Δ

AGI Definition & Capability Goalpost Auditor

Empirical diagnostic evaluating AI claims vs formal cognitive architectures

Archetype Presets:

Cognitive Radar & Definition Goalpost

Blue = System | Amber = Threshold
Goalpost Definition Regime: Strict General Cognitive Architecture
Strict Cognitive (L5) Economic Parity Narrow Task Doublespeak
Strict General Cognitive Architecture (Legg-Hutter / DeepMind L5): Requires continuous production weight adaptation, self-generated intrinsic agency, complete cross-domain transfer, and epistemic self-verification.

Foundational Capability Axes

0 (None) → 100 (Human/Superhuman General)
0

Weights continuously adapt from live runtime experience without requiring offline batch pre-training or frozen checkpoints.

12

Formulates, prioritizes, and pursues novel intrinsic objectives without waiting for an external human prompt.

45

Generalizes abstract models from one domain (e.g. physics) to solve unencountered problems in another (e.g. negotiation).

38

Zero stochastic hallucination drift on verified logical facts across repeated queries.

24

Executes complex multi-step workflows over weeks/months with measurable economic yield and autonomous error-recovery.

15

Accurate internal discernment of what it knows, calibrated confidence intervals, and autonomous refusal to hallucinate.

Empirical Discrepancy & Semantic Drift Telemetry

Level 1 Narrow Task Emulation
Overall General Score
22.3
Weighted across 6 formal cognitive axes
Narrow Competence Score
78.5
Task benchmark completion capability
Semantic Dilution Index
71.6%
Goalpost shift vs formal AGI criteria
Current Threshold Status
Sub-Threshold Gap
Failed on 6/6 formal requirements
Diagnostic Audit Verdict

Claiming AGI based on narrow task execution reflects a 71.6% semantic dilution compared to formal continual-learning cognitive criteria.

Dimensional Architectural Audit

Axis Observed Required Architectural State & Discrepancy Analysis Status

AGI Claim Audit Report

Deterministic technical dossier summarizing architectural gaps, semantic dilution, and threshold compliance. 


    


    

Six assigned axes and chosen thresholds determine a local claim score
Read the explanation
The enterprise fixture assigns six general capability scores: zero, twelve, forty five, thirty eight, twenty four and fifteen. Their sum is one hundred thirty four, divided by six and rounded to one decimal gives twenty two point three. At two pixels per summed score the first three total fifty seven, last three seventy seven and all six one hundred thirty four, measuring one hundred fourteen, one hundred fifty four and two hundred sixty eight. These are manually assigned fixture values, not benchmark observations or an independently evaluated model. Equal averaging gives each axis identical influence regardless of measurement uncertainty. The narrow competence baseline is seventy eight point five. Subtracting rounded general score twenty two point three leaves fifty six point two. At four pixels per score narrow, general and gap measure three hundred fourteen, eighty nine point two and two hundred twenty four point eight. Dividing the gap by the narrow baseline yields about seventy one point six percent, called semantic dilution. That is a relative arithmetic gap between chosen scales, not a measurable change in language meaning or a probability of AGI. A special fixture override also forces its classification and verdict text independently of the generic classification ladder. Under the strict preset all six assigned axes lie below their selected thresholds. Passed axes zero, failed six and total six measure zero, two hundred forty and two hundred forty at forty pixels per axis. Moving the regime changes comparison thresholds without changing candidate capabilities. The chosen classification and thresholds do not prove a scientific or regulatory definition of general intelligence. Offline graph rendering lacks its external library; native controls and report export may still work depending on listener order, which must be observed rather than inferred from the radar title. No model inference or evaluation service is called.


  


  


Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.