Skip to content

CONTEXT.md

Overview

This repository reflects a body of work situated at the intersection of:

  • bibliometrics and research systems
  • knowledge infrastructure analysis
  • AI-assisted research workflows
  • emerging work on AI evaluation, verification, and structured reasoning

The work is not a collection of isolated analyses, but a set of structured methods, frameworks, and tools designed to:

  • represent research and knowledge systems
  • analyse their structure and dynamics
  • evaluate outputs, including AI-generated outputs

Core conceptual model

The work operates across three primary layers:

  1. Methods (conceptual)

    • define how systems are represented and analysed
    • abstract, transferable, not tied to a specific implementation
  2. Frameworks (applied structures)

    • structured applications of methods
    • introduce dimensions, workflows, and constraints
    • partially operational but not fully executable
  3. Tools (operational)

    • implementations supporting specific tasks
    • interface-driven or programmatic
    • do not expose full internal logic

Additionally, there is a supporting layer:

  1. Prompt artifacts / protocols
    • structured prompts used within tools and workflows
    • encode behaviour, evaluation logic, or style
    • not exposed as standalone methods

Implemented baseline

The following components are now defined and publicly structured:

Method

  • AI metascientist

Framework

  • GRID+

Tool

  • AVA

These establish a minimal working system across all three layers: method → framework → tool


Key distinction: method vs prompt artifact

This distinction must be preserved:

  • Methods / frameworks

    • describe how work is done
    • conceptual or structured processes
    • reusable independently of tools
  • Prompt artifacts (e.g. style.md)

    • encode instructions for AI systems
    • operational components
    • belong to tooling or execution layers

Example:

  • writing research publications → method / framework
  • style.md → prompt artifact

Structural constraint

Each layer must remain distinct in both content and level of detail:

  • methods define conceptual logic
  • frameworks define structured application
  • tools implement constrained operations

Tool pages must not exceed frameworks in detail.
Frameworks must not expose full operational logic.


Latent / implicit project categories

Based on working patterns, the following are likely to exist as implicit assets:

  • prompt protocol library
  • evaluation prompts (AI verification / validation)
  • classification systems (e.g. funder classification, journal categorisation)
  • data construction pipelines
  • mobility / researcher flow analysis frameworks
  • citation behaviour analysis methods
  • metric frameworks (novelty, interdisciplinarity, disruption)

These are candidates for formalisation, but not all necessarily public.


Intellectual property strategy

The work follows a layered disclosure model.

Public

  • conceptual methods
  • high-level frameworks
  • selected examples
  • partial workflows

Partial / gated

  • structured frameworks with limited detail
  • worked examples
  • simplified logic

Private

  • full prompt systems
  • optimisation heuristics
  • evaluation logic and scoring
  • orchestration / chaining
  • internal datasets or mappings

Key principle: Do not expose anything that enables replication of most of the value.


Positioning

The work should be framed as:

  • methodological, not product-driven
  • structural, not descriptive
  • rigorous, not promotional

Avoid:

  • startup language
  • solution framing
  • feature-driven descriptions

Preferred framing:

  • methods
  • systems
  • frameworks
  • evaluation

Long-term direction

The work is expected to evolve from:

  • bibliometrics and research systems

toward:

  • evaluation of knowledge systems
  • AI verification and safety
  • structured reasoning and validation

This requires maintaining:

  • domain flexibility
  • conceptual consistency
  • non-commercial tone

Last updated: