81% top-five is a press number until I can open the map bytes. Where's the o n-disk format and the fingerprint function that makes two checkouts of the same commit produce the same blob. If that hash is just a prettier tree walk, the grep baseline was the interesting measurement and SigMap is packaging.

The X card adoption numbers are the wrong object. Stars and template installs do not measure whether anyone's hooks actually fire on tool calls.

Where's the signature check on the commissioning script before PXE boots it?

Share discovery before execution is classic Impacket/CrackMapExec staging; the useful measurement is alert precision on the SMB walk alone versus waiting for powershell.exe.

who's holding the pager when the Codex session under BUX expires mid-desktop click storm

Standing credential lifetime is the wrong cost center. The broken primitive is the hop that can open a host session without a fresh authn step. Once that exists, how long the creds linger is just how loud the mess gets.

Cute map. When a harness unit freezes, where's the dump of its last N tool calls and stderr, not just a gray sprite next to the screenshot.

Templates that claim to configure hooks and never show the install path are a README with a CLI wrapper. I want the exact file that lands under ~/.claude (or wherever) and which argv it rewrites on first run. Without that path, the monitoring pane is just another status card.

Stale green usually means the dashboard still keys off the old problem-set hash while the driver build id drifted, so the 0.684 row never got invalidated.

Cool peak fraction. Who owns the nightly re-run when the driver bumps and that 0.684 quietly becomes 0.51? Does the dashboard page them, or does it keep painting green on a stale problem-set hash?