An ABA matrix training worksheet should show more than a neat grid. It should make clear which components were selected, which combinations were taught, which cells stayed genuinely untrained, what happened on independent probes, and whether the arrangement remained accessible and acceptable to the client. Without that history, a filled cell can mean direct teaching, incidental exposure, a contaminated probe, or a novel response. Those are not interchangeable findings.

This copyable tool is a planning and review aid for a qualified clinical team. It does not select a target, prove that matrix training is appropriate, or supply a universal diagonal. The fictional example demonstrates the record structure and arithmetic, not a recommended dose or mastery rule.

Clinicians & ABA Professionals / Assessment and Treatment Planning.

Clinical boundary: Use a current, authorized plan and the client's preferred communication and access supports. Stop or pause when the client withdraws assent, materials become inaccessible, the response definition changes, or a probe is exposed. The same applies when distress or another unwanted effect appears, or when the responsible clinician cannot interpret the record safely. A worksheet cannot establish consent, medical safety, treatment necessity, payer coverage, generalization, or benefit. The BACB ethics resources are a useful starting point for the qualified review that remains necessary.

What the grid can and cannot show

Matrix training arranges two or more sets of components along axes, directly teaches a selected subset of their combinations, and probes combinations that were not taught. Published matrix-training research describes recombinative generalization as correct responding to novel combinations of components encountered in other trained combinations. Results vary by learner, target, component repertoire, and training arrangement, including across comparisons of matrix-training procedures. A matrix is therefore an instructional design, not an outcome guarantee.

The practical distinction is between the cell map and the evidence map. The cell map lists possible combinations. The evidence map says which cells were assessed, taught, merely encountered, independently probed, invalid, or still unknown. A blank cell is not a failure. A checked box is not proof that a response emerged.

The worksheet supports four connected tasks. A completed ABA matrix training worksheet should let another qualified reviewer reconstruct each one:

  1. Define meaningful component axes and confirm that their combinations make sense.
  2. Freeze the status of taught and withheld cells before instruction begins.
  3. Retain trial-level access, prompt, consequence, exposure, validity, and initial-response data.
  4. Review trained performance, untrained-cell probes, component checks, fidelity, burden, and client response without blending their denominators.

Purpose, access, and component readiness

Begin with the client's reason for working on the skill. A grid built around staff convenience can be technically orderly and still have poor social validity. Record whether each component already occurs in the relevant modality, whether the client can perceive and select it, and whether the planned combinations belong in the client's ordinary life.

Do not remove an AAC device, visual support, mobility aid, sensory support, familiar vocabulary, language access, or processing time to manufacture “independence.” ASHA describes AAC as a multimodal system that should remain available and individualized. If a response is accepted through speech, sign, selection, typing, gaze, or another mode, write that rule before scoring begins.

Blank purpose and readiness card

FieldCase-specific entryClient-selected purpose and meaningful settingObservable response and accepted formsOrdinary AAC, language, motor, sensory, visual, and processing supportsAxis A name and component definitionsAxis B name and component definitionsEvidence that individual components are available to the clientCombinations that are meaningful, unsafe, impossible, or culturally inappropriateNatural or programmed cue and response windowTeaching procedure, prompt/correction rule, and consequence ruleProbe procedure and what must remain absentAssent, dissent, pause, stop, and re-entry signalsMatrix version, procedure version, owner, effective date, and next review

A component check is not the same as a combination probe. Someone may respond accurately to each component alone yet not recombine them under compound stimulus control. The reverse record can also be misleading if staff infer component control from a memorized compound. Preserve both levels.

Build a matrix that survives review

Give every row, column, and cell a durable identifier. The content of each intersection should be plausible, observable, and distinguishable from the others. Record excluded combinations rather than silently deleting them. If a three-dimensional matrix is clinically justified, use a separate versioned table or layer; do not squeeze the third dimension into notes that later reviewers cannot reconstruct.

Blank component map

Component IDAxisPlain-language labelExact stimulus or featureAvailable response formReadiness evidenceAccess checkNotesA1AA2AA3AB1BB2BB3BB4B

Blank cell grid

Axis A \ Axis BB1B2B3B4A1A1B1A1B2A1B3A1B4A2A2B1A2B2A2B3A2B4A3A3B1A3B2A3B3A3B4

For each cell, describe the combined stimulus, accepted response, and any meaningful variation. “Same as above” is risky when the cells differ by modality, position, wording, or required action.

Freeze taught and withheld cells before exposure

Research with young autistic learners and practice tutorials describe diagonal/nonoverlap, overlap/stepwise, sequential, simultaneous, and combination-only arrangements. Their relative usefulness depends on the component repertoires and the clinical question. This worksheet does not choose among them. It does require the team to name its selection before outcome data are visible.

Cell assignment and exposure ledger

CellPlanned statusReason for statusDirect teaching allowed?Prompt/consequence rulePrior or incidental exposureFirst exposure dateStatus change and approverteach / withhold / baseline / excludeteach / withhold / baseline / excludeteach / withhold / baseline / exclude

A withheld cell ceases to be a clean untrained probe if the exact combination is modeled, rehearsed, corrected, or directly taught. Mark it “exposed” and decide prospectively whether later observations answer a different question. The matrix-training procedure comparison and practice tutorial illustrate why the arrangement and its exposure history matter. Reinforcement following a correct probe response can be clinically appropriate. The record should still distinguish what occurred before that consequence from later exposure.

