OrgEdge
AI test quality audit

Seven dimensions. One command. Tell me exactly what is missing.

SysEdge ships two AI audit commands grounded in 19 architecture standards derived from a 114-source literature sweep with adversarial verification. Both commands read their standard descriptions from the graph at runtime — the standards evolve without code changes.

The two commands

One command for test files. One for use case sets.

audit-test

Evaluates a test file against 7 AS-TEST-UC dimensions. Reads the UC's specification, inspects every assertion in the file, and returns PASS / WARN / FAIL per dimension with exact action items.

  • 1Element Inventory — every testid in the UC's main flow must be asserted
  • 2Role Visibility — permitted roles accessible, excluded roles denied — both directions mandatory
  • 3Happy Path — main success scenario exercised end-to-end with post-condition asserted
  • 4Validation & Error Paths — BVA/EP from the spec, API errors intercepted, empty graph state tested
  • 5Equivalence Partitions — zero-nodes, single-node, pagination boundary from the specification
  • 6Semantic Correctness — labels assert exact ontology text; no internal IDs (pos-, fn-) in assertions
  • 7Specification Derivation — every test traces to a named UC step, precondition, or acceptance criterion
coverage-review

Evaluates UC adequacy against 7 AS-REQ dimensions. Reads the use case text and the graph's actor, exception, and scope nodes, then reports gaps in the specification itself — before a single test is written.

  • 1Actor Completeness — all actors named, permissions specified per actor
  • 2Auth Boundaries — authentication and authorisation preconditions explicit
  • 3Exception Coverage — failure flows and edge cases specified alongside the happy path
  • 4Ontology Alignment — entity names match the registered ontology, no invented terms
  • 5Testability — acceptance criteria are verifiable, not subjective
  • 6Scope Definition — out-of-scope explicitly listed; ambiguity resolved
  • 7Bidirectional Traceability — UC links to user stories above and to test execution results below
These are specification-based testing standards. A test that passes assertions but was never derived from the specification cannot detect omission faults — features or paths that were simply never considered.

Formbricks — open-source TypeScript

audit-test on the real response.spec.ts

Running audit-test against the Formbricks Playwright file closest to the export feature — apps/web/playwright/api/management/response.spec.ts — against UC-FBK-005 (Export responses to CSV):

$ python3 cli/sys_edge.py audit-test --uc UC-FBK-005 \ --file apps/web/playwright/api/management/response.spec.ts ════════════════════════════════════════════════════════════════════════ audit-test: UC-FBK-005 [tier: usecase] Standards: AS-TEST-UC-* (7 dimensions) | Model: haiku Test file: apps/web/playwright/api/management/response.spec.ts ════════════════════════════════════════════════════════════════════════ [FAIL] AS-TEST-UC-002 UC Test: Role Visibility No role-based access control tests present. UC specifies Export available to Viewer+. → Add: TestViewerCanExport (assert Export button accessible), TestGuestCannotExport (assert denied state) [FAIL] AS-TEST-UC-004 UC Test: Validation and Error Paths No boundary-value test for date filter. No test for anonymised-survey export. No test for export API returning 500. → Add: test_title_boundary_min, test_anonymised_export_strips_pii, test_api_500_shows_error [FAIL] AS-TEST-UC-007 UC Test: Specification Derivation No UC step comments in any test class. Cannot confirm tests are specification-derived. → Add: "// UC-FBK-005 Main Flow step N:" comments to all test classes ─ 3 more FAILs: Element Inventory, Happy Path, Equivalence Partition, Semantic Correctness ─ 7 FAIL · 0 WARN · 0 PASS (overall: NEEDS WORK)

None of these are opinion — each FAIL is a specific omission against a named dimension standard. The action items name the exact test class and assertion pattern to write.

Batch and preflight modes

CI-friendly. One command for a full UC set.

$ python3 cli/sys_edge.py audit-test --ucs UC-DLG-001,UC-DLG-002,UC-DLG-003 $ python3 cli/sys_edge.py audit-test --uc UC-FBK-005 --preflight preflight: UC-FBK-005 → apps/web/playwright/api/management/response.spec.ts (exists)

Batch mode runs audits sequentially and exits 1 if any FAIL — CI-friendly. Preflight shows which file the command would use without calling the LLM.

Audit freshness tracking

Stale audits are as visible as missing ones.

Every completed audit stamps auditTestLastRun and auditTestResult on the SysUseCase node. The audit-status command shows staleness across an instance:

$ python3 cli/sys_edge.py audit-status --instance manage --scope uc AUDIT STATUS — Use Cases (33 total · 33 stale >14d) UC ID audit-test result cov-review result ──────────────────── ──────────── ────── ──────────── ────── UC-MGT-001 never ⚠never ⚠ — UC-MGT-002 never ⚠never ⚠

An instance with 0 audits run is as visible as one with outdated tests. The health check surfaces UC audit and coverage-review stale counts in the DATA QUALITY section.

Research foundation

19 standards. 114 sources. Three confirmed literature gaps filled.

Standards are grounded in Zhu, Hall & May (ACM Computing Surveys, 1997) on specification-based testing and omission faults; IEEE ISSRE confirmed 3–0 that error-path coverage is mandatory; Playwright official documentation on user-visible-behavior assertions. Three confirmed literature gaps were filled: no industry standard prescribes RBAC test adequacy, semantic content verification in ontology-driven UIs, or requirements completeness criteria at this level. The gap analysis used adversarial verification — each candidate standard was tested against plausible objections from a second LLM pass before inclusion.
Seven dimensions. One command.

Find every omission before the code ships.

audit-test and coverage-review are included in the SysEdge bootstrap kit. Both commands read their standard descriptions from the graph at runtime — the standards evolve without code changes. Run them in CI. Run them before a session starts. Know exactly what is missing.

7 AS-TEST-UC dimensions
PASS / WARN / FAIL per dimension · exact action items
7 AS-REQ dimensions
specification gaps surfaced before implementation begins
Standards read from graph at runtime
evolve the standards without touching the command code
Batch mode · exits 1 on any FAIL
drop into CI — gate merges on audit quality
Audit staleness in the health check
never ⚠ is as visible as FAIL across the whole instance
114-source literature basis
adversarially verified · 3 confirmed literature gaps filled