Theory layers let AGISystem2 maintain multiple, potentially contradictory views of the world simultaneously. Each layer can override or extend the knowledge below it. This enables counterfactual reasoning, context switching, and safe experimentation without corrupting base knowledge.

How Layers Work

Each layer contains facts that can add to or override facts in lower layers. When the Reasoner processes a query, it searches layers from top to bottom. The first matching fact wins. If no layer has a match, the query returns unknown.

Layers support three types of changes:

• Additions – New facts not present below (e.g., "Aspirin TREATS Headache")
• Overrides – Different value for existing fact (e.g., "Water IS_A Gas")
• Retractions – Explicit removal of a fact from view (marker that blocks lookup)

Use Cases

Use Case	How Layers Help
Counterfactuals	Push a layer, add hypothetical facts, reason, then pop to restore reality
User sessions	Each user gets a layer with their preferences and context
Domain switching	Load a "medical" or "legal" layer with domain-specific knowledge
Safe experimentation	Test changes in a layer before committing to base
Versioning	Save layers as snapshots, load to restore previous states

Implementation

The TheoryStack and TheoryLayer components in the Knowledge Layer manage this:

Operation	Description
push(name)	Create new layer at top of stack
pop()	Remove and discard top layer
peek()	Get current top layer
save(filename)	Persist current stack to disk
load(filename)	Restore stack from disk
merge()	Flatten top layer into the one below

Sys2DSL Commands

# Push a new layer
THEORY_PUSH name="experiment"

# Add facts to current layer
ASSERT Water IS_A Gas
ASSERT Water TEMPERATURE_AT Celsius200

# Query (will find Gas, not Liquid)
ASK "Water IS_A ?"

# Pop to discard hypothetical
THEORY_POP

# Now Water IS_A Liquid again

Related Documentation

• TheoryStack Implementation – Stack management
• TheoryLayer – Individual layer structure
• Counterfactual Reasoning – Using layers for "what-if"
• Non-Monotonic Logic – How layers enable belief revision
• Syntax Reference – THEORY_PUSH, THEORY_POP commands