Section 7.3, paragraph 4Recall 8/15 and restoration accuracy measure different things
Informal interpretation Motivation, interpretation or scope; not a formal claim.
1 Paper text
Different metrics. The recall measures the fraction of nonzero nim-sums detected by the exact least-significant-bit channel. Restoration accuracy instead measures whether a complete selected move cancels the full nim-sum. A resemblance between the recall and a training plateau is not evidence that the same learned mechanism produces both.
Generated from paper_jmlr.tex, lines 1189–1190.
2 Explain this
Two numbers near 0.53 appear in the paper: the single-head recall and the restoration plateau near 0.54. They measure different things (detecting nonzero nim-sums through the lowest bit versus choosing a complete cancelling move), and their resemblance is not evidence of a shared mechanism.
3 How this is formalised
An interpretive caution; nothing is formalised.
| Paper anchor | par-results-p4 (website anchor; the manuscript has no label here) |
|---|---|
| Displayed as | Section 7.3, paragraph 4 |
| Claim | Recall 8/15 and restoration accuracy measure different things |
| Coverage category | Informal interpretation |
| Statement source | paper_jmlr.tex lines 1189–1190; SHA-256 of the normalised block source 9423c80bcc4f5067… |
| Explanations | explain/c-different-metrics, formal/c-different-metrics |
| Build evidence | fresh isolated rebuild, audit report |
4 Lean statement and source
No Lean statement corresponds to this claim. This is interpretation or motivation, not a formal mathematical claim.
Build evidence: fresh isolated rebuild of all 38 bundled modules with Lean (version 4.26.0, started 2026-10-10T15:21:56Z, finished 2026-10-10T15:27:00Z: 38/38 modules compiled, 1588 axiom reports, no errors. See Build for the commands and the independent audit of 2026-10-10 (dated evidence).