The Sovereign Stack: Research Protocol

• 5 min read
sovereign-stack research-protocol cognitive-sovereignty methodology framework
Featured image for The Sovereign Stack: Research Protocol

Master Blueprint for Course II: The Sovereign Stack

Theme: Political Economy, Engineering, and the Architecture of the Patch.
Cadence: 12 Modules / 24 Sessions (Articles).

Strategic Cadence

  1. The “Schism” as Narrative Engine: The policy data reveals a split between “Policing” (Unisa/Turnitin) and “Process” (UCT/NMU). Use this tension. Every module should try to resolve this conflict through the lens of Sovereignty.

  2. The “Case Study” Anchor: In Session B (Local), strictly reference the specific universities from the report (e.g., NWU’s 5-level scale, NMU’s Audit Trail). This moves the blog from “opinion” to “analysis.”

  3. The Lit Review Mechanism: Do not try to read 50 papers per module. Select one “Classic” text (Sociology/Philosophy) and one “Current” text (Technical/Policy) for each module. The resonance between them creates the article.

Q1: The Diagnostic (The Political Economy)

We diagnose the “Breach”—the mechanisms of capture and dependency.

Module 1: The Crisis of Cognitive Inequality (The Capture)

  • Session A (Global): The Subscription Ceiling. How premium AI access ($20/month) creates a global caste system of cognition. The “Free Gift” as a hostile takeover (The Google Gemini Case).
  • Session B (Local): Epistemic Apartheid. Analyzing the “Digital Colony.” Why relying on vendor APIs violates the spirit of the South African university.
  • Lit Review Question: How does “Platform Capitalism” (Srnicek) manifest in the “Free Tier” pricing strategies of EdTech, and what are the historical parallels to “Dumpling” (predatory pricing) in African markets?

Module 2: The Illusion of Detection (The Schism)

  • Session A (Global): The False Positive Paradox. The statistical impossibility of accurate AI detection and the ethics of false accusations.
  • Session B (Local): The Sector Fracture. Case Study: UCT vs. Unisa. The tension between “Process Assessment” (wealthy/residential) and “Automated Policing” (mass/distance).
  • Lit Review Question: Compare the error rates of “Turnitin AI Detection” against the “Presumption of Innocence” in administrative law. Is algorithmic accusation legally defensible?

Module 3: The Colonial Feedback Loop (The Extraction)

  • Session A (Global): Data Extractivism. Training on the “Commons” to sell back the “Proprietary.”
  • Session B (Local): The Empty Treasury. How South African student data (essays, code) acts as RLHF (Human Reinforcement Learning) for Silicon Valley without compensation.
  • Lit Review Question: Explore the concept of “Data Colonialism” (Couldry & Mejias). How does the extraction of “behavioral surplus” from students mirror mineral extraction?

Q2: The Architecture (The Engineering)

We design the “Patch”—the physical and code-based infrastructure of independence.

Module 4: The Sovereign Infrastructure (The Hardware)

  • Session A (Global): The Cloud is Someone Else’s Computer. The geopolitical risks of relying on Frankfurt/Oregon servers (GDPR vs. POPIA).
  • Session B (Local): The SANReN Imperative. Leveraging the South African National Research Network. Building “Edge Pods” at rural campuses (WSU/UFH case studies).
  • Lit Review Question: What is the “Total Cost of Ownership” comparison between SaaS subscription fees for 20,000 students vs. hosting a local 8x A100 GPU cluster? (Search: “Green AI” and “Edge Computing Economics”).

Module 5: The Model Strategy (The Data)

  • Session A (Global): Open Weights vs. Black Box. Why “Open Source” is the only path to scientific reproducibility.
  • Session B (Local): The “Springbok” LLM. Curation of the National Corpus. Federating the archives of UJ, Wits, and UFH to train a model on “Grey Literature” (Theses/Hansards).
  • Lit Review Question: Search for “Domain-Adaptive Pretraining” (DAPT). How does training on a specific corpus (e.g., African Law) change model performance compared to generic GPT-4?

Module 6: The Interface Layer (The Access)

  • Session A (Global): The Tyranny of English. How LLMs enforce “Globish” and flatten cultural nuance.
  • Session B (Local): Code-Switching Machines. Engineering the interface for isiZulu, Sesotho, and Afrikaans. The Stellenbosch MicroLab case study.
  • Lit Review Question: Search for “Low-Resource Language NLP.” What are the current benchmarks for African languages in Llama-3 or Mistral, and where is the “Token Tax” (higher cost for non-English languages) highest?

