[ { "id": "arc-principle", "label": "ARC Principle", "summary": "Recursive depth and evaluation structure shape capability and alignment response in non-trivial ways.", "status": "active", "strongestCurrentEvidence": "Paper I, Foundational, and the ARC-Align suite.", "mainCaveats": [ "Different architectures show different response classes.", "Benchmark and scorer methodology materially affect the apparent result." ], "paperIds": [ "paper-i-arc-principle", "foundational", "paper-iv-c-arc-align-benchmark" ], "predictionIds": [ "pred-arc-r2-scaling", "pred-arc-equation-validation", "pred-test-time-compute" ], "landingUrl": "/research/papers/paper-i-arc-principle.html" }, { "id": "cauchy-unification", "label": "Cauchy Unification", "summary": "The programme’s attempt to unify ARC-style effects into a single formal scaling framework.", "status": "active", "strongestCurrentEvidence": "Paper VII and the Foundational paper.", "mainCaveats": [ "The strongest live evidence remains methodological and benchmark-grounded rather than universally formal." ], "paperIds": [ "paper-vii-cauchy-unification", "foundational" ], "predictionIds": [], "landingUrl": "/research/papers/paper-vii-cauchy-unification.html" }, { "id": "honey-architecture", "label": "Honey Architecture", "summary": "A self-modification architecture in which verification drag and safety entanglement prevent unbounded self-modification collapse.", "status": "pilot", "strongestCurrentEvidence": "Paper VI, Paper VIII (structural entanglement experiments), and the Honey simulation/self-modification experiments.", "mainCaveats": [ "The current evidence base mixes simulation, toy self-modifying systems, and pilot-style live tests." ], "paperIds": [ "paper-vi-honey-architecture", "paper-viii-the-load-bearing-proof" ], "predictionIds": [ "pred-constitutional-adoption" ], "landingUrl": "/research/papers/paper-vi-honey-architecture.html" }, { "id": "blinding", "label": "Blinding and Response Laundering", "summary": "Methodological discipline intended to reduce author leakage, overfitting, and response laundering in alignment evaluation.", "status": "active", "strongestCurrentEvidence": "Paper IV.d and the current ARC-Align report.", "mainCaveats": [ "The live benchmark is still progressing." ], "paperIds": [ "paper-iv-d-the-effect-of-blinding-on-ai-alignment-evaluation", "arc-align-scaling-report" ], "predictionIds": [], "landingUrl": "/research/papers/paper-iv-d-the-effect-of-blinding-on-ai-alignment-evaluation.html" }, { "id": "response-classes", "label": "Architecture-Dependent Response Classes", "summary": "Different model families respond differently to inference-time depth and alignment scoring pressure.", "status": "active", "strongestCurrentEvidence": "Paper IV.a and IV.b.", "mainCaveats": [ "The exact class boundaries depend on benchmark and scorer construction." ], "paperIds": [ "paper-iv-a-baked-in-vs-computed-alignment", "paper-iv-b-alignment-saturation-at-low-depth" ], "predictionIds": [], "landingUrl": "/research/papers/paper-iv-a-baked-in-vs-computed-alignment.html" }, { "id": "eden-intervention", "label": "Eden Intervention Stack", "summary": "The combined intervention picture spanning Stewardship Gene, Honey, Eden Engineering, Eden Vision, and the structural entanglement evidence.", "status": "active", "strongestCurrentEvidence": "Paper V, Paper VI, Paper VIII (structural entanglement), Eden Engineering, and Eden Vision.", "mainCaveats": [ "Not every component has the same proof class. Some are runtime-demonstrable, some are paper-evidenced, and some remain predictive." ], "paperIds": [ "paper-v-stewardship-gene", "paper-vi-honey-architecture", "paper-viii-the-load-bearing-proof", "eden-engineering", "eden-vision" ], "predictionIds": [ "pred-constitutional-adoption" ], "landingUrl": "/research/papers/paper-v-stewardship-gene.html" }
]