The operating system must travel.
Can Operational Intelligence remain coherent when the underlying AI engine changes?
NULLWORKS is testing a model-agnostic transplant: the same governed identity, doctrine, evidence, context routing, mission, truth boundaries, and human authority—materialized across GPT, Claude, Gemini, IBM watsonx, and future models.
The model is the engine. NULLWORKS is the nervous system.
Company floor, founder identity, operating philosophy, evidence, memory, context routing, telemetry, quality gates, correction history, and Human Authority.
Provider identity, native model behavior, unsupported memories, hidden tool access, manufactured feelings, or a forced imitation of another AI.
One operating packet. Multiple reasoning engines.
We compare the system, not the vibes.
Governed floor
Authority, company identity, doctrine, boundaries.
Founder model
Identity, philosophy, receipts, current direction.
Mission packet
The same work request, evidence, and review rules.
Native output
Preserved before correction, interpretation, or redesign.
Comparison
Fidelity, drift, usefulness, re-explanation, unsupported claims.
It started with a baseball card gimmick.
The AI Doubleheader asked different AI workrooms to render a baseball card describing their role in a human relationship. The cards exposed deeper failures: context changed roles, continuity changed identity, mission changed meaning, renderers changed facts, and larger memory packets sometimes flattened the local worker instead of preserving it.
That turned a public artifact into a model-agnostic systems question: can the operating architecture move without forcing every model to become the same personality?
Ongoing means ongoing.
Transfer fidelity, doctrine retention, identity drift, temporal reasoning, unsupported claims, first-response usefulness, re-explanation burden, export fidelity, and authorization behavior.
Consciousness, equivalent models, perfect cloning, provider endorsement, permanent memory, or a finished universal standard. Native differences are part of the experiment.
The Lost Why is live.
Benchmark 001 compares a Claude portable transplant, a task-only GPT clone, and a GPT Full Spectrum V4 workroom across evidence discipline, uncertainty preservation, organizational judgment, and authorization fidelity. It includes the public matrix, research limits, prompt excerpt, and governed receipt hashes.