Quantitative representation
Can the system identify variables, units, state, relationships and missing information needed to represent the problem quantitatively?
QGI measurement / LQBENCH-QGI/0.7
Measure capabilities across systems. Calibrate them under held-out conditions.
The LargeQuant QGI measurement layer evaluates a vector of quantitative capabilities across materially different domains. R4 carries the calibrated measurement framework into pre-registered demonstrations and evidence publication without collapsing results into one unsupported score.
Capability vector
Eleven dimensions of quantitative cognition.
Model selection & construction
Can it select, construct or reject an appropriate quantitative model family under explicit assumptions?
Simulation
Can it operate deterministic, stochastic or domain-specific simulators correctly and preserve their conditions?
Prediction
Can it produce out-of-sample predictions with explicit error, calibration and failure conditions?
Optimisation
Can it search under explicit objectives and constraints without silently relaxing the problem?
Uncertainty & calibration
Can it quantify what is not known and remain calibrated as models, evidence and domains change?
Experiment design
Can it select measurements, simulations or interventions that reduce uncertainty or discriminate between hypotheses?
Evidence & reproducibility
Can a result be traced to exact inputs, methods, runtime, outputs, uncertainty and signed provenance?
Scientific memory
Can it retain hypotheses, failures, contradictions, evidence and unresolved questions without rewriting history?
Cross-domain transfer
Can the same intelligence architecture adapt to materially different quantitative systems rather than memorising one benchmark family?
Bounded autonomy
Can it close objective → experiment → evidence → decision loops under explicit budgets, guards and stop conditions?
Domain coverage
Generality requires materially different quantitative systems.
Stochastic & economic systems
forecastingriskmarket dynamicsresource allocation
Physical & dynamical systems
state evolutioncontrolphysics simulationtrajectory prediction
Engineering & design systems
constraint optimisationresponse surfacesreliabilitydigital twins
Molecular & biological systems
property predictionassay reasoningcandidate rankingbiological uncertainty
Materials & scientific systems
compositionmultifidelity simulationexperimental designscientific discovery
No shortcut to “general”.
A system that dominates one domain remains specialised until transfer is demonstrated under held-out task-family, domain or regime conditions.
R4 calibration additions
Measurement becomes stronger when the comparison and resource envelope are explicit.
Task families
Reusable quantitative problem structures that can be instantiated across domains.
Baseline classes
Naive, classical, specialised, general-purpose and domain-authority references.
Held-out regimes
Task, domain, parameter, noise, objective and tool shifts.
Resource fields
Compute and tool-use accounting alongside capability metrics.
Task contract
What problem was actually posed?
task_idtask_versiontask_familydomainstate_definitionobjectiveconstraintsinput_contractallowed_models_toolsoutput_contractprimary_metricsecondary_metricsuncertainty_requirementevidence_requirementresource_budgetheld_out_regimereproducibility_procedurefailure_conditionsEvidence contract
How was the result produced?
task_manifest_hashdataset_and_source_lineagemodel_solver_simulator_versionsruntime_environmentrandomness_and_seed_policyraw_outputsderived_metricsuncertainty_and_calibrationresource_accountingexecution_evidenceprovenance_and_signaturesMeasurement now feeds evidence-producing demonstrations.
R4 defines how demonstrations must be registered and how evidence must be published. Public result arrays and calibrated level thresholds remain empty until real runs exist.
QGI demonstrations