# Horizontal and Vertical Far Transfer Protocol

## Purpose

This protocol defines how IQMindware tries to make transfer testable rather than assumed.

The programme does not treat better in-game performance as proof of far transfer. A training gain is treated as transfer-relevant only when it shows evidence of portability under changed conditions, supports a harder reasoning demand, or later carries into an external outcome check.

Current public programme scope:

- Zone Pulse sets state, readiness, and dose.
- Capacity Gym trains run-time control through adaptive N-back and relational N-back.
- Reasoning Gym asks the user to apply that control to relation matching, constraint solving, rule use, and must-follow inference.
- Tracker checks whether progress appears beyond the training games through pre/post and interval measures.

Future coaching-linked scope:

- Real-world missions will add a further validation layer by asking whether candidate gains support concrete work, study, decision, comprehension, argumentation, or planning tasks outside the app.
- This mission layer is explored further in live coaching and should be described as in development until mission evidence is collected and summarised.

## Definitions used here

### Transfer

Transfer means carry-over from one trained condition to another condition. It is not the same as repeating the same trained task more fluently.

### Near transfer

Near transfer means improvement on very similar tasks, usually with similar rules, stimuli, timing, and response demands.

### Far transfer

Far transfer means carry-over under more changed conditions: changed surface form, changed rule wrapper, changed reasoning demand, delayed re-check, or eventually a real-world task outside the game.

### Horizontal transfer

Horizontal transfer means testing the same underlying control demand across different wrappers at roughly the same level of complexity.

In IQMindware this means the user should not merely become fast inside one familiar game. The same working-memory, attention, inhibition, updating, or relational-control demand is moved across letters, locations, objects, relations, real-world meanings, nonsense meanings, and alternate rule wrappers.

### Vertical transfer

Vertical transfer means using a lower-level trained capacity to support a higher-level cognitive demand.

In IQMindware this means Capacity Gym is not an endpoint. Capacity training is intended to support Reasoning Gym, where the user has to use controlled attention and working memory for relational binding, rule use, inference, constraint solving, and fluid reasoning.

## Why the protocol is conservative

The cognitive-training literature gives a clear warning: people often improve on trained tasks, sometimes improve on closely related tasks, and much less reliably improve on distantly related tasks or everyday outcomes.

For that reason, IQMindware uses a conservative proof posture:

- Do not claim transfer from practice alone.
- Do not count one-game score improvement as far transfer.
- Build wrapper changes and reasoning probes into training.
- Separate training performance from Tracker outcomes.
- Treat real-world mission transfer as a later evidence layer, not as already proven.

## Current programme chain

### 1. Zone Pulse to training route

Zone Pulse uses the Multiple Function Task, masked version (MFT-m), as a daily cognitive-control capacity check.

Its role in the transfer chain is readiness gating:

- avoid heavy learning when the user is out of band;
- adjust route and dose for the day;
- reduce the risk that noisy state produces noisy training;
- route the user into capacity and reasoning targets that fit current readiness.

This is the first vertical step: state regulation supports better quality capacity training.

### 2. Capacity Gym horizontal transfer

Capacity Gym trains run-time control through adaptive N-back and relational N-back.

Horizontal transfer is targeted through:

- wrapper swaps;
- target-modality swaps;
- changes in surface form;
- changes in speed or load;
- relational variants that preserve the control demand while changing the representational wrapper.

The practical rule is stick-then-swap:

1. Stay with one wrapper long enough for a stable strategy to form.
2. Change one dial at a time: wrapper, speed, interference, or N-back level.
3. Test whether performance survives the changed wrapper.
4. If the swap holds, count it as a portability signal.
5. If the swap collapses, treat it as thin automation exposed and return to stabilisation.

The goal is not variety for its own sake. The goal is to test whether a control policy survives transformation.

### 3. Capacity Gym to Reasoning Gym vertical transfer

Capacity Gym is designed to improve the run-time control available for reasoning:

- holding active representations;
- resisting distractors;
- updating bindings;
- keeping task rules active;
- avoiding premature response habits;
- managing speed and accuracy under load.

Reasoning Gym then increases the cognitive demand. It asks the user to apply those control resources to:

- relation matching;
- constraint solving;
- rule selection;
- must-follow inference;
- meaning swaps;
- surface-form swaps;
- boundary and trap probes.

This is the core current vertical transfer route:

Zone readiness -> Capacity control -> Reasoning inference -> Tracker re-check.

