Test-coverage gaps

report_coverage_gaps (MCP, opt-in --preset full) and openlore coverage-gaps (CLI) report important code that no test reaches, ranked by structural significance. It is the deterministic, graph-derived inverse of select_tests: where select_tests walks the call graph backward from a change to the tests that reach it, this walks forward from every test to the reachable set, then reports the internal code outside it.

No test execution. No coverage instrumentation. No working runtime. No LLM. Deterministic — a fixed graph yields a byte-identical report.

What it answers

• "I was told to improve coverage — where do I start?" → the ranked untested hubs.
• "Is the risky part of this change untested?" → scope to a diff (changedSymbols / diffRef).
• "Audit this module's untested surface." → scope to a region (filePattern).

How it works

1. Seeds = every test node, plus the production side of every tested_by edge (a function a test imports/asserts on is associated with that test even when it is not a direct call-graph caller).
2. Forward reachability (BFS over calls + inheritance + synthesized edges, same adjacency as select_tests) → the test-reachable set.
3. Gaps = internal code nodes (non-test, non-external/IaC, with generated / vendored / .d.ts paths excluded) minus the test-reachable set.
4. Ranking reuses the landmark-signals classifiers: load-bearing code (hub / chokepoint) ranks first, then by raw fan-in, then a stable file+name tiebreak. No composite score, no tuning constant — only labels and raw evidence.

Honesty contract

• Gaps-only — never "tested". A symbol with no reaching test definitely has a gap (the falsifiable, sound direction). The report never claims a symbol is tested or covered: structural reachability from a test means a test can reach the code, not that any test asserts its behavior. That distinction is stated in every response's soundness.caveats.
• Untested ≠ dead. A gap that is also unreachable from any liveness root is labeled alsoFlaggedDead: true — that subset is find_dead_code's domain. An untested entry point (a live, framework-invoked root) is reported as a real gap, not dead.
• A scope that matched nothing says so. A changedSymbols/diffRef/filePattern that resolves to zero in-scope symbols returns an explicit note ("nothing matched", NOT "no coverage gaps"), so a typo'd symbol never reads as "my change is fully covered".
• Scoped counts are scoped. analyzedSymbols and reachableFromTest range over the in-scope set, so a scoped call's denominator matches its scoped gaps (never the whole repo behind one gap).
• Partial test detection is disclosed. When some languages have no detected test files, the report names only those languages (not the well-tested ones) and flags that their gaps may be over-reported. With no tests detected at all, testDetection: "none" says the surface looks untested because detection found nothing — not because the code is genuinely untested.
• Over-report is the safe direction. Dynamic dispatch, reflection, and DI can make a symbol reachable-by-test through an edge static analysis cannot see; such a symbol may be falsely reported as a gap. Pass directResolvedOnly: true to ignore synthesized edges and get a stricter (more gaps, more certain) report; the alsoFlaggedDead label is computed on the same edge basis.

Distinct from get_test_coverage

get_test_coverage is a spec/scenario tag-based report (which OpenSpec scenarios have a tagged test, via generate_tests). report_coverage_gaps is pure call-graph structural reachability. They answer different questions and do not overlap.

CLI

openlore coverage-gaps                              # whole repo, ranked
openlore coverage-gaps --max 50                     # cap the list (default 100, capped 500)
openlore coverage-gaps --file-pattern src/core/auth # region scope
openlore coverage-gaps --base main                  # diff scope: gaps among symbols changed vs main
openlore coverage-gaps --symbols parseConfig,login  # diff scope: only these changed symbols
openlore coverage-gaps --json                       # machine-readable (stable shape) for CI / an orchestrator

Read-only and advisory — it is a report and never blocks.

Scope (single-repo, by design)

Cross-repo / federated coverage gaps are deliberately out of scope (single-repo first); the tool does not take the federation flag. Executed-line/branch coverage (a runtime concern) and test quality (does the reaching test actually assert anything) are also out of scope — this surfaces where a gap is, deterministically, not whether an existing test is good.