MindOS Learning Manual: Source-Monitoring State | You Can Remember the Fact and Forget Where It Came From

MindOS · Source-Monitoring State · Remember Content → Ask Where It Came From → Keep Competing Sources Alive → Discriminate → Bind Provenance → Retrieve Content + Source → Verify When Needed → Fade Prompts → Transfer → Return

Wait, What? You Can Remember the Fact Correctly and Still Remember Its Source Wrong

A student says, “My teacher told us this.”

The statement is familiar. The student is confident. The content may even be correct.

But the teacher never said it.

The student read it in a class chat, heard a classmate repeat it, saw a similar sentence online, or generated the inference themselves. Over time, the content survived while its origin blurred.

This is not the same learning problem as forgetting the fact. It is a source-monitoring problem: the learner has information but is uncertain, mistaken or overconfident about where that information came from.

Quick Answer

Owned Learner Job: preserve or reconstruct enough provenance around important information to know whether it came from the textbook, teacher, peer, experiment, source document, AI tool, prior memory or the learner’s own inference—and use that distinction when credibility, verification, citation, correction or later help-seeking depends on it.

The RFE is not “memorise a citation for every sentence.” It is:

When the origin of information changes what the learner should trust, verify, cite, compare or do next, the source must remain recoverable enough to guide that decision.

Content Memory and Source Memory Are Not the Same Measurement

Imagine a learner studies twelve statements from three different sources.

Later, two questions can be asked:

Item memory: Was this statement studied before?
Source memory: Where did this statement come from?

A learner can answer the first correctly and the second incorrectly. That separation is central to source-monitoring research. Remembering what was encountered does not guarantee remembering who said it, where it appeared, whether it was seen or imagined, or which context gave it meaning.

Why Source Memory Matters in Learning

Sometimes the source is irrelevant. If a learner knows that water freezes near 0°C under ordinary atmospheric conditions, they do not need to recover the exact lesson in which they first learned it before using the fact.

But source becomes load-bearing when the next decision depends on provenance:

one source is authoritative and another is speculative;
two sources disagree;
a statement came from a classmate rather than the teacher;
a remembered “fact” was actually the learner’s own inference;
an AI tool generated the claim and it still requires verification;
the learner must cite or quote a source;
an old source may be superseded;
the learner needs to return to the right person or document for clarification.

The operation is therefore selective. MindOS does not turn every learning episode into archival work. It protects source information when losing it would damage a later learning decision.

The Owned Boundary: This Is Not Memory-Conformity State

Memory-Conformity State owns what can happen when another person’s report changes a learner’s later memory report.

Source-Monitoring State owns the attribution problem itself: which origin does this remembered information belong to? Memory conformity can create a source-monitoring challenge, but source-monitoring errors can also occur without social influence.

The Owned Boundary: This Is Not Illusory-Truth State

Illusory-Truth State owns the risk that repetition increases perceived truth through familiarity.

Source monitoring asks a different question. A learner may know that a claim feels familiar yet still need to ask: “Familiar from where?” Repetition can make that question harder because the content becomes highly accessible while source-specific details remain weak.

The Owned Boundary: This Is Not Claim–Evidence Reasoning

Claim–Evidence Reasoning State asks whether a fact actually supports a claim and how the reasoning link works.

Source monitoring comes earlier or alongside that operation: what source supplied this information, and how certain am I about that attribution? A perfectly remembered source can still provide weak evidence; strong evidence can still be misattributed to the wrong source.

The Owned Boundary: This Is Not the Citation Interface

The Student/Studying Interface can own whether a citation, reference manager, copied quotation or study artifact preserves the external source correctly.

MindOS owns the learner operation: can the learner distinguish what they know from where they think they learned it, notice uncertainty, and decide when external verification is required?

How Source Monitoring Works

The classic Source Monitoring Framework developed by Marcia Johnson, Shahin Hashtroudi and D. Stephen Lindsay describes source judgments as reconstructive decisions. A memory does not always carry an explicit permanent label saying “SOURCE: textbook page 42.” Instead, people evaluate features associated with the memory—perceptual detail, contextual information, thoughts, feelings, semantic knowledge and records of cognitive operations—and infer the most likely origin.

That process can be accurate. It can also fail.

If two sources were highly similar, if source details were never encoded, if the content was repeated many times, or if the learner relies on familiarity rather than source-specific evidence, the item may survive while provenance degrades.

