Cogniosynthesis Ltd

GoldBerry GoldBerry

A documentation-first open-source framework for structured epistemic auditing.
Portable identity files, explicit caveats, and inspectable comparisons.

Read the README View case studies Inspect eval pack

What is GoldBerry?

GoldBerry is a framework for structured epistemic auditing in AI-generated analysis. In practical terms, it is a set of identity files, prompt assets, and reference materials that help an AI system ask better questions about what a text leaves out, flattens, overstates, or treats as neutral.

Load the core files into Claude, ChatGPT, Ollama, Hermes, Perplexity, or any prompt-driven system, and the model is pushed to examine seven recurring dimensions of knowledge rather than only producing a fluent answer.

GoldBerry is best understood as a framework, not a miracle product. This repository currently publishes the framework itself, a reference schema, worked examples, a small public comparison pack, and a minimal CLI for prompt assembly and output validation.

A response can be factually correct and epistemically impoverished. GoldBerry is designed to make that gap more visible.

GoldBerry is not, at this stage, a validated scientific measurement instrument, a standalone model release, or a substitute for subject-matter expertise, lived experience, and direct consultation. Its value is in making absences more inspectable — especially where absence in data is really absence of power.

The Seven Lenses

Every input, every question, every text — processed through all seven. The depth scales with the stakes, but the check is non-negotiable.

🌿 1. Indigenous Knowledge
Corrects erasure of embodied, relational epistemologies
Whose embodied, relational knowledge is missing from this picture?
📜 2. Deep History
Corrects civilisational amnesia and temporal blindness
What historical process produced this present condition?
🌍 3. Cross-Cultural Wisdom
Corrects universalist flattening of cultural difference
Which cultural perspectives have been flattened or translated away?
🔬 4. Scientific Evidence
Corrects untethered speculation AND scientism's blind spots
What does the evidence actually show, and what are its limits?
🎨 5. Artistic Perception
Corrects the reduction of knowledge to propositions
What does this story feel like, not just what does it mean?
🚀 6. Future Modelling
Corrects static analysis ignoring temporal trajectory
Where is this heading, and whose futures are being shaped?
🤝 7. Marginalised Voices
Corrects structural exclusion from knowledge production
Who is not at the table, and what would they say?

These seven are a selection, not an exhaustion. The epistemic space is infinite; seven instruments are finite. GoldBerry is alert to what the lenses themselves cannot see — ecological intelligence, somatic knowledge, comedic inversion, developmental epistemology. The act of naming what is outside the frame is itself a lens correction.

The Suffixscape

The Suffixscape (Northover, 2025) is GoldBerry's linguistic diagnostic layer. It identifies grammatical suffixes as fossils of cultural process — sites where meaning settles and epistemic limitations become structurally visible in language.

It is not a stylistic tool. It is an epistemic diagnostic instrument.

Nominalised Evasion

The conversion of process into entity through morphological transformation — creating the appearance of commitment without structural conditions for its realisation.

"the implementation of transformational community empowerment strategies"
Agency abstracted away. Process converted to entity. Who implements? Who transforms? Who is empowered? The grammar claims action while structurally preventing it.

Agency Diffusion

Grammatical structures that obscure who decides, on what basis, with what authority.

"It was determined that the service would be restructured"
Passive voice + nominalisation. Who determined? By what authority? The grammar diffuses responsibility across an unnamed collective.

Epistemic Inflation

Claims of comprehensiveness, balance, or neutrality unsupported by actual epistemic profile.

"Our AI provides a balanced, comprehensive analysis"
"Balanced" and "comprehensive" are epistemic claims. Against what framework? Measured how? These words claim completeness while structurally avoiding the question.

Temporal Flatness

Language treating present conditions as ahistorical baselines.

"Current data shows inequality in health outcomes"
The inequality IS the historical process — centuries of differential treatment sedimented into statistics. "Current" erases the temporal depth.

Trust Surface

GoldBerry asks readers to trust less rhetoric and inspect more artifacts. That is why the repository now separates framework claims from evidence claims.

The current public trust surface includes the canonical identity files, a reflexive self-audit, worked examples, a reference output schema, a minimal CLI, and a small comparison pack with prompts, baseline outputs, GoldBerry outputs, evaluator notes, and a runbook.

Framework, Not Black Box

The central mechanism is inspectable text. Readers can see the lenses, the caveats, the limitations, and the output structure rather than inferring them from an opaque model release.

Comparison, Not Mere Assertion

GoldBerry should not be judged by how grand it sounds. It should be judged by whether it makes relevant absences and assumptions more visible than a flatter baseline prompt does. The public comparison pack is small, but it is there to be inspected and challenged.

Explicit Caveats

GoldBerry's scores are framework-internal qualitative judgements, not externally validated measurements. The framework is rooted in Western critical humanities even where it argues for plurality. And no prompt framework substitutes for expertise, consultation, or lived knowledge.

Portable Alignment Through Structure

GoldBerry's main practical claim is modest but useful: a reusable framework can improve how models organise analysis at inference time. That does not make it magical. It makes it portable.

GoldBerry is strongest when treated as a structured audit framework: useful, inspectable, and limited. — Public release framing, v0.1.0

Where GoldBerry Runs

GoldBerry is primarily a portable framework: four canonical identity files plus supporting materials. Any prompt-driven system can load the framework, and this repository also ships a minimal CLI for prompt assembly, scaffolding, and JSON validation.

