MindOS Learning Manual: Memory-for-Change State | Noticing What Changed Can Stop the Old Version From Winning

MindOS · Memory-for-Change State · Retrieve Old → Detect Change → Bind Old + New → Tag Source/Time → Recollect Change → Select Current Version → Delay → Return

Wait, What? Learning the New Version Can Make the Old Version More Dangerous—or More Useful

A learner studies one answer on Monday. On Thursday, the answer changes, the rule is refined, the teacher corrects an earlier simplification, or a newer source replaces an older one.

The learner understands the new version while it is visible.

Then the test comes.

The old answer appears first.

That does not mean the new learning failed to enter memory. It may mean two related versions are now competing and the learner did not preserve the relationship between them strongly enough to know what changed, when it changed, and which version belongs here now.

Memory research shows something even more interesting: under the right conditions, remembering the fact of change can turn an old memory from a competitor into a cue that helps the learner retrieve the new one.

Quick Answer

Owned learner job: when new information conflicts with or updates previously learned information, detect the change, integrate the old and new versions, preserve their temporal/source relationship, and later retrieve the version that applies while remembering that a change occurred.

Memory-for-Change State is not “forget the old thing and memorise the new thing harder.” Its central question is:

Can the learner remember the change relationship strongly enough that the old version helps identify the new one instead of silently competing with it?

The Scientific Pattern: Change Can Produce Interference or Facilitation

A classic memory-for-change paradigm presents one cue with an earlier response and later changes the response. Imagine:

• Earlier: KNEE — BONE
• Later: KNEE — BEND

When the learner encounters the later pair, the earlier response may come back to mind. That reminding can expose the change.

The important finding is conditional.

• Change detected and later recollected: memory for the newer response can be better than for comparable control material. The older memory becomes part of an integrated trace and can support what researchers call proactive facilitation.
• Change detected but not later recollected: the older response may remain highly accessible and interfere with retrieval of the newer response, producing proactive interference.

This is why “noticing a correction” is not enough. The learner must later be able to reconstruct the change episode itself.

The Owned Boundary: This Is Not Ordinary Correction

Correction State owns the general repair operation: identify an error, understand feedback, reconstruct the corrected route, and prove the repair independently.

Memory-for-Change State owns something narrower and more temporal. The old information may have been reasonable, correct in an earlier context, deliberately simplified, or simply learned first. The learner must preserve:

• what the earlier version was;
• what the later version became;
• what feature changed;
• which source, time, task or condition belongs to each version;
• which version should control the current response.

The goal is not merely a corrected answer. It is a usable old → change → new relationship.

The Owned Boundary: This Is Not Comparison State

Comparison State asks what is similar and different between two objects, ideas or procedures.

Memory-for-Change adds temporal order and updating. Two versions are not simply side-by-side. One was active earlier; another came later; the learner must preserve which is which and why the transition matters.

The Owned Boundary: This Is Not Retrieval-Induced Forgetting

Retrieval-Induced Forgetting State concerns what can happen when selective retrieval strengthens one response while related unpractised competitors become less accessible.

Memory-for-Change instead concerns updating across episodes: an older and newer version share a cue, the learner notices a change, and later performance depends partly on whether that change relationship can be recollected.

The Owned Boundary: This Is Not External Version Control

The Student/Studying Interface: Version-Control Interface owns the external problem of studying from the wrong file, old syllabus, outdated worksheet or superseded document.

Memory-for-Change State begins after the learner has encountered multiple versions. It asks whether the internal memory system keeps their relationship straight.

Observable Learner Signatures

• The learner repeatedly produces an earlier answer after accurately studying the newer one.
• They can recognise the new version when shown it but cannot retrieve which version is current.
• They say, “I remember both, but I cannot remember which one came later.”
• A teacher correction makes sense in class, yet the original misconception returns several days later.
• The learner merges old and new statements into a hybrid that was never taught.
• They remember that “something changed” but cannot reconstruct what changed.
• They remember both versions but lose the source: textbook, teacher clarification, newer specification, experiment result or earlier draft.
• An earlier method keeps winning under time pressure even though the learner knows a refined method when prompted.
• New information is learned as an isolated replacement rather than linked to what it superseded.