Observable Learner Signatures

The learner remembers a correct fact but cannot say whether it came from a teacher, textbook or peer.
A classmate’s suggestion is later reported as if it appeared in the original notes.
An AI-generated explanation becomes “something I already knew” after several days.
The learner remembers a quotation but attributes it to the wrong text or speaker.
Two websites with similar layouts collapse into one remembered source.
The learner says “I read somewhere that…” but cannot recover enough provenance to evaluate the claim.
The learner is highly confident about the content and equally confident—but wrong—about the source.
Source accuracy falls much more sharply than item recognition after a delay.
A more distinctive source cue immediately restores the correct attribution.
The learner asks the wrong peer for clarification because they cannot remember who supplied the original information.

These are educational observations, not diagnoses. Source confusion can occur in ordinary memory and does not by itself imply a memory disorder or other clinical condition.

Keep Multiple Explanations Alive

“You forgot the source” is not yet a sufficient diagnosis. Several different weak links can produce the same behaviour:

Source never encoded: attention went entirely to content.
Source encoded weakly: the learner noticed it but did not bind it to the claim.
Sources were too similar: the distinguishing features collapsed.
Retrieval cue is weak: source information exists but is not accessible under the current prompt.
Content was repeated: fluency is high while provenance is not.
Another person’s answer intervened: route toward Memory-Conformity State.
The artifact lost provenance: route toward the relevant Student/Studying Interface.
The learner misunderstood the concept: source work will not repair the content representation.

Discrimination Test 1: Separate “Did You Learn This?” From “Where Did It Come From?”

Present several studied and unstudied statements.

First ask whether each statement was encountered. Only after that decision ask for the source.

This separates item-memory failure from source-memory failure. If the learner does not recognise the information at all, source-monitoring may not be the earliest weak link. If item memory is strong but source attribution is poor, the state is much cleaner.

Discrimination Test 2: Ask for the Evidence Behind the Source Judgment

Do not accept “It just feels like the teacher said it.” Ask:

What detail makes you think that?
Do you remember seeing it or hearing it?
Was it in the slide, the textbook or the discussion after the lesson?
Could you be remembering your own inference?

The learner is not required to reconstruct impossible detail. The goal is to distinguish source-specific evidence from generic familiarity.

Discrimination Test 3: Change the Cue

If “Where did this come from?” fails, provide one non-leading contextual cue: topic, lesson setting, document type or surrounding example.

If source accuracy recovers sharply, retrieval access may be the bottleneck. If the learner still cannot discriminate among sources, the source representation may never have been well differentiated.

Discrimination Test 4: Separate Confidence From Accuracy

After each source judgment, ask for confidence.

Recent reality-monitoring research provides an important warning: incorrect source judgments can sometimes carry confidence comparable to correct ones. Confidence is therefore evidence about the learner’s subjective state, not proof that the source attribution is correct.

Discrimination Test 5: Verify Against the External Record

When notes, browser history, a textbook, shared document or original conversation record exists, compare the learner’s source memory with the record.

This is especially useful during training because source monitoring improves when errors become visible. The point is not to shame an inaccurate memory. It is to calibrate the learner’s future decision to verify when provenance is uncertain.

The Smallest Useful Source-Monitoring Operation

For information where source matters, bind a compact source tag during learning:

CLAIM → SOURCE → WHY SOURCE MATTERS

For example:

“Exam format → official syllabus document → authoritative requirement.”
“Possible explanation → classmate → useful hypothesis, not yet verified.”
“Mechanism → textbook → check edition if specification changed.”
“Worked solution → AI tool → verify method and conditions independently.”

Do not attach ten metadata fields when three words would preserve the necessary distinction. Source monitoring should reduce future uncertainty without overwhelming the learning target.

The MindOS Source-Monitoring Protocol

Step 1 — Decide Whether Source Is Load-Bearing

Will provenance affect trust, citation, verification, interpretation, correction or later help-seeking? If not, do not create unnecessary source-memory work.

Step 2 — Encode One Distinguishing Source Feature

Use a feature that actually separates sources: teacher versus peer, official document versus commentary, experiment versus inference, Source A versus Source B.

Step 3 — Bind the Claim to the Source

Do not merely remember “Source A exists.” Connect the specific information to the source that supplied it.

Step 4 — Retrieve Content First, Then Source

This makes the two measurements visible. The learner should be able to say, “I remember the claim, but I am uncertain about where it came from.” That is a better learning state than inventing provenance.

