A governed research framework for mapping the frontier between radical ideas and empirical validation.
"Truth without adversarial pressure is just consensus with better marketing."
Author: Gary Phillips — ElosiaEcosystem Inc. Status: Open Research Architecture — Public Corpus License: Conceptual layer: CC BY-NC 4.0 | Operational substrate: Proprietary
The SDM framework was built to address a structural failure in modern research:
Paradigm-shifting ideas are filtered out before they can be properly evaluated.
Most research pipelines optimize for reproducibility, incremental improvement, and benchmark conformity. This creates an invisible filtering mechanism where ideas that challenge current paradigms are discarded not because they are wrong — but because they perform poorly against metrics designed for the paradigm they are attempting to replace.
SDM approaches the problem differently.
Instead of asking "Is this idea correct?" the framework asks:
"How far is this idea from empirical validation — and what would it cost to close that gap?"
The output is not a recommendation. It is a Structured Distance Measurement — a precise, historically durable map of the frontier between theoretical possibility and validated truth.
To explore the framework:
| Path | What You Get |
|---|---|
| Read the White Paper | Full conceptual framework — start here |
| Read the Manifesto | Architecture philosophy and design principles |
| Explore the NotebookLM Corpus | Interrogate 31 source documents interactively |
| Read a Live Exchange | See the framework operating on a real research question |
| Read the Canonical Definitions | Precise terminology used throughout |
| Understand the Licensing | What is open, what is proprietary |
To engage with the research:
- Open a Discussion to challenge a concept
- Submit an Issue to flag an inconsistency
- Read CONTRIBUTING.md for how to participate
Modern research systems are structurally biased toward safe convergence.
- Academic publication rewards reproducibility over risk
- Benchmarks reward optimization against existing paradigms
- AI systems optimize statistical confidence against prior distributions
- Corporate pipelines optimize for predictable returns
The consequence: ideas that challenge foundational assumptions are filtered out before they can be meaningfully explored — not because they are wrong, but because the evaluation system was built by the paradigm they are challenging.
The SDM framework treats this as a structural problem, not a model-quality problem.
Instead of forcing convergence, the framework creates controlled adversarial pressure between competing interpretations — and preserves the full reasoning record of that pressure as a durable research artifact.
At the center of the framework is a structured three-role adversarial exchange.
D1 recombines existing tools, theories, and structures into coherent but premature proposals. Its role is disciplined frontier exploration — not randomness. Every proposal must be logically coherent, source-linked, and include a minimal testable experiment. Speculation without a validation path is not an output.
D2 does not merely critique D1. It identifies the hidden assumption D1 failed to escape — then advances a competing proposal from a fundamentally different paradigm. The goal is forced paradigm separation, not debate performance. A critique without a counter-proposal is an incomplete output.
D3 is anchored to current engineering reality. Its role is not philosophical synthesis — it is measurement. D3 evaluates both proposals against empirical grounding, engineering feasibility, and validation cost. The output is a Structured Distance Measurement: the precise gap between each proposal and practical validation.
D3 measures distance. It does not decide whether the distance is worth crossing. That decision belongs to the human.
Step 1 — D1 Proposal
Step 2 — D2 Review + Counter-Proposal
Step 3 — D1 Review Notes (original proposal hash-locked)
Step 4 — Package Assembly (Orchestration Layer only — no interpretation)
Step 5 — D3 Empirical Arbitration
↓
Human-in-the-Loop Review
↓
APPROVE → System Memory
RETURN → Exchange (targeted revision)
ARCHIVE → Dissent Archive
ESCALATE → Governance Review
Every substantive claim must carry a Material Warrant — a typed, source-locked, provenance-chained claim record. Unsourced claims are blocked at the exchange boundary. They do not reach the next agent. There is no flag-and-pass path.
The orchestration layer performs exactly five functions: route packages, enforce role order, preserve provenance, trigger execution, and record movement. It does not summarize, interpret, filter, rank, or reason. Any orchestrator that interprets content becomes an undeclared fourth agent — introducing bias that cannot be audited.
Each agent maintains its own isolated research notebook. Agents cannot read each other's working research. The only material that crosses agent boundaries is what the orchestration layer explicitly passes as structured exchange artifacts.
The Human-in-the-Loop layer is the only authority that can convert validated reasoning into system memory. No automated process — regardless of confidence score or model consensus — can bypass this boundary.
The system measures the frontier. The human decides which distances are worth crossing.
Per-agent long-term memory. Stores validated outcomes, historical patterns, and agent-specific knowledge accumulated across runs. Append-only, governed by HITL approval events exclusively. Agents cannot write to SIM directly.