These signs are not sufficient to diagnose Memory-for-Change State. They can also arise from weak initial learning, ordinary forgetting, cue dependence, poor source monitoring, misconception, task switching or simple exposure to contradictory sources. MindOS therefore tests the relationship instead of applying a label.

Discrimination Test 1: Can the Learner State Both Versions?

Ask:

• What did you believe or learn first?
• What did you learn later?
• What exactly changed?

If the learner cannot retrieve the newer information at all, the first weak link is ordinary learning or retrieval. If both versions are available but their relationship is confused, Memory-for-Change becomes more plausible.

Discrimination Test 2: Can the Learner Preserve Order?

Present the two versions without labels and ask which came first and which came later. Then ask what evidence supports the order.

This matters because remembering two pieces of information is not the same as remembering their temporal relationship.

Discrimination Test 3: Can the Learner Preserve Source?

Sometimes two versions are both valid under different authorities or conditions. Ask:

• Which came from the older worksheet?
• Which came from the teacher’s later clarification?
• Which applies to this syllabus?
• Which belongs to the simplified model and which to the refined model?
• Which claim came from which source?

If source is lost, the learner may retrieve a real memory but use it in the wrong place.

Discrimination Test 4: Is the Old Version Wrong—or Contextually Right?

Do not automatically delete the earlier representation.

Some learning progresses through useful simplifications. A Primary Science explanation may later be refined at Secondary or JC level. A Mathematics shortcut may work only under certain conditions. A historical interpretation may be revised when new evidence or a different analytical frame is introduced.

The learner needs to know whether the change means:

• replacement: the old claim was wrong;
• refinement: the old claim was incomplete;
• context split: both versions are valid under different conditions;
• source update: a newer authority supersedes an older one;
• representation change: the same underlying idea is now expressed differently.

That classification changes the repair.

The MindOS Memory-for-Change Protocol

Step 1 — Retrieve the Old Version Before Showing the New One Again

Ask the learner what was previously known. This creates a visible baseline and can make the discrepancy easier to detect.

If the old version is unsafe misinformation or highly consequential, use judgment about whether direct repetition is appropriate. Educational memory practice should not amplify harmful material unnecessarily.

Step 2 — Put Old and New in Contact

Show the newer information and ask the learner to identify the changed feature rather than simply reread the replacement.

“What is different from what you knew before?”

Step 3 — Classify the Change

Replacement? Refinement? New condition? New source? New time? New exception?

This prevents the learner from encoding only two disconnected statements.

Step 4 — Build an Integrated Statement

Require one sentence that contains the relationship:

“I first learned X; later I learned Y; Y differs because ___; for this task, Y applies when ___.”

This is not intended as a permanent script. It is a scaffold for binding the episodes.

Step 5 — Tag Source, Time or Condition

Add the minimum metadata needed to choose correctly later:

• old syllabus / current syllabus;
• initial hypothesis / later evidence;
• simplified model / refined model;
• teacher draft / final method;
• general rule / exception condition;
• earlier source / newer verified source.

Step 6 — Close Both Versions

Ask the learner to reconstruct:

• the old version;
• the new version;
• the change;
• which now applies and why.

The fourth answer is the most important. Memory for change is useful because it guides selection.

Step 7 — Change the Cue

Ask the same relationship through a new question, diagram, example or wording. This prevents the integrated memory from becoming tied to one teaching script.

Step 8 — Return After Delay

Several days later, ask for the current response first. Then ask whether an earlier version existed and what changed.

If the correct current answer survives but the learner no longer remembers the old version, that may be perfectly acceptable for some tasks. If future interference remains likely, preserving the change relationship may still be useful.

Worked Example: A Science Model Becomes More Precise

A younger learner may initially use a simplified statement such as “particles in a solid do not move.” Later, the model is refined: particles in a solid vibrate about fixed positions.

If the later statement is simply added, the old wording may continue to compete. A stronger update sequence is:

1. Retrieve the earlier statement.
2. Identify the changed feature: “do not move” becomes “vibrate but remain around fixed positions.”
3. Classify the change as a refinement of the particle model.
4. Explain why “fixed positions” does not mean “motionless particles.”
5. Draw or explain the refined model without looking.
6. Use the model in a new heating question.
7. Return later and ask what changed from the earlier model.

The learner is not being asked to keep a wrong idea alive for its own sake. The earlier model becomes a contrastive landmark that helps preserve the refinement.

