Past a handful of boxes you need inventory management — but inventory systems "drift out of sync with reality" as hardware is moved, decommissioned, or repurposed, decaying into a stale snapshot of what you *think* you have. The fix is to make inventory **drive deployment** rather than describe it. When installs and configuration are generated *from* the inventory record (MAC address, role, IP, hostname), the system stops being a guess and becomes the source of truth that guides and instructs your infrastructure — kept honest because it's load-bearing. (iClassify is one such system; it integrates with Puppet for config management.) The capacity payoff: provisioning a new server becomes "enter its details, rack it, power it up," and re-purposing is "assign a new role and reboot." That's what makes the provisioning phase fast and predictable — the controllable slice of the run-out timeline — and keeps your headcount of each role accurate enough to forecast against. --- *Source: [[The Art of Capacity Planning]] (John Allspaw, O'Reilly 2008) — Ch 5 — Deployment*