspace:<name> tag retrieval end-to-end (E4 dogfood covered demo_minimal only)
• RUNBOOKS.md sweep — add a Showcase-workspace section: replay-of-reset=true destructive semantics, non-unique names, workspace rows accumulating (no DELETE endpoint yet), holiday_rush union-window replay note
• DOMAIN_MODEL.md sweep — add showcase_workspace aggregate (status state machine, soft-reference created_objects, no-FK rationale) + ubiquitous-language entry for workspace vs seeder scenario
• Umbrella close-out — tick feat(api,ui): showcase workspace — preserve, restore, and replay showcase runs #389 decomposition + success-criteria checkboxes with evidence links; close feat(api,ui): showcase workspace — preserve, restore, and replay showcase runs #389
• (already satisfied — verify only) replay regression coverage: tests/test_e2e_demo.py::test_demo_replay_same_config_twice green in CI

[w7-mgfcode]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔍 E5 Release Gate — Evidence (2026-06-12)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Base: dev @ c74695c (v0.2.22). Fresh DB via DROP/CREATE + alembic upgrade head clean; GPU ollama (nvidia-smi ✅) + warmed qwen3-embedding:4b (41s cold); local uvicorn :8123 on .env config (ollama:gemma4-agent).

📋 8-preset matrix (Re-seed ✓, seed 42; UI-driven via Playwright)

Preset	Steps	Outcome
demo_minimal (keep e5-gate-minimal)	11	✅ green (reset-skip sanctioned)
retail_standard	11	✅ green
high_variance	11	✅ green
stockout_heavy	11	✅ green
new_launches	11	✅ green
sparse	11	✅ green (the matrix-sanctioned green outcome occurred; no NaN fail this run)
holiday_rush (+ Reset ✓)	11	✅ 11/11; /seeder/status = 2024-10-01..2024-12-31 pinned, no union range
showcase_rich (+ Reset ✓, keep e5-gate-rich)	24	✅ 22 pass + 2 sanctioned ⏭️ (verify V2-roots; agent_hitl_flow gemma4 no-tool) — zero ❌

📋 Workspace-mode proofs
✅ Legacy ephemeral run → green, zero workspace rows (byte-compat)
✅ e5-gate-minimal: 1 completed row; Load → config + 7 artifact links; Replay → green, 2nd distinct row
✅ e5-gate-rich: created_objects carries winning_run_id / v2_run_id / alias / 2 scenario_plan_ids / batch_id; Load → 11 artifact links; Replay (destructive, reset=true) → green, 2nd distinct row — live #146/#324 proof on the 24-step path
✅ GET /scenarios?tags=workspace:e5-gate-rich → 2 plans, tags ["showcase", price|holiday, "source:showcase", "workspace:e5-gate-rich"]
✅ 404 problem+json on unknown workspace id

🧪 Regression + gates
✅ test_demo_replay_same_config_twice — local pass (32s) + CI run 27427250799 (success @ c74695c)
✅ ruff / format / mypy --strict / pyright --strict / 1975 unit tests
✅ Frontend demo-component vitest 61/61

📝 Docs sweep: PR #403 (RUNBOOKS § Showcase workspace + DOMAIN_MODEL aggregate/UL/ER + this PRP tracked)

────────────────────────────────────────────
✅ Result: GATE PASSED — zero deviations; STOP rule never fired
────────────────────────────────────────────

[w7-mgfcode]

Gate complete — zero deviations, STOP rule never fired. Evidence in the comment above; docs sweep lands via PR #403 through normal review. Umbrella #389 ticked (11/11) and closed.