1. What cost the most. The pre-flight probe took four tries:
   • A grammar entry returns a throwing result, so match on it fails; it needs catch.
   • A text(..) view must be cloned when the rule returns String.
   • A match needs an else arm.

   Next came the error columns. I spent two turns finding out that a failed guard reports the position after the element it read, and that a cut does not move it.

2. What the brief could have handed me:
   • That develop already has hex_value from the 0x/0o work, so the draft adds a second one (NK1148).
   • That DoublePart -> String cannot return the t/w view from text(..); it needs .clone().
   • The grammar entry’s call form, G::top(t) catch { .. }, for the probe.
   • The cut syntax (p => q).
   • That a guard’s error column is the position after what it read, so the expected .stdout columns are known in advance.
3. Checks:
   • Useful: the scratch probe with check --asserts, check --asserts on the package after each edit, the package tests (once failing, once with --bless, once after fmt), fmt::every_nika_file and pre-push-check.
   • Wasted: the cut probe (it changed nothing), and arguably the test run before --bless, since the new tests had no expected lines yet.
   • Skipped: clippy, since no Rust changed.
4. One change that would have saved the most: a ready-made probe skeleton in the brief (an entry call with catch, a rule returning a cloned text(..)), plus a line saying “develop already defines hex_value/is_hex; drop the draft’s copies”. That would have saved about five turns.

The report’s Tests: line was rewritten on disk: commit.nika turned \u00e9/\u0020 into the actual characters. The meaning is intact, but the escapes no longer read as written; that is a small defect in commit.nika.

This site is open source. Improve this page.