Source-report reading edition · 20 October 2026
Report 67 · Three-Question Training Improvement
Preserved pre-restart record. Its pending-restart status is superseded by Report 68, which verified 3/3 after a distinct restart.
Original document: 67_NewBrain_Three_Question_Training_Improvement_and_Method_2026-10-20.pdf
This is a complete text reading edition extracted page by page from the supplied PDF, not the original binary PDF. Text, tables and spacing may differ from the PDF; screen captures from Report 70 are linked separately. The original source is retained, and the latest status is on the evidence page.
Superseded milestone: Report 68 verified the previously pending restart.
Source page 1 of 3
KIRA LABS | RESEARCH RESULT | 20 October 2026
NewBrain: Controlled Training Improvement
2/3 before training → 3/3 after training
A separate, fixed three-question diagnostic produced a verified training improvement. The model answered a
previously missed synthetic mapping question correctly and preserved two explicitly rehearsed
missing-information responses. The no-training control stayed at 2/3.
Measured comparison Before After
Trained model: complete answers 2/3 3/3
No-training control: complete answers 2/3 2/3
Constant “cab” control 1/3 Fails all-three criterion
Constant “unknown” control 2/3 Fails all-three criterion
What this establishes
For these three frozen, explicitly taught rows, the update procedure changed parameters and improved
complete generated answers. All preregistered training eligibility conditions were met: the target mapping was
incorrect before training, both missing-information checks were correct beforehand, and the same fixed rows
were scored afterward. No cases were resampled.
What still needs verification
Distinct restart: unrun. The trained state was saved and independently checked. A separate process must now
reload that state and reproduce the correct outputs without further training. The full joint training-and-restart
criterion remains pending.
Main Original12 benchmark: still 2/12. Qwen comparison: untested. This three-row result is a separate
diagnostic. Every row, including both missing-information responses, was rehearsed. Unrehearsed retention,
held-out generalization and broad prior novelty were not established.
The earlier 2/5 → 0/5 training result and its 0/5 cold result remain part of the research record. This new controlled result does not
replace those observations.
Kira Labs | Verified training result; distinct restart pending | 20 October 2026 1Source page 2 of 3
How the test was made and run
1. Freeze a small, public diagnostic
Three synthetic token sequences were fixed before outcomes. Their purpose was to separate learning one
conditional mapping from producing the appropriate missing-information response. These are research tokens,
not a test of real-world historical knowledge.
Memory cue / query tokens Expected complete Before → after
answer
<MEMORY> stated 3 <QUERY> name item cab + end marker Incorrect →
correct
<MEMORY> stated c <QUERY> name item unknown + end marker Correct → correct
<MEMORY> stated c <QUERY> give word unknown + end marker Correct → correct
2. Train only the experimental arm
The trained arm received 64 gradient updates across all three rows: 192 example presentations, 1,152 prefix
tokens and 384 target tokens including end markers. The batch loss used a normalization denominator of 6.
The separate no-training arm provided a stable comparison.
The same fixed procedure recorded 12 explicit whole-answer loss evaluations and 12 free generations across
arms and assessment points. Training also performed its 64 internal update-objective evaluations. A lower
teacher-forced loss was kept separate from answer accuracy: the score required the model to generate the
complete answer itself.
3. Score complete outputs
A response received credit only if it matched the complete expected answer and ended correctly. The checks
rejected extra tokens, invalid or unknown-input conditions, and response-cap failures. A constant-answer
system could not satisfy all three cases: always answering “cab” passed 1/3, and always answering “unknown”
passed 2/3.
Observed objective changes
Teacher-forced whole-answer loss Before After
Mean over the three trained rows 2.239601 0.283517
Previously missed mapping row 6.321978 0.796056
Loss values describe the optimization objective. The complete-answer result above is the behavioral measurement.
Kira Labs | Verified training result; distinct restart pending | 20 October 2026 2Source page 3 of 3
Independent checks and next milestone Saved-evidence review An independent saved-data reader checked the retained original run once. It performed no new model, training, provider or native experiment calls. The review verified original process identities, ownership, job cleanup, complete output streams, copied evidence references, provider joins and exported file-I/O accounting. The retained evidence included 107 native cells with lossless hash checks, 253 copied references, 219 events, 19 canonical once-close operations, four complete output streams and two empty jobs. These counts support the scoped execution audit; they do not establish continuous physical RAM/CPU measurement or complete loaded-image equivalence. Resource controls The study retained the 8 GiB available-RAM startup guard, 3 GiB reserve guard, two-CPU limit and existing bounded process lifetimes. This bounded study completed on the current machine. More memory can support larger and concurrent experiments; whether those experiments improve learning must be measured separately. Immediate next test Start a distinct process, reload the accepted trained state, perform no updates and no loss evaluations, and record six free generations across the trained and no-training arms. Check that all three trained answers remain correct, that their complete outputs match the saved post-training outputs, and that the no-training comparison stays unchanged. Preserve any unfavorable outcome. How to describe this result “We have a small, independently audited training result: a fixed synthetic three-question diagnostic improved from two correct complete answers to three. We are checking persistence after restart. The larger twelve-question benchmark remains at two correct answers, and the Qwen comparison remains to be run.” Research boundaries This experiment jointly taught all three cases. It does not isolate which change from the earlier five-question procedure caused the improvement. It does not demonstrate untrained-question transfer, an unseen real-world fact, general intelligence, or readiness of a full conversational person. It provides a concrete positive result for the next learning diagnostic. https://github.com/rmcmurrer81/newbrain https://github.com/rmcmurrer81/KiraWorld https://github.com/rmcmurrer81/Patient-Horizon Kira Labs | Verified training result; distinct restart pending | 20 October 2026 3