### 4. Reasoning Gym horizontal transfer

Reasoning Gym also has a horizontal transfer layer.

The same inference demand should be tested under changed meanings and surface forms. A user should not only learn one fixed relation family or one visual presentation. The programme therefore rotates the wrapper while preserving the underlying reasoning demand.

Examples:

- same-relation matching across different story or object wrappers;
- real/nonsense meaning swaps;
- relation binding with changed labels;
- must-follow conclusions where surface content changes but logical demand remains;
- boundary probes where familiar shortcuts fail.

### 5. Tracker and delayed checks

Tracker separates training scores from evidence checks.

Current measurement layers:

- SgS-12 A/B for brief pre/post reasoning snapshots.
- Psi-CBS for applied cognitive bandwidth trends.
- MFT-m / Zone Pulse for cognitive-control capacity in bits per second.

The protocol treats these as evidence checks, not as training credit. A training score is stronger when it is accompanied by:

- completed Capacity and Reasoning targets;
- low wrapper dependence;
- delayed re-checks;
- pre/post reasoning signal;
- applied cognition trend signal;
- future mission evidence where available.

## Future mission-linked vertical transfer

The next intended vertical layer is real-world mission validation.

In that layer, the user would apply candidate gains to a concrete external task, such as:

- comprehension of a difficult text;
- argument evaluation or construction;
- decision-making under trade-offs;
- planning and execution under constraints;
- AI-assisted work where the human must maintain judgement and control.

This is not currently counted as core 20-day programme evidence unless a mission workflow is explicitly run and logged. In the public proof posture it should be described as future or coaching-linked validation, not as an already validated product outcome.

Live coaching can support this layer by helping the user:

- select an appropriate mission;
- define a small success criterion;
- identify the operator or strategy being transferred;
- run the mission soon after training;
- record the outcome without turning it into a high-pressure test.

## Operational evidence levels

### Level 1: In-wrapper improvement

The user improves inside one task wrapper.

Status: useful training signal, but not transfer evidence by itself.

### Level 2: Horizontal portability

The user holds performance after a wrapper, target, speed, interference, or surface-form change.

Status: transfer-relevant in-app evidence.

### Level 3: Vertical support

Capacity Gym improvements support Reasoning Gym performance, especially on rule use, relation binding, constraint solving, and inference under changed forms.

Status: core current programme target.

### Level 4: Delayed and external checks

Improvements persist at later checks or appear on Tracker measures.

Status: stronger programme evidence, still interpreted with caveats.

### Level 5: Mission transfer

The user applies a trained operator or control policy to a real-world mission with a defined outcome check.

Status: future/coaching-linked validation layer.

## Claims boundaries

Allowed:

- designed for far transfer;
- tests whether gains survive changed conditions;
- trains capacity and reasoning together;
- uses wrapper swaps, vertical progression, and delayed checks;
- aims for meaningful gains beyond one-game practice effects.

Not allowed:

- proves far transfer for every user;
- guarantees real-world improvement;
- proves intelligence gain from app scores alone;
- treats self-report or mission completion as diagnostic evidence;
- treats one training wrapper as sufficient proof.

## Source anchors

The protocol is an IQMindware design document, not a direct copy of any one published protocol. It is anchored in several strands of transfer research:

- Barnett and Ceci's far-transfer taxonomy frames transfer as a question of what learned content carries to which new contexts and distances: https://pubmed.ncbi.nlm.nih.gov/12081085/
- Simons et al. review brain-training evidence and caution that trained-task gains are common while distant and everyday transfer is much less established: https://doi.org/10.1177/1529100616661983
- Au et al. report a small positive meta-analytic effect of n-back training on fluid intelligence, while noting that transfer evidence is equivocal and moderated: https://doi.org/10.3758/s13423-014-0699-x
- Melby-Lervag and Hulme provide a more sceptical meta-analytic review of working-memory training and far transfer: https://doi.org/10.1037/a0028228
- Morris, Bransford, and Franks' transfer-appropriate processing account supports matching training processes to later test/application demands: https://doi.org/10.1016/S0022-5371(77)80016-9
- Brunmair and Richter's meta-analysis of interleaved learning supports structured variation as a way to improve discrimination and transfer under some conditions: https://doi.org/10.1037/bul0000209
- Butler's retrieval-practice work supports testing/application as a route to transfer beyond restudy: https://doi.org/10.1037/a0019902