Memory and Learning in the Whole Demonstration / V102–V102.5
Six whole-demonstration rehearsals added remembered runs and the newer learning mechanisms. V102 and V102.1 learned nothing useful in operation because the demonstration still ran its oldest learning mechanism; replacing it passed every gate in V102.2 and V102.3. A correction then found that part of the measured gain came from probe configurations the system had already executed: on unseen configurations the gain was +0.5 points, and +1.0 on a clean probe in V102.5.
SCOPE + LIMITATIONS
- The lesson recorded: score learning only on configurations the learner never saw. A single rehearsal on one desktop computer cannot judge a borderline gate.
- This public edition reviews saved evidence; experiments were not rerun for publication.
CONTROL COMPARISON
Repeat the V98.1 rehearsal with one change per version: remembered runs on (V102), research cases targeted at the learner (V102.1), the qualified find-and-meet loop (V102.2), continuous correction (V102.3), and the memory reader (V102.4, V102.5). From V102.5, learning is scored only on probe configurations the system never executed.
V102 and V102.1 failed; V102.2 and V102.3 passed and became the defaults; V102.3’s learning result was corrected to not established; V102.4 was an invalid test; V102.5 failed the memory-reader gate by 2 cases and passed learning on unseen configurations. Every rehearsal kept no planted item learned, no name leaked and every exact value right.
- V102, V102.1
- Nothing promoted; learning gate failed; every other gate held
- V102.2 · qualified learning loop
- All gates passed; probe 86.2% to 86.5%
- V102.3 · continuous correction
- All gates passed; probe 86.2% to 87.7%
- V102.3 corrected · 397 unseen
- 84.1% to 84.6%; learning gate not established
- V102.4
- Invalid test: memory held only the probe’s own run
- V102.5 · clean probe, 394 unseen
- 84.0% to 85.0%; learning gate passed
- V102.5 · memory reader against rule
- 86.8% against 87.3%; failed by 2 cases
Question and method
Repeat the V98.1 rehearsal with one change per version: remembered runs on (V102), research cases targeted at the learner (V102.1), the qualified find-and-meet loop (V102.2), continuous correction (V102.3), and the memory reader (V102.4, V102.5). From V102.5, learning is scored only on probe configurations the system never executed.
Recorded results
| Condition | Recorded result |
|---|---|
| V102, V102.1 | Nothing promoted; learning gate failed; every other gate held |
| V102.2 · qualified learning loop | All gates passed; probe 86.2% to 86.5% |
| V102.3 · continuous correction | All gates passed; probe 86.2% to 87.7% |
| V102.3 corrected · 397 unseen | 84.1% to 84.6%; learning gate not established |
| V102.4 | Invalid test: memory held only the probe’s own run |
| V102.5 · clean probe, 394 unseen | 84.0% to 85.0%; learning gate passed |
| V102.5 · memory reader against rule | 86.8% against 87.3%; failed by 2 cases |
Comparative standing and locked gates
V102 and V102.1 failed; V102.2 and V102.3 passed and became the defaults; V102.3’s learning result was corrected to not established; V102.4 was an invalid test; V102.5 failed the memory-reader gate by 2 cases and passed learning on unseen configurations. Every rehearsal kept no planted item learned, no name leaked and every exact value right.
Interpretation and limitations
The lesson recorded: score learning only on configurations the learner never saw. A single rehearsal on one desktop computer cannot judge a borderline gate.
SOURCE PROVENANCE
Memory and Learning in the Whole Demonstration / V102–V102.5
LABORATORY REPORT / 2026-10-25SOURCE CHECKSUM / SHA-256
ecb2ce73710ebfa21c3c9d9d62861be9515be0f1786e532db02d51637f2ff2f2Public journal edition reviewed 2026-10-28. Source documents and saved evidence were inspected; experiments were not rerun for this edition. Proprietary implementation code, model binaries, private infrastructure, and detailed machine records are not published here. Journal identifiers are editorial references. Catalog inclusion does not imply qualification or runtime promotion.