oam's resolver is hickory-client where Node's is c-ares, and almost none of the work of matching node:dns was about parsing DNS. Six behaviours oam got wrong by being helpful, three bugs that were not DNS at all, and why the reference is c-ares at your Node's version rather than RFC 1035.
Five rules for reading any benchmark or conformance table, stated from the posts that already publish this project's numbers — and deliberately containing none of them. What a row's label hides, which rows the harness makes soft, and why a percentage without a denominator is not evidence.
The one JSON Lines envelope behind every oam diagnostic: what a line carries, why an agent should read it instead of scraping stderr, the two things it must not assume, and what is reserved but not shipped. Plus which engine runs under Node, Deno, Bun and oam.
Four raw-mode traps, checked against the libuv and kernel source Node actually runs: cfmakeraw is not Node's raw mode, a Windows console read already waiting keeps its old line buffering, an exit listener is not an exit path, and macOS writes back a termios bit you did not set.
An audit of what the flag was actually enforcing: 9 of 47 filesystem ops, and child_process not at all. What changed, and the one place oam is deliberately stricter than Node.
What every release signs and with what, what the installer and oam self-update verify, what each signature proves and does not, and how to check a download by hand.