A fresh example, another lesson, and a remembered correction.

A different example can confirm a correction

The isolated artificial-word control deliberately seeded a wrong meaning for “wug”. Owen B made one mistaken action, received a contact-verified correction, then received a confirming demonstration with a different object pair. It did not need to repeat the wrong action on that second pair.

After a separate right-of lesson, the revised meaning transferred to an untrained pair with 6.95 mm endpoint error. The transfer occurred 16.99 minutes after confirmation and 15.62 minutes after the intervening lesson. The artificial fault and its correction stayed in isolated state; no semantic confirmation was copied into primary memory. This is a controlled mechanism test, not proof of naturally occurring English misconception repair.

What the timeout tells us

The first isolated transfer attempt requested a 60-second pause and hit the existing 180-second runtime limit. It committed no completed activity event, sensory-learning update or successful-action credit. The successful retry requested no additional within-session pause. The longer interval between separate runs was measured, but the 60-second isolated variant remains unqualified. No resource limit was increased.

Earlier learning after another lesson

Primary memory received an actual right-of demonstration with a third object pairing. A later left-of test retained the earlier clearance correction at 5.28 mm error after a 60-second pause, 2.87 hours after its previous same-word action and 15.10 hours after the original meaning observations. Right-of error against its earlier reference changed from 7.21 to 6.77 mm; against the newly learned mean it was 7.23 mm. This passed the report’s declared tolerances, but remains a modest single-scene result.

The caller explicitly supplied and retained the instruction, and physics paused during the wait. These are not demonstrations of autonomous intention memory or general English.

Both eyes, one target

Independent recalculation checked 180 paired geometry samples across four primary fixations. Final centring errors ranged from 0.038 to 0.229 pixels per eye, with estimated target range around 46–47 cm. Both eyes followed one shared target using the existing tracker and eye controller. The new measurements add diagnostics, not a replacement eye controller or cortical motor authority. General depth accuracy and identity understanding remain open.

Preserving the earlier record

The report records 25 primary native events and 809 visual pairs, with the earlier 22 events and all 809 earlier pairs preserved as identical prefixes. All 559 qualified visual associations retained identical decoded centres and unchanged mean earlier-pair RMSE of 0.449034 pixels. The cortical learner remains shadow-only under its 2 MiB budget. Fourteen focused tests and ten physical runs completed across staged, isolated and primary work; the extra timed-out attempt is reported separately.

This completes the report that was still pending in the 17:30 implementation update. Later selected changes to contradictory feedback, timing and cortical proposals are documented separately and do not inherit this report’s test status.