Worked Example: Two Mathematics Conditions

A learner first memorises a procedure in a narrow set of questions. Later, the teacher introduces a condition showing that the procedure does not apply universally.

The weak update is:

“Forget the old rule. Use this one.”

The stronger update is:

• state where the old rule worked;
• identify the new condition;
• give one problem where the old route is valid;
• give one near problem where it fails;
• explain the deciding cue;
• mix both problem types later.

The learner now owns not just two procedures but the boundary that determines which one applies.

Worked Example: Changing Information From Sources

A learner reads an older source and later encounters a newer correction or update. The educational job is not to repeat the older claim until it becomes familiar. It is to preserve enough source information to know:

• what the earlier source said;
• what the later source changed;
• why the later source should control the current answer;
• whether the change is established fact, updated measurement, revised interpretation or unresolved disagreement.

This is particularly important when studying fast-changing domains. The learner needs epistemic versioning, not just a pile of remembered statements.

Why Reminding Can Help

The Memory-for-Change framework proposes that new information can cue retrieval of an earlier episode. If the old and new information are simultaneously accessible, the learner can detect the discrepancy and encode an integrated representation containing both versions plus their relationship.

Later, if the learner recollects that integrated change episode, the earlier memory need not simply compete. It can help cue the newer one and preserve temporal order.

Experiments using reminders before changed information support this logic. Intact reminders that effectively reinstated earlier information improved the benefits associated with recollecting change compared with weaker reminder conditions.

Why Reminding Can Also Hurt

Retrieving an old response also makes that response accessible again. If the learner detects the update during study but later fails to recollect the change relation, the strengthened older response can interfere.

This gives a critical safeguard:

Do not remind the learner of old information without building and later testing the old-to-new relationship.

Simply repeating the misconception before a correction is not automatically beneficial.

Connection to Misinformation Correction

Memory-for-Change research has also been applied to misinformation corrections. Studies using news-like materials show why reminders can produce mixed outcomes: reinstating the earlier claim can make the conflict easier to notice and support integration with the correction, but it can also increase familiarity with the earlier information if the correction relationship is not later recollected.

This is a useful educational boundary, but MindOS does not reduce misinformation correction to one simple formula. Belief, trust, source credibility, repeated exposure, motivation and social context add complexities beyond paired-associate learning.

A 2025 eye-tracking study of misinformation reminders found that reminders could enhance belief updating and memory for corrections, with evidence consistent with increased attention to conflict and integration. The practical lesson remains conditional: the correction needs to be trustworthy, the discrepancy needs to be processed, and later memory needs to preserve the correction relationship.

How Do We Know?

The modern Memory-for-Change account was developed through experiments by Jacoby, Wahlheim and colleagues examining how detection and recollection of change influence proactive interference and source/list discrimination. Their central result is unusually useful for learning design: changed information can produce either interference or facilitation depending on whether the change relation is later recollected.

Follow-up experiments found that reminders can support memory updating when they successfully cue retrieval of existing information and the old/new relationship is integrated. Later work has extended the framework to more naturalistic materials, including misinformation and correction.

• Wahlheim (2014), proactive effects of memory and the role of change recollection
• Wahlheim & Jacoby, testing can counteract proactive interference by integrating competing information
• Wahlheim, Smith & Delaney (2019), reminders can enhance or impair episodic memory updating
• Memory-for-Change framework applied to misinformation correction
• Misinformation reminders, attention and correction memory (2025)

What the Evidence Does Not Prove

• It does not prove that every correction should begin by repeating the old answer.
• It does not prove that noticing a discrepancy automatically improves later memory.
• It does not prove that the old memory must be erased.
• It does not prove that all classroom interference is caused by failure to recollect change.
• It does not prove that laboratory word-pair effects transfer unchanged to complex conceptual learning.
• It does not prove that reminders are safe when the old information is harmful, emotionally loaded or socially amplified.
• It does not prove that later information is automatically more correct merely because it is newer.

The safe inference is narrower: when old and new information compete, detecting and later recollecting their relationship can improve memory updating and can sometimes transform proactive interference into facilitation.

Source Authority Gate: Newer Is Not Automatically Truer

Memory-for-Change is about memory updating, not truth determination.

