Sheaf — example coverage reports

Among the best-documented large CLIs anywhere, so it tests whether thorough docs are enough on their own. Commands are 100% documented and 82% tested, but flag-level usage falls to 29% and 441 commands are “asserted” (documented, with no test and no example) — usage is the bottleneck across 835 gaps.

xDS is how Envoy and every xDS-based control plane exchange config, so this small service is a load-bearing contract far beyond its size. The 7 covered methods are documented and tested but have no worked examples: examples are the bottleneck, the largest single gap at 54% of the debt.

fuchsia.io is the filesystem protocol at the center of Fuchsia, and this is the gallery’s two-tier flagship: a FIDL contract cross-referenced against its C++ VFS implementation. The test gap is stark: of 212 elements, 100 are documented but only 11 are tested.

The scale case: a sprawling generated CLI where docs mostly auto-exist. Sheaf surfaces the test gap that scale hides: docs cover 99 to 100%, but only 4% of flags carry any test evidence and 1,429 commands are “asserted” (documented, untested, unexemplified), 93% of the debt in one group.

pw_rpc is Pigweed’s embedded RPC system, the contract devices use to talk to hosts, common in embedded code that ships without an API doc site. Sheaf shows it barely bridged (5% of functions documented, 6% tested), leaving 125 functions “unclaimed”, 90% of the worklist in one group.

The dogfooding report, Sheaf run against its own cobra surface, and the gallery’s honesty check. A mostly-bridged surface for once: 53 of 76 elements are “completed” (concept doc, test, and example all aligned on the same call), with tests the bottleneck on the remaining 23 gaps.