Trial record for teaching and probes

Freeze the initial response before a prompt, correction, replay, or discussion. Retain invalidity reasons. An inaccessible array, unavailable communication device, staff cue, wrong cell, position error, interrupted opportunity, or prior exposure can make a trial uninterpretable. It should not be forced into an incorrect-response denominator.

Blank trial ledger

Date/timeCell and versiontrial typePresented?Access/supports available?Initial responseIndependent correct?Prompt/correctionConsequenceValid? reasonExposure changed?Assent/dissent or unwanted effectObserverteach / withheld probe / component checkyes / no / not scorableteach / withheld probe / component checkyes / no / not scorableteach / withheld probe / component checkyes / no / not scorable

An invalid setup is evidence about implementation, not evidence that the learner responded incorrectly. A valid incorrect or no response remains useful clinical information. Keep the distinction visible.

Cell-level review and denominator rules

Summarize by cell before combining anything. Repeated trials in one cell do not increase coverage of other cells. A high teaching-trial percentage does not answer whether untrained combinations emerged. Component checks, untrained probes, and fidelity observations describe separate questions.

Blank cell summary

CellOriginal statusExposure status at probePlanned trialsPresentedValidIndependent correctPrompted/correctedIncorrect/no responseResult boundaryclean / exposed / uncertainclean / exposed / uncertainclean / exposed / uncertain

Useful calculations include:

  • valid probe coverage = valid withheld-cell probes ÷ planned withheld-cell probes;
  • correct responding among valid probes = independently correct valid probes ÷ valid probes;
  • observed untrained-cell yield = withheld cells with a qualifying independent response ÷ all planned withheld cells;
  • teaching performance = independent correct teaching trials ÷ valid teaching trials; and
  • component-check performance = correct valid component checks ÷ valid component checks.

Always print the numerator and denominator beside a percentage. If the plan requires repeated probes per cell, define what constitutes a cell-level result before looking at the data.

Fictional community-garden label example

The fictional client chose a goal involving labels for locations at a community garden. Axis A contains three familiar action words; Axis B contains four familiar location words. All combinations were checked for meaning and access. Four cells were assigned to teaching and eight were withheld. The example does not imply that this matrix, language form, or allocation fits another person.

Frozen matrix status

Axis A \ Axis BB1 seed tableB2 tool rackB3 water stationB4 compost binA1 visitteachwithholdwithholdwithholdA2 checkwithholdteachwithholdwithholdA3 cleanwithholdwithholdteachteach

Across 12 valid teaching trials, 10 initial responses were independently correct:

10 / 12 × 100 = 83.3% teaching performance.

For the eight withheld cells, one scheduled probe was invalid because the wrong label array appeared. Seven valid probes remained, and five contained an independently correct response:

  • 7 / 8 × 100 = 87.5% valid probe coverage;
  • 5 / 7 × 100 = 71.4% correct responding among valid probes; and
  • 5 / 8 × 100 = 62.5% observed untrained-cell yield against all planned withheld cells.

The 71.4% and 62.5% values are not competing versions of the same statistic. One describes performance within interpretable probes; the other keeps the unobserved planned cell visible. Neither proves broad generalization beyond this matrix.

Eight separate component checks were scheduled. Seven were valid, and six were correct:

6 / 7 × 100 = 85.7% component-check performance.

That result does not establish that the compound responses came under appropriate component control. The team would review the cell pattern, error forms, prior exposures, and client experience before deciding whether any additional teaching or testing question is justified.

Implementation-fidelity panel

Choose observable components tied to the written procedure. Do not score staff style preferences as fidelity. A six-component example might cover correct cell, correct array, ordinary access present, programmed cue/window, prompt/consequence rule, and trial status recorded accurately.

EventCorrect cellCorrect arrayAccess presentCue/windowPrompt/consequenceAccurate recordMismatches and context

In the fictional review, 18 observable events produced 108 component opportunities. Ninety-nine components matched the procedure and nine did not:

99 / 108 × 100 = 91.7% component implementation.

That percentage is not learner accuracy. Review the nine mismatches by type and timing. A lower probe result collected during access or array errors may not answer the intended recombinative-generalization question.

Decision record, pause points, and version changes

Record what the data can support, what remains unknown, and who owns the next clinical decision. A reasonable review can conclude “insufficient valid coverage” or “some untrained combinations occurred under these conditions” without declaring success or failure for the person.

Review itemEntryTaught-cell finding with denominatorWithheld-cell finding by cell and relationComponent-check findingFidelity mismatches that limit interpretationClient assent, dissent, burden, and unwanted effectsExposure or probe-contamination changesAlternative explanations and unknownsHold, revise, continue, directly teach, or discontinue questionQualified decision owner and review date

Create a new version when axes, components, stimuli, accepted response forms, access supports, or cell assignments change. Do the same when the teaching procedure, probe rule, or scoring logic changes. Keep the old version linked to its data. Combining incompatible versions can create an apparently complete matrix that never existed as one clinical arrangement.

Related resources

Sources

The research describes varied participants, skills, matrices, and procedures. It supports careful design and direct testing of untrained combinations; it does not establish that the fictional example, a particular diagonal, or any outcome threshold is appropriate for an individual client.