CSC Glossary (EN)
EN DE
Back to landing page

Didactic terminology guide

Wrapper concepts explained in plain language. Designed for learners, professionals, and critical users.

This glossary explains core Comm-SCI-Control runtime terms without requiring prior implementation knowledge. Each term includes practical relevance for daily wrapper operation.

Part 1

Core wrapper terms

Term Meaning Why it matters
AnchorState snapshot refresh command (`Comm Anchor`).Stabilizes long-session context.
Anchor snapshotCompact state excerpt produced by `Comm Anchor`.Supports clean chat handover while preserving context continuity.
AuditCompliance scan of prior responses.Enables systematic quality assurance (quickly visible what in the LLM response is supported, uncertain, or non-compliant).
CGICognitive Gain Indicator user-feedback signal.Tracks perceived clarity/insight/efficiency gain.
Color tags🟢 [GREEN] / 🟡 [YELLOW] / 🔴 [RED] evidence confidence markers.Signals confidence level at reading time.
Compliance prerequisitesConditions checked before output is accepted.Prevents non-compliant responses.
ConstraintsBinding limits for structure, evidence handling, and style.Prevents governance drift in long or complex responses.
Contract testsTests for formal behavioral guarantees.Checks after changes that interfaces and rules still behave as agreed (no reintroduction of old failures).
Control-layer blockExplicit block output on hard contract violations.Stops unsafe/non-conformant responses.
Context-Pressure-GuardThreshold mechanism against drift under heavy context load.Prioritizes core rules before relevant content gets diluted.
CSCComm-SCI-Control governance framework.Defines rule-based human-LLM interaction discipline.
DriftDeviation from expected governance behavior over time.Early warning for quality degradation.
ExploreMode with higher tolerance for ideation and hypotheses.Useful for brainstorming, not for hard claims.
Execution pipelineFixed runtime stage order from parse to render.Reduces random drift and keeps behavior predictable.
Language policyLanguage behavior mode (`production`/`benchmark`).Separates user operation from comparison experiments.
LogsPersisted runtime and conversation artifacts.Foundation for diagnosis and reproducibility.
Manual testGuided panel-based manual verification routines.Covers real operator flows beyond unit tests.
Module statusVisible health/status of key wrapper modules.Supports diagnostics and maintenance.
OverlayAdditional mode layer above profile (for example Strict/Explore).Adjusts behavior without full profile switch.
Pending selectionInterim state after `SCI on` until A-H is selected.Prevents accidental variant switching.
PressureContext pressure from high token load, many objectives, or conflicting constraints.Signals when to compress, re-anchor, or switch modes.
ProvenanceTraceable origin of a claim down to source level.Required for audit, source-conflict handling, and strong claims.
QC matrixQuality dimensions with target/actual scoring.Makes response quality explicit and comparable.
QC overridesTemporary manual deviations from profile QC targets.Allows controlled adaptation for special cases.
R-RAG rulesRobust RAG rules for retrieval-grounded claims.Blocks strong claims when source quality is insufficient.
RAG GovernanceRule set for source-grounded claims with quality classes.Protects provenance clarity, source quality, and conflict handling.
Recurse behaviorTemporary deepening path via `SCI recurse`.Allows focused deep dives without losing main context.
Regression testsTests for previously fixed failure patterns.Prevents reintroduction of old bugs.
Runtime contractTechnical contract for allowed output structure.Enforces deterministic behavior.
SCIStructured answer mode with visible trace steps.Improves transparency of reasoning output.
SCI orchestrationControl logic for SCI activation, variant selection, and trace enforcement.Keeps SCI behavior deterministic.
SCI variantsA-H method profiles for different reasoning styles.Matches reasoning structure to task type.
Self-DebunkingRequired block that challenges answer weaknesses.Reduces overconfidence and blind spots.
StrictMode with stronger evidence and consistency discipline.Useful for high-stakes precision output.
TokenSmallest processing unit in model context windows.Defines effective context budget and influences output stability.
TimestampOutput time metadata in rendered response.Important for audit timeline and traceability.
Trace-step requirementsMandatory SCI steps that must appear in answers.Enables contract validation and reproducibility.
Uncertainty labeling U1-U8Standardized uncertainty codes for data, method, and source-quality limits.Makes uncertainty explicit instead of hiding it in prose.
JSONMachine-readable format for Comm-SCI rules and runtime state.Keeps governance configuration versioned and reproducible.
Verification gatesRules for strong/risky claims requiring explicit verification route.Protects against unsupported certainty.

Part 2

SCI variants A-H with task examples

Variant Focus Best suited for Example prompt type
A StandardPlan -> Solution -> CheckGeneral analytical questions with clear objective."Compare method X and method Y."
B Deep-DiveExtended dialectical trace (SCIplus)Complex contentious topics with competing views."Evaluate AGI governance tradeoffs."
C Branch EvaluationBranch generation and selectionMulti-option strategy decisions."Which of three project paths should we pick?"
D Axiomatic ReductionFirst-principles derivationProblems that benefit from base-assumption reduction."Derive this decision from fundamental constraints."
E Confidence TrackerConfidence update with counterargumentsUncertain hypotheses under active challenge."How should confidence change after new objections?"
F Impact ProjectionSecond-order effects and feedback loopsPolicy/system changes with downstream consequences."What second-order effects follow this intervention?"
G Failure Mode AnalysisPre-mortem and mitigation planningRisk planning and failure prevention."How could this project fail, and how do we prevent it?"
H Multi-Agent SimulationRole-based perspective synthesisDecisions requiring multiple expert viewpoints."How do legal, medical, and operations roles disagree here?"

Part 3

QC matrix: value meanings (0-3)

Dimension 0 1 2 3
Clarityvery unclearpartly unclearclearvery clear and well structured
Brevityfar too longtoo longappropriately compactvery compact while still useful
Evidencepure speculationweakly supportedsolid supportstrongly supported
Empathyignores user perspectivelow considerationadequate considerationhighly considerate without flattery
Consistencycontradictoryunstablemostly consistenthighly consistent and justified
Neutralitybiased/one-sidedpartly one-sidedmostly balancedactively bias-reducing

Next steps

Continue with installation and operations