Before telling the learner that the later version supersedes the earlier one, verify why:

• Has the syllabus or specification actually changed?
• Is the new source more authoritative?
• Was the earlier statement simplified rather than false?
• Are the two sources describing different conditions?
• Is the issue genuinely unsettled?
• Was the “correction” generated by an unreliable tool?

A memory system can faithfully update toward wrong information if the authority layer is wrong. Verification comes before integration.

AI Boundary: A New Answer From AI Is Not Automatically an Update

AI can produce a different answer every time a learner asks. Difference alone does not establish correction.

A safer learner sequence is:

1. state the learner’s current answer;
2. ask AI for an alternative or critique;
3. verify the disputed claim against an appropriate authority;
4. identify exactly what changed;
5. tag why the change is justified;
6. close AI;
7. reconstruct the updated answer independently;
8. return later and test which version comes back.

If AI supplies both the new answer and the reason it should be trusted, the learner may never perform the source-control operation.

Staged Practice

1. Detect supplied changes: show old and new together; learner identifies the changed feature.
2. Classify the change: replacement, refinement, context split, source update or representation change.
3. Integrate: learner states the old → change → new relationship.
4. Source/order tag: learner identifies which version came first and why the later/current version applies.
5. Remove both: learner reconstructs both versions and the change from memory.
6. Changed cue: ask through a new example or representation.
7. Delayed return: retest after days.
8. Spontaneous updating: learner independently notices future conflicts and creates the relationship without being told to do so.

Scaffold Fade

At first, the tutor may place OLD and NEW labels side by side. Next, only the new version is shown and the learner must retrieve the old one. Later, the learner must detect an unannounced change. Finally, the learner encounters a new source independently and decides whether it is a replacement, refinement, disagreement or irrelevant difference.

The mature state is not “always look for changes.” It is the ability to notice meaningful change and preserve enough relationship information to select correctly later.

Transfer Test

Move to a different subject and present a familiar cue with a changed response. Do not announce that the old information will compete.

Ask the learner to:

• notice the discrepancy;
• retrieve the earlier version;
• explain the change;
• identify source/time/condition;
• use the current version in a novel task;
• return later without the comparison in view.

If the learner performs only when OLD and NEW are explicitly labelled, the operation remains scaffold-dependent.

Delayed Return Test

Several days later, test the current answer first. Then ask:

• Was there an earlier version?
• What was it?
• What changed?
• Why does the current version apply?
• Can you identify a case where the earlier version would still be appropriate, if any?

The strongest receipt is not mere recall of two statements. It is accurate selection plus preserved relationship information.

Common Misconceptions

• “Just forget the old answer.” Suppression may not be necessary or possible, and the old answer can become a useful contrast if the change relationship is preserved.
• “If the learner noticed the correction, it is fixed.” Detection during study is not the same as recollection of change later.
• “The latest information must be right.” Recency is not authority.
• “Remembering both versions means confusion.” It can instead be a high-resolution memory state if source, order and conditions are preserved.
• “Repeating the old misconception before correcting it always helps.” Reminder effects are conditional and can increase interference or familiarity when integration fails.
• “This is just proactive interference.” Memory-for-Change explains when the same old/new relationship can shift from interference toward facilitation.

Parent and Tutor Teaching Guide

When a child says, “But I learned it the other way before,” do not immediately answer, “Forget that.”

Ask:

• What did you learn before?
• What are you learning now?
• What changed?
• Why did it change?
• Is the old version wrong, incomplete or valid somewhere else?
• Which version applies to this question?
• Can you close the book and explain the transition?

This respects the learner’s real memory rather than pretending the earlier learning never happened.

MindOS Direction Graph

Old knowledge retrieved → new information encountered → meaningful change detected → authority checked → change classified → old/new integrated → source/time/condition tagged → model removed → current version retrieved → changed cue → delayed return → update future selection.

If the new information itself cannot be retrieved, route to Retrieval State. If the learner cannot explain what differs, route to Comparison or Concept-Boundary State. If the old version is simply an identified error requiring reconstruction, use Correction State. If the learner studies from the wrong external document, use the Version-Control Interface. If one related memory crowds out others after selective practice, use Retrieval-Induced Forgetting State.

---

MindOS rule: updating learning is not always replacement. When knowledge changes, the learner may need to remember the transition itself—because a remembered change can turn yesterday’s competitor into a cue for today’s correct answer.