`test/vdrive-walk-order.test.js` imported fs, path and os as bare specifiers. Bare cannot resolve those, so `npm run test:bare` died at module resolution before a single test ran: ModuleError: MODULE_NOT_FOUND: Cannot find module 'fs' imported from .../test/vdrive-walk-order.test.js It was not one line but three, and it arrived with 1847ca0 — my own commit, so the vdrive walk-order test has never actually run under Bare. This file was the only offender matched by the test:bare glob (`test/*.test.js`), which is why nothing caught it. `cross-machine-autobase.js` and `keet-pattern-e2e.js` use bare specifiers too but are not `*.test.js`, so the glob skips them and they abort nothing. The package already declares the mapping — `#node:fs` -> bare-fs / fs, and the same for path and os — and the rest of the suite uses it consistently (94 `#node:path`, 89 `#node:os`, 88 `#node:fs`). Nothing here needed inventing. Verified: 4/4 tests, 11/11 asserts under `bare ./node_modules/.bin/brittle`, and no bare node specifier remains anywhere in `test/*.test.js`. The version bump is not cosmetic. ReleaseRegistry's apply does an unconditional `view.put` on `release:<name>:<version>`, so publishing a changed tree under an existing version overwrites the entry in place, and `latest:` only advances on timestamp — anyone already installed on 0.1.234 would get no signal to re-fetch. Publishing pgv3 is a manual step, so leaving 0.1.234 here would arm that trap for whoever publishes next. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>