Tick: took the one entry in the findings queue. Verifying it before acting reversed half of it. The queued line carried a second-hand report that the CLI had been renamed to `tg` and noted tick.md names `tang` in thirteen places plus both skill shims. It correctly declined to guess. The machine contradicts the report: `which tg` finds nothing, `which tang` resolves. The proposed sweep would have rewritten every documented command across tick.md and the two skill shims to a binary that does not exist. No sweep was performed. Defense: tick.md owns the issue-filing workflow, so the correction lands there. It now records that `tang` is current, dates the contradiction, and tells the next tick to confirm with `which` rather than sweep on a rumor — the queue's re-verify rule is what caught this, and the page should carry the same instruction so it is caught again. The path half was real but imprecisely stated. The chain is `~/.bun/bin/tang` -> `~/.bun/install/global/node_modules/tangled-cli` (a bun link back to the fork) -> `~/code/tangled-cli/dist/src/index.js`. So tick.md's intent was right — the fork is the live source and editing it changes what runs — while the entry point it named was wrong. `dist/index.js` exists in the fork, is a stale Jul-9 sibling, and never executes, so the old sentence would have sent anyone debugging the CLI to a plausible-looking dead file. Both hops, the real entry point, and `readlink -f "$(which tang)"` as the check are now documented. Queue line deleted, coverage row added, session log at wiki/log/2026-07-27-tangled-cli-path.md. `tang` is still unauthenticated (browser OAuth, human-only), so no issue was filed and the decision-made queue could not be read at intake.