Brain has completed its first accepted learning, reload, cognition, retention, and benchmark cycle as one locally controlled system. A compact Brain-owned cortex now works with typed cognition, governed memory, evidence, and exact QSA state computation without a donor LLM at runtime.
Brain does not ask one neural weight matrix to store, retrieve, reason, verify, remember, and govern everything. The cortex works inside a controlled architecture whose other parts remain explicit and testable.
Accepted trained state
A compact Brain-owned cortex now completes the full learning cycle.
The accepted build completed 1,024 optimizer steps across a fixed budget of 29,930,048 Brain-owned mutable scalars. The final candidate reloads from its own checkpoint, operates without the temporary teacher at runtime, and remains recoverable under the project checkpoint authority.
01Brain-owned state
02Bounded learning
03Acceptance gates
04Exact reload
Verified or implemented
29,929,950 scalars changed during accepted training
64 of 64 whole-Brain acceptance cases passed
106 of 106 retention checks passed
No donor model required at runtime
LanguageUnderstand the request
SemanticsBind roles and constraints
EvidenceRecall with provenance
CognitionEvaluate candidate states
AuthorityCheck truth and action limits
RecordAnswer, retain, or roll back
Benchmark observatory
Fresh gains at a fixed neural budget
These are bounded benchmark records, not a claim that Brain is broadly equivalent to one conventional model size. Select a result to see the cohort, controls, and exact point where the claim stops.
200 fresh ARC questions
Grounded evidence doubled the fresh ARC result.
Inherited Brain43 / 200
Measured result92 / 200
Measured change+49 correct / +24.5 points
What changed
Candidate-specific evidence was evaluated through question anchoring, candidate relation, and grounded support before the answer was selected.
No retraining
No target labels in the reasoning path
Protected Brain state unchanged
No donor model at runtime
QSA and the information-identical classical product selected the same answers. This proves a strong architecture gain, not a uniquely quantum accuracy advantage.
Conventional model context
Three tasks / published zero-shot references
Current profile estimate~500M class
Several-hundred-million-parameter performance from a ~30M Brain.
Closest three-task referenceQwen1.5-500M
Near Brain's current ARC-Easy, PIQA, and WinoGrande estimates.
Parameter ratio16.7x
Qwen1.5-500M carries roughly 16.7 times Brain's parameter count.
SmolLM2-135M reports 43.9% ARC average, 68.4% PIQA, 51.3% WinoGrande, and 42.1% HellaSwag. Brain is about 4.5 times smaller; its current profile is stronger on ARC and WinoGrande and competitive on PIQA. Brain's recorded HellaSwag route gained 6.25 percentage points over its own inherited state; that gain is not presented as a protocol-matched SmolLM2 score.
Evidence ledger
Proof beside the boundary, not buried beneath it.
Brain's current result is stronger than the previous public page, but it is still research. Every record states what was measured, what was protected, and what the test does not establish.
The live checkpoint, private corpus, model topology, memory contents, thresholds, and security controls remain protected.
ARC grounded evidence V33Fresh target-sealed result
43 / 200 became 92 / 200 with no retraining, no target labels in the reasoning path, no donor runtime, and no protected-state mutation.
The information-identical classical control tied QSA, so the measured gain belongs to the cognitive architecture rather than a claimed quantum advantage.
Wino object-state V35Fresh train-holdout result
523 / 1,024 became 550 / 1,024. The exact object route answered 26 / 28 resolved rows; actual overrides produced 14 gains and 2 losses.
This is not an official validation score and is not directly protocol-matched to external model cards.
Quantum attributionControl retained
Public QSA executes and certifies admitted structured states, and benchmark records preserve information-matched controls and fallback status.
No current Brain benchmark establishes universal quantum advantage, physical-QPU advantage, or broad superiority to conventional LLMs.
Current record
From sealed learning to measurable cognition
The important change is not a new visual label. Brain has crossed from an integration project into an accepted end-to-end system with a continuing, falsifiable benchmark program.
Aug 2001
Structured procedure layer sealed
A single fresh 2,120-row blind closed the accepted structured cognition fallback with its own immutable evidence record.
Aug 2102
Brain completed accepted training
The Brain-owned state completed 1,024 optimizer steps, reloaded exactly, passed 64 / 64 whole-Brain acceptance, and retained 106 / 106 prior behaviors.
Aug 2203
Fixed-parameter benchmark gains replicated
Fresh ARC, PIQA, HellaSwag, and WinoGrande routes improved without increasing the neural parameter budget or reconnecting the teacher runtime.
Now04
Semantic world grounding
Current work is compiling evidence into typed entities, roles, effects, constraints, causal links, uncertainty, and candidate-specific possible worlds.
Brain / QELM / QSA
Connected systems, separate authorities
QELM is the language-model research foundation. QSA supplies exact quantum-state execution. Brain owns language, memory, evidence, learning, checkpoint adoption, and the final decision path.
Broader protocol-matched external comparison and independent reproduction packages
Evaluation contract
How a Brain result becomes evidence
A finished run is not automatically a capability gain, and a QSA-backed gain is not automatically quantum advantage. The record follows the exact test that was performed.
01
Freeze
Bind the method, runtime, knowledge identities, parameter state, and complete predictions before target access.
02
Score
Open the target vault once, compare against the inherited Brain answer, and record gains, losses, ties, and coverage.
03
Control
Run information-matched classical, dephased, reset, or fallback controls appropriate to the mechanism being tested.
04
Protect
Verify accepted Brain state, protected checkpoints, donor disconnection, and production authority remain unchanged.
05
Retain
Keep positive, neutral, failed, and revoked paths in the same evidence history so a weak route cannot return as a claim.
