1. What cost the most: reading the winnow-grammar source to see how frames are made: push_rule, the live rules stack, .item(n) in rt.rs, the code for rule calls, extern calls and template inlining. That took about 12 of my ~20 turns. The key fact, that __nikaia_list gets no frame, only showed up once I ran the probe and read the generated probe.rs. My first probe run failed because I cut the brief’s lines one off and kept the code fence: one wasted turn.

2. What the brief could have handed me:
   • Template rules (rule X<T>(item: Rule<T>)) are inlined by the macro and push no frame.
   • item N frames are pushed only for Backtrack errors, never for Cut (rt.rs ~1005/1082/1248).
   • Every members[S] failure is a Cut.

   The brief’s suggested route (“push a marked segment frame [1]”) rests on frames that can carry a run-time index, and there are none. The brief also said in item N “are not the list index” without saying they vanish on Cut.

3. Checks:
   • I ran the old binary on the probe, read the generated Rust, and ran the disk check.
   • None of them was wasted apart from the fence slip.
   • I ran no build, lower-std or tests, which was right because nothing could be built.
4. One change: a brief that says “one way needs no backend change” should rest on a probe of that way by the architect. Here that would have meant the existing probe’s generated Rust plus one list-of-objects case with a cut. Whether a run-time index survives a Cut was the whole question, and a one-line note under Locations would have settled it, or turned the brief into a backend decision before any builder started.

This site is open source. Improve this page.