The system-wide governance layer. Defines role expectations, orchestration constraints, behavioral mandates, and structural boundaries. Not a research database — a constitutional layer. Agents read it at initialization to orient their role. They do not consult it during active research.
SIM accumulates experience. SMK enforces alignment. They do not exchange data.
Rejected and unresolved proposals are not discarded. They enter the Dissent Archive — a structured preservation system with three functions:
- Temporal reference — ideas archived as premature today may become viable as infrastructure evolves
- Calibration signal — patterns in archived proposals reveal systematic tendencies in the system's evaluation criteria
- Cross-run analysis — enables comparison of how different model-role configurations approach the same research domain
The archive accumulates not just knowledge, but the history of how knowledge was evaluated. That history grows more valuable as the world changes around it.
The framework is governed by a domain-agnostic behavioral enforcement substrate whose founding principle is:
The pattern stays the same. The output changes.
The same five-layer enforcement pattern that governs agent behavior in a research exchange also governs servicer behavior in financial compliance, provider behavior in healthcare authorization, and employee behavior in enterprise governance. The actor changes. The domain vocabulary changes. The enforcement pattern does not.
Applied to FPGA resource allocation for edge AI CNN deployment. The exchange surfaced a critical BRAM saturation anomaly at 25-thread configurations — a counter-intuitive non-linear trade-off that conventional optimization pipelines consistently missed. The dissent archive produced an auditable record of rejected configurations satisfying ISO 42001 Clause 9 transparency requirements.
Two complete Disruptor Exchange runs with different model-role permutations explored competing paradigms for replacing transformer attention. Full transcripts are in the /exchanges folder.
Run A: Recursive Binding Propagation (D1) vs Continuous Attractor Dynamics (D2) Run B: Same question, rotated model assignments
Cross-run finding: Both permutations independently produced the same paradigm split — discrete-structural versus continuous-dynamic computation — confirming the split as a genuine feature of the problem space rather than a model artifact.
Calibration finding: The Empirical Arbiter role systematically weighted engineering tractability over paradigm depth. This is documented as an open governance question: the D3 evaluation criteria may need adjustment for domains where paradigm depth matters more than near-term feasibility.
The framework is presented with its failure modes named explicitly.
| Limitation | Description |
|---|---|
| Shared Blind Spots | All participating models may share training-embedded assumptions the exchange cannot surface |
| Arbiter Calibration | D3 evaluates against a moving empirical target — its judgment is point-in-time, not permanent |
| Systematic Bias | Live runs revealed D3 may systematically favor tractability over paradigm depth |
| Human Bottleneck | HITL review creates scaling constraints — this is a feature, not a flaw |
| Corpus Pollution | Open research environments accumulate low-quality material; provenance enforcement is the primary defense |
| No Breakthrough Guarantee | The framework raises the ceiling. It does not guarantee what exists beyond it |
SDM/
├── README.md ← This file
├── MANIFESTO.md ← Open Research Architecture Overview
├── LICENSING.md ← Licensing framework and proprietary boundary
├── CONTRIBUTING.md ← How to participate
├── docs/
│ ├── white-paper.md ← Full conceptual framework v1.0
│ ├── canonical-definitions.md ← Precise terminology definitions
│ └── technical-architecture.md← System architecture overview
└── exchanges/
├── README.md ← Exchange format guide
├── run-a-attention-mechanism.md ← Live exchange Run A
└── run-b-attention-mechanism.md ← Live exchange Run B
The full source corpus — including architecture specifications, exchange transcripts, and supporting documents — is available for interactive interrogation via NotebookLM:
→ Open the Elosia SDM NotebookLM Corpus
This is the recommended starting point for deep engagement with the framework. You can ask questions, trace concepts across documents, and interrogate specific architectural decisions directly.
Conceptual layer — released under Creative Commons Attribution–NonCommercial 4.0 International (CC BY-NC 4.0)
This includes: the SDM thesis, the Disruptor Exchange model, the Consensus Trap theory, the Recursive Creativity framework, and all conceptual architectural descriptions.
Operational substrate — proprietary to ElosiaEcosystem Inc.
This includes: orchestration runtime systems, BDRM enforcement substrate, MicroManager/TAG infrastructure, SIM/SMK technical architecture, and all production governance implementation layers.
See LICENSING.md for complete terms.
Gary Phillips ElosiaEcosystem Inc. Structured Distance Measurement (SDM) Framework May 2026
Conceptual framework materials may be used for academic research, educational discussion, and non-commercial adaptation with attribution.