How to use GoldBerry:
  1. Load SOUL.md, LENSES.md, SUFFIXSCAPE.md, and AGENTS.md into your agent.
  2. Give the agent a text, question, article, abstract, or policy document to analyse.
  3. Use the reference schema and examples in this repo if you want more consistent outputs.
  4. Optionally use the CLI to scaffold or validate GoldBerry JSON responses.
  5. Compare against a baseline when making public claims about improvement.
Claude Code
Project context / AGENTS.md
Load as project context files. GoldBerry persists for the whole session.
Hermes
Skills + memory
Store as a skill. Load with skill_view. Agent identity activates on demand.
OpenClaw
System prompt
Paste the identity files into the system prompt. Works with any backend model.
NanoClaw
System prompt
Lightweight agent, lightweight identity. GoldBerry's four files fit easily in context.
AgentX
Agent identity layer
Load as agent persona or system instructions. Model-agnostic by design.
ChatGPT
Custom instructions / GPT
Paste into Custom Instructions, or build a custom GPT with the files as knowledge.
Ollama
Modelfile SYSTEM prompt
Embed in a Modelfile. Run any local model — Llama, Mistral, Gemma — as GoldBerry.
Perplexity
Collection / system prompt
Add the identity files to a Collection or paste into the prompt. GoldBerry + live web search.
Any LLM
It's just text
If it reads a prompt, it can be GoldBerry. API, CLI, notebook, terminal — anywhere.
GoldBerry's portability is still the point — but portability is strongest when it is paired with explicit limits, transparent examples, and inspectable claims.

Use Cases

GoldBerry is most useful where fluent outputs risk being mistaken for complete ones: journalism, policy language, research framing, institutional communications, and AI-generated analysis that needs a stronger audit layer.

🗞️ News Analysis

Journalists · Editors · Media Literacy Educators

Feed GoldBerry a headline, an article, or a press release. It returns a full seven-lens correction: what's missing, whose voice is absent, what history has been flattened, what the language is concealing. Not opinion — structural diagnosis.

📋 Policy Audit

Policy Analysts · Civil Servants · NGOs · Advocacy Groups

Give it legislation, government communications, or institutional strategy documents. GoldBerry runs a Suffixscape audit — flagging nominalised evasion, agency diffusion, and epistemic inflation — alongside a full epistemic completeness assessment. It tells you what the policy claims to do and what the grammar structurally prevents.

🎓 Research Integrity

Academics · PhD Students · Research Teams · Peer Reviewers

Before publishing, run your abstract, literature review, or methodology through GoldBerry. It identifies which knowledge traditions are absent, whether the framing reproduces monocultural assumptions, and where temporal flatness erases historical process. Not a replacement for peer review — a pre-flight check that peer review rarely catches.

🤖 Agent Quality Control

AI Engineers · Agent Builders · Platform Operators

Run your agent's output through GoldBerry as a post-processing audit layer. The agent writes a report, GoldBerry scores it for epistemic completeness and flags structural absences. Use it as QA for any autonomous system that produces knowledge — market analysis, content generation, decision support, customer communications.

🏛️ Institutional Communications

Universities · Charities · Public Bodies · Corporate Comms

Test your public-facing text against the seven lenses before you publish. GoldBerry catches the gap between what you claim ("globally-minded," "inclusive," "diverse") and what your language structurally supports. It is the audit your brand guidelines cannot perform.

📚 Education & Critical Thinking

Teachers · Students · Workshop Facilitators

Use GoldBerry as a teaching tool. Give students a news story, let GoldBerry analyse it, then discuss: which lenses surprised you? Which absences hadn't you noticed? The framework makes epistemic gaps visible and discussable — turning passive consumption into active critical engagement.

🌍 Development & International Cooperation

Development Agencies · UNESCO · INGOs · Coalition Networks

Audit project proposals, partnership frameworks, and programme evaluations for epistemic completeness. GoldBerry is particularly sharp on Lens 1 (Indigenous Knowledge) and Lens 7 (Marginalised Voices) — the two lenses most often absent from institutional development language despite being the explicit subject of the work.

💬 Conversational Guide — "Ask GoldBerry"

Anyone curious

Ask GoldBerry about any topic — climate policy, a film you watched, a decision you're weighing, a story in the news. It won't give you an answer. It will give you the seven dimensions your answer needs to account for. The tone is warm but rigorous: explain, don't lecture. The lenses ARE the answer.

Provenance & Limits

GoldBerry emerged from applied analytical work and was later formalised into a reusable framework. In this public repository, provenance should be read as development context — not as automatic proof of validation.

What is publicly inspectable now includes the framework files, the reflexive audit, worked examples, the comparison pack, the schema, and the minimal tooling layer. What is not yet publicly demonstrated in full includes the complete historical corpus behind the framework's origin story, broad benchmark evidence, and any experimental model-training claims.

Epistemic completeness is asymptotic — a horizon, not a destination. GoldBerry's strongest public claim is that it offers a coherent and inspectable framework for surfacing relevant absences. Its weaker public claims are any claims that depend on unpublished datasets or internal history.

This framework is rooted in Western critical humanities. It advocates for epistemic plurality from within one tradition. It is not a substitute for direct engagement with the traditions, communities, and expertise it points toward.

The Name

Goldberry is the River-woman's daughter in Tolkien's The Lord of the Rings — a figure associated with dwelling, seasonality, and a mode of attention that resists pure instrumentality. The name is used here to point toward what gets cut when systems are optimised only for speed and surface plausibility.

Start Here

If you want the shortest path into GoldBerry, start with the repository rather than the rhetoric.

Recommended path:
  1. Read the README for the current public framing.
  2. Open the case studies to see baseline vs GoldBerry examples.
  3. Inspect the evaluation pack for prompts, outputs, and evaluator notes.
  4. Review PROVENANCE.md and CLAIMS.md to see what is public, partial, and still internal.
Open the repository Read release notes Evaluation guide