// evaluation.control.point
The system layer where evaluation captures structured intelligence — persisting what scores discard.
// the.collapse.problem
WHAT SYSTEMS STORE
72 / 100
evaluation collapsed to output
WHAT GETS LOST
Without capturing learning signals, education has no memory of learning. This is why scores don't capture learning signals.
// system.architecture.shift
TRADITIONAL SYSTEM
LEARNING SIGNALS SYSTEM
This is not a gradual improvement — it is a shift in system architecture.
// category.primitives
01
LEARNING SIGNALS
Atomic, structured units of learning generated during evaluation — not captured or structured by existing systems.
02
EVALUATION CONTROL POINT
The stage where responses are interpreted against rubrics, policies, and human judgment — where signals are validated and captured.
03
PERSISTENT LEARNING MEMORY
A structured record of learning signals accumulated across time — enabling longitudinal understanding, not just per-assessment scoring.
// signal.flow.architecture
EVALUATION CONTROL POINT
Responses are interpreted using rubrics, policies, and human oversight.
SIGNAL CAPTURE
Learning signals are extracted at the question level.
SIGNAL STRUCTURING + STORAGE
Signals are converted into structured, machine-readable formats and stored as a cumulative dataset.
MEMORY LAYER FORMATION
Signals accumulate and evolve into persistent learning memory across time.
// learning_signal.json
{
"response_id": "r_8421",
"signal_type": "conceptual_gap",
"rubric_alignment": "partial",
"icc_score": 0.93,
"confidence_band": 0.87,
"teacher_override": false,
"feedback_payload": {
"gap_identified": "Newton's Third Law",
"recommendation": "review_motion_concepts"
},
"memory_persisted": true
}// capabilities.unlocked
CONSISTENT EVALUATION
Reliable assessment across evaluators and time — no evaluator drift.
FULL AUDITABILITY
Every evaluation decision traceable, reviewable, and defensible.
RELIABLE PROGRAM DECISIONS
Policy and program choices grounded in structured learning data.
PATTERN-BASED PERSONALIZATION
Recommendations derived from actual learning signals, not assumptions.
SYSTEM-LEVEL INTELLIGENCE
Aggregate signals across cohorts, subjects, and time.
Learning shifts from isolated events to a continuous, structured signal layer.
// capability.shift
MULTIMODAL INTERPRETATION
Handwritten and complex responses now interpretable at evaluation quality.
POLICY-AWARE EVALUATION
Rubric-driven systems that apply institutional policies consistently at scale.
LARGE-SCALE SIGNAL CAPTURE
Learning signals extracted across thousands of responses in a single workflow.
Evaluation can now reliably capture learning signals at scale — something previously not possible.
Without this shift, increasing scale will only increase inconsistency.
// category.differentiation
NOT SCORING TOOLS
Scoring tools produce scores.
Do not capture learning signals.
NOT ASSESSMENT PLATFORMS
Manage assessments.
Do not structure learning signals into structured learning intelligence.
NOT REPORTING SYSTEMS
Reporting systems interpret signals after evaluation.
Do not capture signals at the source.
Learning Signals defines the mechanism through which learning becomes structured and usable — the foundation of what the AI grading system captures.
It defines the fundamental data layer for learning systems.
// product.position
// CGF.OPERATION
CAPTURE
At the evaluation control point
STRUCTURE
Machine-readable learning signals
STORE
Persistent learning memory
This layer becomes the foundation on which evaluation infrastructure is built.
This category is already forming — CrazyGoldFish is building it.
// icp.routing
EDTECH PLATFORMS
See how learning signals transform platform-scale evaluation.
See EdTech Solution →COACHING & TEST-PREP
See how signals replace inconsistent subjective grading at volume.
See Coaching Solution →LMS / ERP PLATFORMS
See how learning signals integrate into existing infrastructure.
See LMS Solution →INSTITUTIONS & PUBLIC PROGRAMS
See how signals enable audit-ready evaluation at scale.
See Institutions Solution →// COMMON.QUESTIONS
Learning signals are structured, atomic units of learning captured during evaluation — not scores. They capture how a student reasoned, where they made mistakes, and what concept gaps emerged. CrazyGoldFish captures and structures these signals as persistent learning memory, forming the foundation of evaluation infrastructure.
What are learning signals?
Learning signals are atomic, structured units of learning data generated when a student response is evaluated. They capture what the student understood, where mistakes occurred, which concept gaps emerged, and how the evaluator interpreted the response. Unlike scores, learning signals persist as structured, machine-readable data that can be reused across systems and time.
How are learning signals different from assessment scores?
A score collapses a response into a single number — once assigned, the underlying learning information is lost. Learning signals are the structured data underneath the score: concept gaps, rubric alignment, confidence levels, and feedback context. Scores are outputs; learning signals are the data layer that makes those outputs reusable.
What is an evaluation control point?
The evaluation control point is the moment when a student response is interpreted against rubrics, policies, and human judgment. It is the only stage in the learning workflow where signals can be captured. CrazyGoldFish operates at this control point to extract and structure learning signals before they disappear into a score.
What is persistent learning memory?
Persistent learning memory is a cumulative record of learning signals accumulated across evaluations and time. It enables education systems to understand how a student's learning evolves — not just what score they received on a single assessment. CrazyGoldFish builds this memory layer by structuring and storing signals at the evaluation control point.
Why can't existing systems capture learning signals?
Existing grading tools and assessment platforms are designed to produce outputs, not capture structured data. They store scores, not the signals inside responses. Capturing learning signals requires operating at the evaluation control point — interpreting responses, structuring output as machine-readable data, and persisting it as a cumulative dataset. This is infrastructure, not a grading feature.
// RELATED · RESEARCH & PLAYBOOKS
See how CrazyGoldFish captures learning signals at the evaluation control point — turning responses into structured intelligence.