Step 5 — Compare Near-Neighbour Sources

If two sources are easily confused, place them side by side and ask what distinguishes them. Similarity needs an explicit boundary.

Step 6 — Verify When Consequences Are High

Do not demand perfect internal source memory. When the stakes justify it, use the external record. Mature source monitoring includes knowing when memory is insufficient.

Step 7 — Fade the Source Prompt

Remove visible SOURCE labels. Ask the learner to decide independently when provenance matters and to create a source tag only where useful.

Step 8 — Change the Environment

Move from classroom notes to a website, AI conversation, group discussion or research task. The operation should travel.

Step 9 — Return After Delay

Test content and source separately several days later. The final receipt is not perfect provenance for everything; it is accurate discrimination where source changes the next decision.

Worked Example: Science

A learner hears three explanations for an unexpected experimental result: one from the teacher, one from a classmate and one generated by an AI study tool.

Two days later, the learner remembers all three explanations but says the AI-generated hypothesis was “what the teacher told us.”

Do not merely correct the label. Rebuild the learner operation:

retrieve each explanation;
attribute each source;
state confidence;
check the lesson record;
decide which explanation is actually supported by the evidence;
return later without the source table.

The learning target is not obedience to the teacher. It is provenance-aware scientific reasoning.

Worked Example: English

A student remembers an interpretation of a poem but cannot tell whether it came from the poem itself, the teacher’s commentary or their own inference.

Source monitoring protects the distinction:

text evidence: what the poem explicitly gives;
teacher interpretation: a model reading;
learner inference: a claim that still needs textual support.

This strengthens interpretation because the learner can now explain what comes from the source text and what has been constructed around it.

Worked Example: Mathematics

A student uses a shortcut they remember “from tuition.” The tutor does not recognise it. The student later discovers it came from a social-media video.

The issue is not that social media is automatically wrong. The issue is that the source was misattributed, which prevented the learner from asking the right verification questions about assumptions and validity.

After verification, the learner must still solve changed problems without the source label. Source monitoring supports method selection; it does not replace mathematical understanding.

AI Boundary: Did the Learner Know It, or Did the Model Say It?

AI makes source monitoring unusually important because fluent dialogue blurs boundaries between:

the learner’s prior knowledge;
the learner’s inference;
the model’s generated explanation;
an external source quoted by the model;
a source the learner independently checked.

A generated answer can become familiar enough that the learner later experiences it as self-known.

Use one simple technology rule:

If the tool supplied a load-bearing claim, preserve that fact until the learner has independently verified, reconstructed and used the knowledge.

Then fade the provenance scaffold where it is no longer needed. Technology succeeds when the learner’s capability survives after the conversation is closed.

How Do We Know?

The foundational Source Monitoring Framework was set out by Johnson, Hashtroudi and Lindsay in Psychological Bulletin in 1993. It treats source attribution as a reconstructive judgment based on qualitative characteristics of remembered experience and flexible decision criteria rather than as a perfect source label permanently attached to every memory.

A 2021 review in Psychology of Learning and Motivation summarised behavioural methods and findings on remembering and reconstructing episodic context, reinforcing the distinction between item information and its source.

A scoping review first published in 2024 and appearing in the 2025 British Journal of Developmental Psychology identified 141 studies measuring source monitoring in children and documented multiple ways researchers separate recognition from source discrimination.

Recent educational work also demonstrates why the operation matters beyond laboratory word lists. A 2026 Frontiers in Education study of collaborative learning examined source memory for information supplied by different learning partners and argued that remembering who supplied information can support later credibility judgments and help-seeking. The study also reported that source memory was generally imperfect, reinforcing the need not to assume content learning preserves provenance automatically.

Research on misinformation adds an important boundary. A 2023 experiment and meta-analysis found that repetition increased suggestibility, while varying the number of misinformation sources did not independently increase the effect. The authors interpreted the pattern as evidence that retrieval fluency can dominate source-specifying information. This is useful precisely because it prevents a simplistic rule that “more sources” automatically means stronger source memory.

Evidence Boundary

Source monitoring is reconstructive and fallible; an attribution is not guaranteed to be correct because it feels vivid.
Content memory and source memory can dissociate.
Confidence in a source judgment is not the same as source accuracy.
Source-memory paradigms simplify real educational environments; classroom and online learning involve more complex sources and repeated exposures.
Not every piece of information requires source rehearsal. The operation is most useful when provenance changes trust, verification, citation, interpretation or action.
Source monitoring does not determine whether a source is credible. Credibility evaluation is a separate reasoning job.
Source confusion in ordinary learning is not a clinical diagnosis.
A learner should be allowed to say “I know the content but I am not sure where I learned it.” Uncertainty is preferable to fabricated provenance.

