How we think about learning at LoreGraph

LoreGraph started with a practical problem:

Organizations already have knowledge.

It lives in SOPs, policies, manuals, handbooks, presentations, and internal documents.

But owning a document is not the same as helping someone learn from it.

That distinction shapes how we think about the product.

Today, LoreGraph helps teams turn existing documents into structured course drafts, lessons, quizzes, and trackable training workflows with human review before publishing.

Source: LoreGraph. But our view of learning is broader than document conversion.

The principles below describe how we evaluate the problem and the direction we want to follow. They are design principles—not a claim that every idea is already implemented in the product today.

Our starting definition: learning is a change in capability

We do not define learning as opening a file.

We do not define it as watching a video.

We do not define it as finishing every page in a course.

Those activities may contribute to learning, but they are not interchangeable with it.

Our Week 1 learning-science foundation uses a practical definition:

Learning is a relatively durable change in knowledge, skill, judgment, strategy, or capability that develops through experience and can be demonstrated later, ideally in a relevant context.

That definition creates a high bar.

A learner can complete a course without retaining it. A learner can pass an immediate quiz because the wording is familiar. A person can understand a procedure but still fail at work because the environment makes the correct behavior difficult.

Learning has to be inferred from evidence, not declared by the platform.

From that starting point, several product principles follow.

Principle 1: Start with what someone should be able to do

Training should begin with an intended change.

Not:

Teach the employee the incident policy.

But:

When a reportable event occurs, the employee should recognize it, begin the correct reporting process, and know where to find the required details.

The second statement gives the learning experience a destination.

It also helps distinguish essential knowledge from reference material.

This matters when converting workplace documents.

Documents are normally organized to preserve information, satisfy policy requirements, define responsibilities, or support reference.

That order is not automatically an effective learning sequence.

Before generating more content, we should ask:

• What decision should improve?
• What action should become possible?
• What explanation should the learner be able to give?
• What evidence would reveal the change?

The capability comes first.

The course is one possible route to it.

Principle 2: A source document is not yet instruction

A good policy can be a bad lesson.

A complete SOP can still overwhelm a new employee.

A technically accurate manual may contain many details that belong in a searchable reference rather than human memory.

That does not mean the source is poorly written.

It means reference information and instruction serve different jobs.

Turning workplace knowledge into learning requires transformation:

source → structure → outcome → explanation → example → practice → evidence

AI can accelerate parts of that transformation.

But summarization alone is not instructional design.

A system must decide which concepts are prerequisites, which details matter now, where examples are needed, what should remain visible as support, and what the learner should actually practice.

Principle 3: Learners must do something with knowledge

Information enters the learning environment.

Learning requires the person to work with it.

That work may include:

• recalling;
• explaining;
• comparing;
• classifying;
• predicting;
• choosing;
• practicing;
• correcting;
• applying.

Retrieval is particularly important because retrieving knowledge can strengthen later retention rather than merely test what was learned.

Source: Roediger and Butler, “The Critical Role of Retrieval Practice”. This changes how we think about quizzes.

A quiz is not valuable because it produces a percentage.

Its value depends on what the question asks the learner to do.

A weak question may reward recognition of a sentence that appeared moments earlier.

A stronger activity might ask the learner to choose an action in a new scenario and explain which principle applies.

The score is secondary.

The thinking is the point.

Principle 4: Performance during training is not the final target

A learner can look successful because the environment provides support.

The answer is visible.

Hints are available.

Examples are nearly identical.

Feedback appears before a serious attempt.

The learner can search the source while answering.

Those supports are not necessarily bad.

Beginners often need them.

But the conditions change what the result means.

Research distinguishes performance during practice from durable learning. A learner may perform well in the moment without building an equally durable capability.

Source: Soderstrom and Bjork, “Learning Versus Performance”. For us, this means learning evidence should eventually become richer than one immediate quiz score.

Useful questions include:

• Was support available?
• Could the learner perform independently?
• What remained after a delay?
• Could the learner handle a changed example?
• Did the capability appear in a realistic context?

A dashboard should describe what was measured rather than overstate what the metric proves.

Principle 5: Not everything belongs in memory

Good workplace learning is not a competition to memorize the company handbook.

Some knowledge should become readily retrievable from memory.

Other information should remain available in the environment.

An employee may need to remember:

• that a situation requires escalation;
• the first safety action;
• the principle behind a decision;
• when to stop and verify;
• where approved information can be found.

They may be better served by a checklist, searchable SOP, decision aid, or reference for:

• rare exceptions;
• frequently changing information;
• long tables;
• exact contact details;
• complex procedures used infrequently.

This distinction is especially important for AI.

AI can become external cognitive support.

