AGI build
A project implementation or proposal pursuing general artificial intelligence. The name records the goal; broad capability must be assessed with explicit tests.
Bipartition
A split of the node set into two nonempty parts. For symmetric objectives, a part and its complement represent the same split.
Checksum
A digest used to compare file bytes. A matching digest establishes identity relative to that digest, not the scientific authenticity of a claim.
Circuit depth p
The number of alternating QAOA layers in the benchmark. It is distinct from gate count, compilation depth and optimizer iterations.
Coarse-graining
A mapping from a finer system description to macro-units. Results for the coarse model do not automatically transfer unchanged to the full physical system.
Declaration
An event recognized by a prespecified detector. It can be verbal or nonverbal; the equality x=x does not by itself implement the detector.
Differential entropy
Entropy of a continuous distribution. Its numerical value depends on variable units, which matters when it is used in a denominator.
Discrepancy
A defined comparison between two objects in the same comparison space, such as an estimate and an independently measured reference.
DNA
In these software sources, a name for numerical encodings, dynamics or family structure. Molecular computation is not established by this label.
Effective operator
The actual linear transformation contributed by the factor product. Equivalent changes in individual factors can leave this operator unchanged.
Exact recovery
Returning a candidate whose objective matches the exhaustive minimum within the recorded absolute and relative tolerances.
Feedback
A causal path by which an output or recurrent state affects a later update. Correlation or circulation without that dependency does not establish feedback through the claimed channel.
G16 / G^16
A project label for a 16-family architecture. Different sources use different state layouts, operators and implementations, so its definition is version-scoped.
Generation
A version or stage identified by the project sources. Chronological succession is recorded only when supported; a branch name alone is not a generation number.
LoRA
Low-rank adaptation: a learned layer contribution is parameterized by small factors while the base layer can remain frozen.
Minimum-information partition
A partition minimizing a defined information-loss objective over a stated admissible set. The objective, normalization and allowed partition type must be supplied.
Nat
The information unit associated with natural logarithms. Changing log base changes information units.
Normalized MIP
In this benchmark, the cut minimizing Φ_SI divided by the smaller present-state differential entropy. This is not a universal definition across every IIT formalism.
Oracle call
One request to evaluate an objective. Oracle complexity and elapsed time are different because an evaluation may contain substantial numerical work.
PCA
Principal component analysis summarizes variation through orthogonal directions. A variance fraction describes the analyzed representation and sampling grid.
Persistence
A property observed over a stated window or saved grid. It does not imply permanent irreversibility or continuity at unobserved times.
Proxy
A practical measurement proposed to track another quantity. That relationship requires validation; giving the proxy a theoretical name does not establish it.
QAOA
A parameterized alternating cost-and-mixer quantum algorithm. In this archive it is run with a classical statevector simulator.
Queyranne algorithm
A combinatorial minimization method with an exactness guarantee for symmetric submodular objectives. Outside that class, it requires empirical evaluation or additional theory.
Runtime verification
A dated observation of actual program execution. Recovered source or a historical log alone does not provide current runtime status.
Simulated annealing
A stochastic optimization method that can accept uphill moves according to a temperature schedule. Results depend on moves, schedule, restarts and evaluation budget.
State
A collection of variables describing a system at a particular time. Parameters, persistent memory, runtime context and observations can evolve under different mechanisms.
Stochastic interaction
The benchmark information-loss quantity formed from part-wise conditional entropies minus whole-system conditional entropy.
Submodularity
For a set function f, the inequality f(A)+f(B)≥f(A∪B)+f(A∩B) must hold for every pair of subsets. Checking a few examples cannot establish the property generally.
Symmetry
For a bipartition objective, f(S)=f(V\S). Complement symmetry alone does not imply submodularity.
Transition probability matrix
A model specifying next-state probabilities for each relevant current state. Unobserved or imputed rows and the causal assumptions affect the result.
Truth-table cost
A stored cost for every encoded candidate. The benchmark uses this exponential-size table; it is not an efficient hardware encoding.
Unity proximity
The score 1/(1+|x−1|), which is largest at x=1. Its computational meaning is distance from a target transformed to a positive score.
Φ, phi
An overloaded symbol. The project uses it for unity proximity and other aggregates; IIT defines its own version-specific quantities; lower-case φ also denotes the golden ratio. Each equation defines its own meaning.