Common Misconceptions

“If the fact is correct, the source does not matter.” Sometimes true; sometimes dangerously false. Ask whether provenance changes the next decision.
“A confident source memory is probably accurate.” Confidence and accuracy can diverge.
“Good students remember where everything came from.” Source memory is selective and imperfect even when content learning is strong.
“Source monitoring means memorising citations.” Citations are one external provenance system; source monitoring is the learner’s attribution operation.
“If AI gave the answer, the learner did not learn it.” Not necessarily. The key question is whether the learner can later reconstruct, verify and use the knowledge independently.
“One trusted source removes the need to think.” Source credibility and claim validity remain distinct questions.

Staged Practice

Visible provenance: study a small set of claims with source labels present.
Content/source split: retrieve the claim first and source second.
Near-source discrimination: compare two similar sources.
Confidence calibration: rate source confidence before checking.
External verification: compare attribution with the original record.
Mixed sources: teacher, peer, text, AI, self-generated inference.
Prompt fade: remove the visible SOURCE field.
Selective provenance: learner decides which information actually needs a source tag.
Changed environment: repeat the operation in a new subject or tool.
Delayed return: test content and source separately after time.

Scaffold Fade

At first, the tutor asks, “Where did that come from?” and provides visible source labels.

Next, the learner must retrieve the source before looking. Then labels disappear. Eventually, the learner should independently notice: “This claim matters enough that I need to know its provenance,” or “The source is uncertain, so I should verify before relying on it.”

The final target is not a child who constantly asks an adult to confirm every source. It is a learner who knows when provenance is load-bearing and acts accordingly.

Transfer Test

Move the learner to a new setting with several information channels—for example, a textbook, a class discussion and an AI study tool.

Do not instruct them to track sources. Later ask:

Which claims require provenance to use safely?
Can you retrieve the source?
What source-specific evidence supports your attribution?
Where are you uncertain?
Which uncertain claims require external checking?

Transfer is present when the learner initiates the source-monitoring operation without a source worksheet.

Delayed Independent Return Test

Several days later, present the information in a different order and without source labels.

A strong return has four parts:

content is still usable;
important source distinctions remain accurate enough;
the learner can express uncertainty instead of guessing;
the learner knows when to verify externally.

If source accuracy disappears as soon as the table disappears, the scaffold did not yet become learner capability.

Examination Implications

Source monitoring is not equally important in every examination. Many Mathematics and Science questions test concepts or procedures without asking where the learner learned them.

It becomes directly relevant when the assessment requires quotation attribution, source comparison, document-based reasoning, experimental provenance, historical evidence, textual evidence or evaluation of competing accounts.

Even when source is not directly marked, the operation can still protect revision by preventing a peer’s guess, an old note or an AI answer from silently becoming “what the syllabus says.”

Parent and Tutor Teaching Guide

When a learner produces a claim, avoid turning every conversation into interrogation. Use source questions only where they improve the next decision:

“What do you know?”
“Where do you think that came from?”
“What makes you think that was the source?”
“Could it have come from another place?”
“Does the source matter for this question?”
“How confident are you about the source?”
“Should we verify, or is this source detail unnecessary?”
“Can you still explain the idea after we close the source?”

The learner should become more epistemically careful without becoming paralysed. Good source monitoring supports learning; it does not turn every remembered sentence into a forensic investigation.

MindOS Direction Graph

Claim remembered → does source matter? → no: continue learning → yes → source known? → uncertain: keep alternatives alive → retrieve source-specific detail → compare near sources → verify if consequential → use claim with provenance → fade source prompts → changed setting → delayed return.

If another person’s answer changed the remembered content, route to Memory-Conformity State. If repetition is making a claim feel true, route to Illusory-Truth State. If the learner needs to evaluate what a fact actually supports, use Claim–Evidence Reasoning State. If the information itself must be found or independently checked, use Search State.

MindOS rule: a remembered claim and a remembered source are different achievements. When provenance changes what the learner should trust, verify, cite or do next, learning is stronger when both survive—and stronger still when the learner knows when memory is not enough.