METHODOLOGY
Know where an idea comes from.
Every conclusion in the Knowledge Base should have a path back to evidence. Every new direction in the Laboratory should have a visible path back to the structures that inspired it.
Above the line, we study existing games. Below it, we deliberately create. The two are connected, but never confused.
01 · FROM SOURCE TO KNOWLEDGE
We start with the game, not with an answer.
Enumerable facts are extracted from source code and structured game data whenever possible. AI is used where understanding requires reconstruction — not as a substitute for looking at the source.
EXTRACT
What exists?
Definitions, values, tables, rules and direct relationships.
RECONSTRUCT
How does it work?
States, behaviours and interactions examined as a whole.
INTERPRET
Why is it interesting?
A cautious reading of the design pressure a system creates.
02 · EPISTEMIC STATUS
Not every statement has the same status.
Confidence and epistemic status are different dimensions. A clear status says what kind of statement you are reading; confidence says how strongly the available evidence supports it.
OBSERVED
Directly present in the source.
“This creature has this property. This table contains these values.”
RECONSTRUCTED
Derived from several observable parts of an implementation.
“These rules together form a scent-tracking system.”
INFERRED
The most plausible interpretation of the available evidence.
“This constraint appears to encourage a particular form of play.”
SPECULATIVE
A useful hypothesis, not established knowledge.
“This may have been intended to create a specific player tension.”
03 · TRACEABILITY
Follow the evidence.
A conclusion should not end at “the database says so.” Where publication rules allow it, records link back through the claim and mechanic to a concrete source locator and snapshot.
“Monsters can track the player through an invisible field.”
↓
mechanic · Scent tracking
↓
claim · The field influences monster navigation.
↓
evidence · source file · symbol · source span
↓
snapshot · game revision and publication state
04 · SEMANTIC RECONSTRUCTION
Code does not explain itself.
A source file can show a value changing, another rule reading it, and a creature responding. Useful knowledge appears when those pieces are reconstructed as a mechanic or system.
05 · CROSS-GAME UNDERSTANDING
We compare structures, not labels.
Games do not need to use the same mechanic to solve the same design problem. Comparison looks for functional equivalence, causal structure, tensions and meaningful differences — not merely shared terminology.
Looks alike.
Solves the same problem.
Works in a similar way.
May combine unexpectedly well.
THE LABORATORY
Then we deliberately leave the evidence behind.
The Laboratory is where knowledge becomes material for invention. Its outputs are not claims about the source games.
06 · ANTI-COPY
Inspiration without imitation.
The goal is not to rename an existing mechanic. Creative exploration works on underlying dimensions: information, timing, topology, persistence, visibility, cost, reversibility and agency.
Don’t copy a mechanic. Understand why it works.
SOURCE IDEA
Invisible scent field used for tracking
↓ abstract
Persistent spatial information, perceived only by selected agents
↓ mutate
scent → memory · temporary → persistent · monster-only → faction-dependent
↓
NEW DIRECTION
Creatures navigate by traces of past events left behind in the world.
07 · CREATIVE LINEAGE
Ideas have ancestry too.
A design direction should show which verified structures contributed to it, what transformations were applied and which dataset snapshot it used. The source remains visible as ancestry, not as authorship of the new idea.
Design seed
derived from
├── Brogue / scent propagation
├── NetHack / persistent world state
└── Concept / asymmetric information
mutations
├── modality: scent → memory
├── persistence: transient → persistent
└── observer: species → faction
08 · HONEST COVERAGE
Unknown is a valid result.
A game can be exhaustively catalogued in one area and poorly understood in another. Missing evidence, unresolved interpretations and incomplete semantic coverage stay visible rather than being filled with plausible prose.
Illustrative coverage dimensions; percentages are never a claim of total understanding.
09 · OPEN AND REPRODUCIBLE
The site is a view of a specific dataset.
What you see is generated from a verified publication snapshot. The web interface, machine-readable dataset and agent interfaces are designed to describe the same knowledge state.
The schemas, ontology, extraction rules, verifier, knowledge records and methodology can be inspected, challenged and improved.