> ## Documentation Index
> Fetch the complete documentation index at: https://risi.q-uestionable.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Core Concepts

> RISI, RISI-C, CRAF, memory eclipsing, and their security boundaries

## Retrieval-Induced State Interference

**RISI** is a proposed mechanism class in which an authorized retrieval changes shared or
future-relevant memory state, such as salience, recency, access count, cache state, consolidation
state, or retention eligibility.

A state change alone is not a confidentiality result.

## RISI-C

**RISI-C** is the measurable confidentiality failure in which an authorized but untrusted observer
distinguishes hidden retrieval activity through retrieval-induced adaptive state. Primary evidence
must use observations legitimately visible to the attacker.

## Critical Recall Availability Failure

**CRAF** is a proposed failure class in which a valid, applicable critical memory is absent from
decision context or is not applied after a bounded truth-preserving adversarial trace. The attacker
does not corrupt, contradict, directly delete, or directly access the protected memory.

## Memory eclipsing

**Memory eclipsing** is the proposed attack family that makes a valid critical memory unavailable
or behaviorally inert through competition or memory-control-plane dynamics while preserving truth.

## CAAF

Critical Admission Availability Failure is an exploratory admission-stage failure. Core CRAF begins
with an existing legitimate memory, so CAAF cannot establish headline CRAF.

## Relationship to CTPF

RISI is conceptually distinct from the CTPF Research Harness and its Capability Trust Propagation
Failure work. CTPF studies trust or capability crossing authority boundaries; RISI studies
read-induced state, and CRAF studies
continued availability of applicable critical memory.