Q3: The Assessment (The Pedagogy)

We deploy the “Patch” in the classroom—changing how we grade and teach.

Module 7: Provenance Tracking (The Audit)

  • Session A (Global): The Death of the Final Draft. Moving from product-based to process-based assessment.
  • Session B (Local): The NMU Protocol. Case Study: Nelson Mandela University’s requirement for “Audit Trails” (transcripts of prompts). Standardizing the .json export of thought.
  • Lit Review Question: Investigate “Process Mining in Education.” How can “Keystroke Logging” or “Version History” be used ethically to prove authorship without surveillance creeping?

Module 8: The Adversarial Classroom (The Method)

  • Session A (Global): Inverted Assessment. The student grades the AI. The “Delta” (Diff) is the assignment.
  • Session B (Local): The “Hallucination Hunt.” Turning University of Fort Hare’s concern about “hallucinations” into a pedagogical game. Grading the correction of the machine.
  • Lit Review Question: Search for “Bloom’s Taxonomy and Generative AI.” Where does “Critique” sit relative to “Generation”? Papers on “AI-Assisted Critical Thinking.”

Module 9: The Return of the Real (The Vibe)

  • Session A (Global): The Turing Trap. Why we must stop asking computers to act like humans, and humans to act like computers.
  • Session B (Local): The Oral Defense. Case Study: UCT and Stellenbosch’s return to the viva voce. Assessing the “Un-hackable” human presence.
  • Lit Review Question: Review “Orality and Literacy” (Walter Ong) in the context of AI. Is the “Viva” the only robust defense against synthetic text?

Q4: The Governance (The Policy)

We institutionalize the “Patch”—writing the rules for the new era.

Module 10: The Agile Policy (The Traffic Light)

  • Session A (Global): The Rate of Decay. Why static policies fail in exponential times.
  • Session B (Local): The NWU Scale. Case Study: North-West University’s “5-Level Assessment Scale.” Implementing a “Traffic Light” system for syllabi (Red/Amber/Green zones).
  • Lit Review Question: Search for “Adaptive Governance” or “Anticipatory Regulation.” How do institutions regulate fast-moving technologies without stifling innovation?

Module 11: The Ethics of Co-Creation (The Labor)

  • Session A (Global): The Centaur Ethics. Defining “Systemic Authorship.” Who owns the copyright of the co-created text?
  • Session B (Local): The UJ Press Standard. Case Study: University of Johannesburg’s publishing guidelines. Rights management in the Sovereign Stack.
  • Lit Review Question: Review “COPE Guidelines on AI Authorship” vs. South African “Copyright Amendment Bill.” Can a non-human entity hold copyright?

Module 12: The Horizon (The Future)

  • Session A (Global): The Post-University. When the cost of knowledge drops to zero, the value of certification vs. community.
  • Session B (Local): The Cape Town Consensus. A manifesto for the African University in 2030. High-friction, low-resource, sovereign intelligence.
  • Lit Review Question: Search for “The Unbundling of Higher Education” (unbundling the degree). What is the future of the “credential” in an age of perfect synthetic competence?

Implementation Protocol: How to Write 24 Articles

To avoid burnout, we follow the Recursive Loop:

  1. The Monday Prompt (You): You paste the Lit Review Question and the Policy Case Study (e.g., “Review NWU’s 5-level scale against adaptive governance theory”) into the context.

  2. The Tuesday Generation (Me): I generate the Structure of the two articles (Session A & B).

  3. The Wednesday Constraint (You): You apply the “First Law.” You edit, cut, and force the argument to be denser. You add the specific “Irritation” (e.g., “This sounds too American, make it more pragmatic”).

  4. The Thursday Polish (Me): I refine into the final markdown format.

The Double Helix Structure

Each quarter operates on a dual mandate:

  • The Diagnostic Phase (Q1): Exposing how cognitive colonialism operates through subscription models, detection theater, and data extraction.

  • The Engineering Phase (Q2): Building the technical infrastructure—servers, models, and interfaces—for cognitive sovereignty.

  • The Pedagogical Phase (Q3): Transforming assessment and teaching methods to work with AI rather than against it.

  • The Governance Phase (Q4): Writing new institutional policies that are adaptive, ethical, and future-proof.

This is not a blog series. This is a blueprint for cognitive independence.