Raise test coverage to 85% while retaining the 80% CI floor

[apstndb]

Goal

Raise the existing full-suite statement coverage to at least 85% on main, while retaining the CI acceptance floor at 80%. The previous main CI result was 80.6%.

Scope

Add meaningful regression tests in small, independently reviewable PRs. Start with uncovered completion logic, offline feature behavior, and statement/execution paths. Prefer existing fixtures and in-process transports. Minimal production test seams require a concrete justification.

Do not change .octocov.yml, coverage exclusions, generated-file filtering, the measurement command, dependencies, or unrelated production behavior. Do not inflate coverage with assertion-free calls or broad refactoring.

Acceptance

• make test-coverage and go tool cover -func=tmp/coverage.out report at least 85% on the integrated source; verify the main CI result too.
• .octocov.yml continues to specify acceptable: 80%.
• Each PR passes make check, relevant focused race checks, current-head review, and all CI checks.
• Tests assert outputs, state changes, request parameters, and error contracts as appropriate; they do not introduce network/credential dependencies into offline tests.
• Preserve existing test isolation, especially global logger/configuration state.

Use Refs in implementation PRs until the combined result is verified. Record final coverage evidence before closing through the final PR.

[apstndb]

Candidate pool and stopping rule

Prepare more independent candidates than are needed for 85%; do not require every candidate to finish. Cursor owns prioritization, lane assignment and combined/post-merge measurement. Reconcile active ownership before splitting an existing lane. Start roughly 3-5 non-overlapping lanes when useful, with one reviewable PR per bounded change.

Candidate	Exclusive production area and corresponding tests	Useful behavior to investigate
A	internal/mycli/fuzzy_finder.go	Candidate queries, cache invalidation, variable values, selection/error handling
B	internal/mycli/feature/llm/	Offline configuration, prompts, fake-model execution, cleanup/error contracts
C	internal/mycli/feature/bigquery/ and feature/cql/	Validation, result conversion and fake-client error behavior
D	internal/mycli/execute_ddl.go	Async results, cancellation/wait failures, buffering and operation handling
E	internal/mycli/statements_explain_describe.go	Explain/describe errors, plan export and formatting contracts
F	internal/mycli/cli_readline.go, cli.go, meta_commands.go	Input/meta-command behavior, EOF/errors, output capture and cleanup
G	internal/mycli/execute_sql.go, row_processor.go	Result/row processing, empty results and execution errors
H	internal/mycli/statements_transaction.go, session.go, transaction_manager.go	Existing fake/emulator-supported transaction/session error paths; lower priority if setup or concurrency cost is high

These are candidate areas, not permission for broad production edits. The original three lanes may already own combinations A, B+C and D+E; keep that ownership and use F/G/H as additional independent choices rather than editing their files concurrently. Prefer a high-value subset over large test scaffolding. Each worker must establish its baseline and likely attainable benefit; uncovered statements are an upper bound, not a promised gain.

Use the existing CI make test-coverage output, coverage HTML artifact and octocov report tied to the exact head. Cursor collects the source SHA, run URL, executed step, scope and percentage, including the final main run after merges. Codex inspects that evidence instead of repeating identical local coverage measurements. Keep mandatory pre-push validation gates.

Stop starting new candidates once integrated main CI reports at least 85%. Preserve unstarted candidates for later. For in-flight surplus work, report its state and retain recoverable commits/patches; stop safely without deleting work. Keep .octocov.yml at acceptable: 80% throughout.

[apstndb]

Completed the coverage wave through #942, #943, #945, #944, and #946.

Final main commit b84fe5ed04aef890e3cbeae66f8a238def11c93e passed CI run 34706208619. The full make test-coverage / go tool cover -func total is 86.4%, up from the initial local full-suite baseline of 80.7%. This is the Go tool total, separate from octocov's metric.

The existing 80% CI floor, coverage scope, workflow, dependencies, and production code are unchanged. All 16 changed files are tests. Review fixes covered deterministic DDL cancellation, validation/iterator assertions, SDK environment isolation, and terminal/output isolation.

The target is met: stop starting new coverage candidates. Retain the remaining candidate pool and draft #947, whose timing-dependent follow-up has a review finding and is deferred. No worktree or branch cleanup is included.