Phase 2F — Test Fixture Infrastructure: Synthetic External Workspaces + Multi-Tenant E2E Suite

Context

Sub of Phase 2 Epic #11626 (meta-Epic #11624). Graduated from Discussion #11623 §8.

Substantial test scaffolding — synthetic external-workspace fixtures + end-to-end multi-tenant suite. Filed as separate sub because the fixture-authoring effort is non-trivial (each fixture mimics a real workspace shape).

The Problem

Phase 2 integration tests need synthetic external-workspace fixtures to validate cross-repo ingestion without requiring real external repos. We have one real repo (Neo); need to fake the rest.

The Fix

Create fixture directories under test/playwright/integration/ai/kb-ingestion/fixtures/external-workspaces/:

1. mini-neo-workspace/ — npx neo app-shaped workspace; validates default-inheritance flow (useDefaultSources: true + custom additions)
2. mini-es5-workspace/ — pure ES5 code; tests custom parser registration (ES5 needs different acorn options or babel)
3. mini-cpp-workspace/ — C++ source files + pre-parsed parsed-chunk-v1 JSONL records (client-side parser simulation)
4. mini-custom-source/ — non-standard source type (e.g., .proto files) with mock custom Source class

Each fixture includes:

• Minimal package.json (or non-JS marker file)
• Sample source files (3-5 per fixture)
• Expected chunk-output JSONL (golden file for byte-equivalence regression)
• Optional kb-config.yaml for tenant config bootstrap

Plus end-to-end multi-tenant integration suite at test/playwright/integration/ai/kb-ingestion/multi-tenant.spec.mjs:

• Push pipeline: mock hook → ingestSourceFiles → ingestion → query → tenant isolation
• Default-inheritance + default-exclusion variants
• Cross-server transport parity (server-side parse vs client-side parse byte-equivalence)
• Schema-version mismatch handling
• Tombstone flow
• Tenant path collision isolation
• Backup-record rejection
• MCP threshold → bulk path response
• Force-push reconciliation (revision-boundary)

Acceptance Criteria

• fixtures/external-workspaces/mini-neo-workspace/ exists with sample files + expected chunks
• fixtures/external-workspaces/mini-es5-workspace/ exists with sample ES5 files + custom parser config
• fixtures/external-workspaces/mini-cpp-workspace/ exists with sample C++ + pre-parsed JSONL
• fixtures/external-workspaces/mini-custom-source/ exists with mock custom Source class
• multi-tenant.spec.mjs covers all 9 integration scenarios listed above
• Each fixture has README explaining its purpose + how to extend
• CI integration: fixtures tested against current dev; regressions break build
• Performance budget: full E2E suite runs in < 5 minutes (per Neo test-substrate norms)

Out of Scope

• Real external repo integration (synthetic fixtures only for V1)
• ML-driven test generation (manual fixture authoring for V1)
• Performance benchmarking suite (separate from correctness E2E; future ticket if needed)

Related

• Parent: #11626
• Blocked-by: Phase 0/1 Epic completion (need stable schemas + registry)
• Consumed by: Phase 2A, 2B, 2C, 2D, 2E (integration test substrate)
• Discussion source: #11623 §8 test substrate (all 13 scenarios from §8)

Origin Session ID

7360e917-1733-4cdd-a6f3-5ac51c34b838

Handoff Retrieval Hints

• test/playwright/ is the test substrate root; follow existing pattern conventions
• Memory anchor: feedback_npx_bypass_test_isolation.md — use npm run test-unit not npx playwright
• Memory anchor: feedback_mcp_test_location.md — MCP server tests under test/playwright/unit/ai/mcp/server/; this sub adds integration tests under test/playwright/integration/ai/kb-ingestion/
• whitebox-e2e skill applies for any Neural Link-shaped E2E tests