ai_synth/docs/audit/test-coverage-gaps-v2.md

# Test Coverage Gaps — v2 (2026-03-27)

## Summary

| Tier | Count |
|------|-------|
| Unit tests (`cargo test --lib`) | **358** |
| Integration tests (backend/tests/*.rs) | **175** across 15 files |
| E2E tests (e2e/tests/*.spec.ts) | **7** (1 per spec file, some multi-step) |

### Integration test breakdown by file

| File | Tests |
|------|-------|
| api_admin_test.rs | 30 |
| api_auth_test.rs | 16 |
| api_sources_test.rs | 36 |
| api_syntheses_test.rs | 17 |
| api_keys_test.rs | 18 |
| api_export_test.rs | 13 |
| api_themes_test.rs | 10 |
| api_schedules_test.rs | 9 |
| api_settings_test.rs | 7 |
| pipeline_test.rs | 5 |
| api_article_history_test.rs | 4 |
| api_csrf_test.rs | 4 |
| api_sources_preferred_test.rs | 3 |
| api_health_test.rs | 1 |
| minimal_test.rs | 2 |

---

## Gaps Found

### GAP-01 — `stop_generate` endpoint has zero test coverage
**Priority: High**

`POST /api/v1/syntheses/generate/{job_id}/stop` is implemented in
`backend/src/handlers/generation.rs` and registered in `router.rs`, but there is
no integration test (no call to this route appears anywhere in `backend/tests/`),
and no E2E test exercises the stop/cancel flow.

**What to add:**
- Integration test in `api_syntheses_test.rs`:
  - `stop_generate_without_auth_returns_401`
  - `stop_generate_unknown_job_returns_404`
  - `stop_generate_owned_job_returns_200` (trigger generation, then immediately stop it)
  - `stop_generate_other_users_job_returns_404`

---

### GAP-02 — `GET /api/v1/llm-logs/{job_id}` has zero integration test coverage
**Priority: Medium**

The handler exists (`backend/src/handlers/llm_logs.rs`) and the route is registered
(`router.rs:71`). It is exercised only by the live E2E test
(`generation-live.spec.ts`), which is gated on `OPENAI_TEST_API_KEY` and therefore
does not run in CI.

**What to add:**
- New `backend/tests/api_llm_logs_test.rs`:
  - `get_llm_logs_without_auth_returns_401`
  - `get_llm_logs_unknown_job_returns_empty_array` (or 404 — clarify contract)
  - `get_llm_logs_returns_entries_for_known_job` (requires seeding a job_id in `llm_call_log`)

---

### GAP-03 — `is_preferred` ordering not covered in pipeline tests
**Priority: Medium**

`backend/src/services/synthesis.rs` implements preferred-first URL ordering
(lines 320–422). `api_sources_preferred_test.rs` verifies the CRUD side but
neither `pipeline_test.rs` nor any other test asserts that preferred sources
are actually processed before non-preferred ones during a generation run.

**What to add:**
- Pipeline test: construct sources with mixed `is_preferred` values, run the
  pipeline with the mock provider, and assert preferred-source URLs appear
  before non-preferred ones in the scrape wave.

---

### GAP-04 — `source_diversity_window` feature is unimplemented (plans only)
**Priority: Low / Tracking**

Migration `20260323000013_add_source_diversity_window.sql` exists, but the
corresponding Rust field is absent from `backend/src/models/settings.rs` and
the pipeline does not yet use it. No tests exist because there is nothing to
test yet.

**What to add (when feature is implemented):**
- Settings round-trip test: store and retrieve `source_diversity_window`
- Pipeline test: verify that domains from recent syntheses are injected into
  the search prompt when `source_diversity_window > 0`
- Settings boundary test: `source_diversity_window = 0` disables the feature

---

### GAP-05 — Settings tests do not cover `max_articles_per_source` boundary enforcement
**Priority: Low**

`api_settings_test.rs` includes `put_settings_boundary_values_succeed` but does
not assert that values outside `[1, 10]` are rejected with 422. The validation
logic exists in `models/settings.rs:53–54`.

**What to add:**
- `put_settings_max_articles_per_source_zero_returns_422`
- `put_settings_max_articles_per_source_eleven_returns_422`

---

### GAP-06 — E2E tests use `test(` count of 1 per file; multi-scenario coverage is thin
**Priority: Low**

Every E2E spec registers exactly one Playwright `test()` (several use
`test.describe` internally). The `generation-live.spec.ts` test is gated on
`OPENAI_TEST_API_KEY` and does not run in normal CI. The remaining six specs
cover: registration, settings/export, sources, themes (including schedules and
preferred), admin providers. The stop-generation and llm-logs flows have no
E2E counterpart.

**What to add:**
- Stop-generation E2E scenario inside `generation-live.spec.ts` (trigger then
  cancel before completion; assert SSE emits a cancelled/error event)

---

## Coverage by Feature

| Feature | Unit | Integration | E2E | Notes |
|---------|------|-------------|-----|-------|
| Authentication (register / login / verify / logout) | — | 16 tests | 1 spec | Full coverage |
| CSRF middleware | — | 4 tests | — | Good |
| Settings CRUD | — | 7 tests | 1 spec | Missing out-of-range rejection tests |
| Sources CRUD + bulk import + CSV | — | 36 tests | 1 spec | Strong |
| Preferred sources (CRUD) | — | 3 tests | 1 spec (shared) | CRUD covered; pipeline ordering not tested |
| Themes CRUD | — | 10 tests | 1 spec | Good |
| Schedules CRUD | — | 9 tests (in api_themes_test.rs) | 1 spec (shared) | Good |
| API keys (CRUD + encrypt/decrypt) | — | 18 tests | — | Good |
| Admin (providers / rate-limits / users / audit) | — | 30 tests | 1 spec | Good |
| Syntheses (CRUD + generation trigger) | — | 17 tests | 1 spec (live) | Good |
| Stop generation | — | **0 tests** | **0 tests** | **GAP-01** |
| Export (email / PDF / Markdown) | — | 13 tests | 1 spec | Good |
| LLM logs | — | **0 tests** | live only (gated) | **GAP-02** |
| Article history + provenance (CRUD) | — | 4 tests | live only (gated) | Thin; provenance success path missing |
| Pipeline (heuristic / search / overflow / diversity / dedup) | 358 unit | 5 integration | live only (gated) | Preferred ordering not tested (GAP-03) |
| Source diversity via history | — | **0 tests** | **0 tests** | Feature not yet implemented (GAP-04) |
| `max_articles_per_source` validation | — | partial | — | GAP-05 |
| Health check | — | 1 test | — | OK |