1. Most turns: working out which parts of the brief v1 had already done. Part I 6.5 had already dropped “returns it by value”, Part I already cited NK1333–NK1336, and NK2302 already had the wording step 2 asked for. That took about five turns of grep and reads. An off-by-one in a Python line-index edit cost one more turn.
2. Looked up that the brief could have handed me:
   • how tests/specification is run: cargo test -p nikaia --test it -- specification::, in crates/nikaia/tests/it/specification.rs;
   • the format of the ADR README’s “Changed by” table and index rows;
   • whether any registry of NK codes must match Part III (there is none, which cost one wasted grep);
   • where windows/chunks and the byte buffer sit in Part III (line 382);
   • that the brief’s “grammar inputs” use Bytes and contain no ref Array.
3. Checks I ran: specification:: twice (25 s each), pre-push twice, and the cache and disk check. Wasted: the grep for a code registry and the check that NK1137/NK1158 exist.
   • Gate: green is not backed by a gate run. The brief orders the gate, while builder.md says the driver runs it, and no Rust changed, so I ran neither cargo fmt nor clippy. The supervisor should know that the field means only my cheap checks.
4. One change: a “State after v1” line in the brief listing what v1 already wrote (6.5 list, the Part I codes, NK2302), plus the exact test command. That would have saved about eight turns. Also settle in builder.md what Gate: means for a builder, now that the driver runs the gate.

This site is open source. Improve this page.