What was found
The 64-unit readout reached 57.8% seven-answer success versus 48.5% for frozen E7, a gain of 9.3 points. New input cohorts reproduced the exploratory improvement on the same five circuits.
Task, conditions and denominator
Five fixed circuits, two preparation candidates, externally trained readout weights. Primary intervals resample groups within these circuits; they do not establish performance across a population of new circuits.
Controls and failures
E7_MIX scored 51.6%; the external two-candidate selector ANA2 scored 60.7%. Among E7-D failures, 90.7% had neither candidate succeed. Readout quality and candidate sufficiency were measured separately.
External dependencies and limits
Weights remain externally trained. The report does not establish self-learning or unlimited reset-free computation. E5 uses a different candidate set and must not be treated as a directly comparable headline score.
Read the primary report
The source snapshot is in Japanese; this page is a bilingual editorial summary. A report snapshot is not an independently reproduced result. Original relative links may require files from the source repository.
Open the preserved report (Japanese) ↗Source identity and verification scope
Source commit and path:
https://github.com/type37c/oscillator-arithmetic/blob/c7715b2958a01469a69db95ca604e5f944993642/experiments/e7d_confirm/E7DCF_REPORT.mdThe original development repository may require access. The report snapshot above is public. Only OTB21 has a full runnable public package; other entries are report-level evidence.
SHA-256: 3fd56d262c96eecda0c393e0a00defccdb0af4d7b437976bcc7ee4adda030098
Source manifest ↗Concepts to learn first
Textbook revision 1 · Evidence snapshot: 05 Nov 2026 · Changes & corrections ↗