v1.0 Read paper ↗
COG / THEORY MAP
COG Theory Map
Explore the four constituents of AI engineering and their connections to ASI, transgenerational AI evolution, and the KSC series.
Select a paper to explore its role and direct relationships.
Enclosure: program constituentsSolid line: theory and methodDashed line: relevant evidence
MRAC Research and Evaluation
ASI
Transgenerational AI Evolution
AI Engineering Program
v1.0Organizes research on reliable delivery around four constituents: a target definition, a reliability relation, a reliability observation mechanism, and foundational coordinates.
Direct relationships
- Target definition: Engineering Definition of AGI · Program constituent
- Reliability relation: SCR · Program constituent
- Reliability observation mechanism: MRAC · Program constituent
- Foundational coordinates: KSC1 · Program constituent
Related concepts
Engineering definition of AGI · SCR · MRAC · KSC1
Engineering Definition of AGI
v1.1Defines engineering AGI through autonomous, reliable delivery of complex projects with independently established acceptance boundaries, distinguishing it from open-frontier innovation.
Direct relationships
- Target definition: Engineering Definition of AGI · Program constituent
- Engineering Definition of AGI ↔ KSC2 · Comparison of capability frontiers
Related concepts
Project-level capability · No continuous human steering · Verifiable delivery
KSC1
v3.3.1Provides AI engineering with coordinates for intelligent capability and human–AI collaboration, while informing research into ASI and transgenerational AI evolution.
SCR
v2.6.2Relates task search space, system coverage capability, and delivery reliability, motivating search-space governance.
MRAC
v2.1.9Observes search-space coverage through issue trajectories across audit and repair rounds, supporting repair, rollback, and decomposition. Convergence offers conditional process evidence, not a proof of correctness.
Direct relationships
- Reliability observation mechanism: MRAC · Program constituent
- SCR → MRAC · Reliability relation underpinning the mechanism
- MRAC → MRAC L1–L5 · Research agenda extension
- MRAC → MRAC Bench · Mechanism applied to evaluation methods and infrastructure
Related concepts
Second-order observation · Convergence and non-convergence · Recursive control
MRAC Bench
v1.0Applies MRAC to software engineering evaluation using audit-and-repair trajectories, convergence within budget, and resource consumption. Tasks are constructed, screened, and calibrated from real repositories and change requests. The specification-audit pilot offers preliminary observations, not proof of actual coverage or a general capability ranking.
Direct relationships
- MRAC → MRAC Bench · Mechanism applied to evaluation methods and infrastructure
Related concepts
Real-repository tasks · Specification audits · Convergence trajectories · Difficulty calibration · Budgets and resource consumption
MRAC L1–L5
v1.0Develops MRAC into a layered research agenda, distinguishing claims, conditions, evidence requirements, and limits of extrapolation.
Direct relationships
- MRAC → MRAC L1–L5 · Research agenda extension
Related concepts
Non-convergence diagnosis · Local coverage · Recursive decomposition · Project loops · Cross-domain extrapolation
KSC2
v1.2Studies structural autonomy in increasingly open tasks, extending the account of structure generation. AGI and ASI describe different capability requirements, rather than strictly sequential stages.
Direct relationships
- KSC1 → KSC2 · Framework extension
- Engineering Definition of AGI ↔ KSC2 · Comparison of capability frontiers
Related concepts
Structural openness · Structural autonomy · AACC
KSC3
v2.2Explores how structural experience passes through real-world selection, propagation, and effective training exposure to influence later models' initial structural tendencies.
Direct relationships
- KSC1 → KSC3 · Framework extension
- Public Evidence for ES–DSI → KSC3 · Relevant evidence, not causal proof
Related concepts
DSI: Distributional Structural Internalization; ES–DSI: External Selection–Distributional Structural Internalization
Public Evidence for ES–DSI
v1.1Examines capability changes at fixed model scale and evidence concerning the proposed mechanisms. It advances a testable hypothesis; current evidence does not establish the full causal chain.
Direct relationships
- Public Evidence for ES–DSI → KSC3 · Relevant evidence, not causal proof
Related concepts
Fixed model scale · Generational comparisons · Mechanism testing
The Three-Axis Model
v0.1Views AI progress through scale, paradigm, and engineering, providing a complementary perspective across the research directions.
Direct relationships
A complementary perspective, not a dependency between papers.
Related concepts
Scale axis · Paradigm axis · Engineering axis
The KSC Series: A Foundation and Two Research Extensions
The three papers form a related research series. KSC1 establishes the framework; KSC2 and KSC3 extend it toward structural autonomy and transgenerational inheritance, respectively.
Two extensions of the foundational framework
Extending structure generation S · ASI
When a task's problem representations, abstractions, and exploration directions are not given in advance, can AI autonomously form, evaluate, and revise those structures?
Extending knowledge K · AI evolution
How can structural experience pass through selection, propagation, and effective training exposure to become candidate structures that later models generate more readily?