Evaluation systems reduce complex student thinking into a single number.
They are created during evaluation. When systems store only scores, they are lost permanently.
This is where responses are understood, judgments are made, and learning signals are generated.
If signals are not captured here, they are lost permanently. No downstream system can recover them.
Signals must be captured, structured, and stored — not discarded after scoring.
This is not an optimization. It is a structural requirement for modern learning systems.
Not as a workflow. Not as a feature. But as the layer where learning signals are created and stored.
Each answers a failure mode from the model above.
// COMMON.QUESTIONS
Scores don't capture learning because evaluation systems collapse student responses into a single number, discarding the learning signals generated in the process — concept understanding, error patterns, rubric alignment, and reasoning. CrazyGoldFish captures these signals during evaluation and structures them as persistent learning memory, achieving 93% reliability in total scoring alignment (ICC).
Why don't scores capture learning?
Scores collapse student responses into a single number, discarding everything generated in the process — concept understanding, error patterns, step-wise reasoning, and rubric alignment. The evaluation produces these signals, but the scoring layer throws them away. What persists is only whether the student passed or failed, not what they actually understood.
What are learning signals in evaluation?
Learning signals are the structured outputs that evaluation generates beyond the score itself — concept-level understanding, error type classifications, step-wise reasoning quality, and rubric alignment markers. CrazyGoldFish captures these during evaluation and structures them as persistent learning memory, making them reusable for personalisation, remediation, and platform intelligence.
What happens when evaluation only produces scores at scale?
Three structural failures emerge at scale. First, inconsistency — evaluators disagree, and scores carry no confidence signal. Second, loss of information — everything below the surface of the score is permanently discarded. Third, inability to reuse — scores are terminal; they cannot feed personalisation, remediation, or platform memory. CrazyGoldFish addresses all three.
How does CrazyGoldFish capture learning signals?
CrazyGoldFish intercepts evaluation at the point where learning signals are generated and structures them before they are discarded. Every response produces a LearningSignal object — concept_gap, confidence_score, error_type, memory_id — stored as persistent learning memory. Reliability: 93% alignment with human scoring (ICC), 89.7% adjudicated correctness vs teacher 82.8% (+6.9%).
What does evaluation infrastructure mean?
Evaluation infrastructure is the system layer that captures learning signals during assessment and stores them as persistent learning memory. Unlike grading tools that produce scores and stop, CrazyGoldFish builds the infrastructure layer that converts every evaluation into structured, reusable intelligence — making evaluation the control point where learning is interpreted and remembered.
// RELATED · RESEARCH & PLAYBOOKS
See how CrazyGoldFish structures learning signals into persistent evaluation infrastructure.