The goal is not to eliminate reliance on tools. Professionals have always used manuals, checklists, calculators, documentation, and colleagues.

The design question is:

What capability should belong to the person, and what information should reliably belong to the system?

Principle 6: Evidence should match the claim

Every metric has a limit.

Opening a course supports an access claim.

Finishing it supports a completion claim.

A satisfaction survey supports a reaction claim.

An immediate quiz supports a claim about performance under those test conditions.

A delayed assessment can support a retention claim.

A changed scenario can provide transfer evidence.

Repeated workplace behavior can support a workplace-performance claim.

These distinctions are central to the Seven Levels of Learning Evidence framework developed from our Week 1 research.

The rule is simple:

Do not make a claim stronger than the evidence.

This also affects language.

We prefer:

Course completion: 94 percent.

over:

94 percent learned the material.

We prefer:

Immediate quiz performance: 87 percent.

over:

Mastery: 87 percent.

unless “mastery” is backed by a documented evidence model.

Honest measurement is not weaker marketing.

It is stronger learning engineering.

Principle 7: Human review remains part of the system

AI-generated training can be fast and useful.

It can also be incomplete, poorly sequenced, misleading, or inappropriate for a specific workplace.

LoreGraph's current workflow lets creators review and edit generated lessons, quiz questions, answers, and explanations before publishing.

Source: LoreGraph. We view that human control as important, especially for sensitive operational knowledge.

The person reviewing the course knows things the model may not:

• which exception creates real risk;
• which wording carries legal significance;
• which local process has changed;
• which example is unrealistic;
• which task an employee actually performs;
• which information should never be simplified.

AI should reduce the cost of creating and maintaining learning.

It should not remove responsibility for what gets taught.

Principle 8: Learning should connect back to work

A training system becomes more useful when the distance between the course and the job gets smaller.

Transfer means using what was learned in a new context. The National Academies treats the ability to extend knowledge beyond the original context as a central learning goal.

Source: National Academies, “Learning and Transfer”. For workplace training, that means practice should eventually resemble the decisions people actually face.

An employee learning a refund policy should handle customer cases.

A supervisor learning feedback skills should respond to difficult conversations.

A caregiver learning incident reporting should distinguish ambiguous situations and begin the correct process.

A technician learning troubleshooting should diagnose changed symptoms rather than repeat one memorized example.

Context also matters after training.

Research on workplace transfer has found that supervisor, peer, and organizational support can influence whether new capability is actually used.

Source: Hughes et al., “The Role of Work Environment in Training Sustainment”. Training does not operate in isolation.

Sometimes the right next step is a better lesson.

Sometimes it is a better checklist, workflow, manager behavior, tool, or process.

What these principles mean for LoreGraph

LoreGraph currently focuses on turning company documents into trackable training: upload the source, generate a structured course draft, review and edit it, assign it, and track learner progress and quiz results.

Source: LoreGraph. The longer-term learning problem is larger.

A modern learning system should increasingly be able to connect:

• source knowledge;
• intended capabilities;
• learning activities;
• practice;
• assessment evidence;
• learner progress;
• workplace context;
• content changes.

That direction is why we think of LoreGraph as more than an AI content generator.

The interesting question is not:

How many courses can AI create?

It is:

How can software help organizations turn changing knowledge into capability they can actually observe and improve?

What we are careful not to claim

These principles are a design philosophy and research direction.

They do not prove that every LoreGraph-generated course produces durable learning.

They do not mean a quiz score is a direct measure of workplace performance.

They do not mean AI can determine mastery simply because it generated the assessment.

And they do not mean every workplace problem should become training.

A missing tool, contradictory process, poor incentive, unclear responsibility, or unrealistic workload may require an operational solution rather than another course.

The standard should remain:

Make the learning claim match the evidence.

Next step

Take one piece of workplace training and run this eight-question review:

1. What should someone be able to do differently?
2. Is the source organized for reference or for learning?
3. What must the learner actively think or practice?
4. What support is available during performance?
5. What belongs in memory and what should remain searchable?
6. What does the current evidence actually prove?
7. Who reviews the training before people rely on it?
8. Where should the capability appear in real work?

If those questions are difficult to answer, adding more content probably is not the first solution.

Clarifying the learning system is.

Sources and further reading

• National Academies, How People Learn II: Learners, Contexts, and Cultures.
• Soderstrom and Bjork, “Learning Versus Performance: An Integrative Review”.
• Roediger and Butler, “The Critical Role of Retrieval Practice in Long-Term Retention”.
• National Research Council, “Learning and Transfer”.
• Hughes and colleagues, “The Role of Work Environment in Training Sustainment: A Meta-Analysis”.
• LoreGraph, AI-powered training from existing documents.