Loose parts and teacher mediation in early childhood education

1. Figure: affordances, mediation, and artifacts (schema)

Affordances of open material → situated actions a loose piece makes possible: combine, contain, carry, align, hide, undo, leave at rest.

Teacher mediation → read those actions before intervening; provoke without hijacking play; document processes rather than products; enter as co-explorer only when presence helps; secure real material access.

Criticized artifacts → a “STEM loose parts” catalogue; a generative prompt pack of provocations; a dashboard of “creative exploration minutes”; a “creativity score” checklist. They skip the middle term and therefore do not constitute support.

The schema orders a craft, not a model with coefficients. Affordance names here the action a piece makes available in a concrete situation, not a virtue stored in the object or a lesson pasted on beforehand. A tube can contain, roll, join, or remain unused. A cloth can cover, drag, or serve as a boundary. The list illustrates the concept: it is not an inventory or a STEM catalogue (frame). Loose parts, or open materials, are detachable pieces without a single instruction for use, which can be moved, combined, and left. Part of the corpus treats them as a medium toward a goal —number, pattern, interest, a creativity percentage. Another part treats them as a situation: negotiation, sensory attention, immersion, encounter. What counts as mediation depends on that difference (frame).

Observing before intervening means looking at what the material is already doing with that child, on that floor, and with those peers. Provoking without hijacking is a brief invitation that does not replace the direction. Documenting processes means keeping combinations, pauses, undoings, and returns, not a score. Genuine access requires pieces within reach, enough of them to combine, and permission not to produce. Co-exploration is conditional: the teacher enters when her body adds something the child cannot reach —hold, bring closer, fetch— and leaves when play resumes (frame).

The arrow toward the artifacts does not claim a measured reduction. It claims that a catalogue, prompts, a minutes dashboard, or a score list, if they take the place of mediation, are not support for play with open materials. The figure carries no effect sizes, percentages, or samples, and it recommends no optimal time. The middle term fails by excess, when the adult takes over the play, or by default, when the material is locked away or nobody looks. The criticized artifacts tend to automate one of those two failures (frame; inference).

2. State of the art: loose parts and the teaching craft

The literature from 2021 to 2026 does not describe a single craft. It describes two uses of the same name. In one, loose parts are a didactic medium, prepared so that patterns, numbers, interest, early literacy, or a creativity percentage will appear. In the other, they are an open situation: the piece resists, the peer negotiates, the body attends, and the adult sustains without closing the script. They share the vocabulary of exploration and provocation, and they do not authorize the same judgment of a catalogue app (frame).

The open situation appears in four scenes that later sections must not merge. Nome (2024) observes negotiations in two Waldorf kindergartens: resistance, in the sense the study takes from Biesta, comes from non-industrial material, from scripts already begun, and from play partners, and failure of agreement is more present than with industrial toys. Kaneva (2026) reads children aged five and under, in an art museum in England, as rearrangers, makers, and movers, co-producers rather than an audience. Park, Pack, and Chung (2025) follow touch, sight, and hearing in one-year-olds with natural materials. Ko and Cho (2022) distinguish, in a class of two-year-olds, personal immersion and attention to a peer. None of the four delivers a creativity score. Their object is the encounter with the piece (empirical finding, developed below only as far as each design allows).

The teaching craft appears most clearly when a study looks at the adult and not only at the product. Araújo, Esteves, and Marta (2023) find it in preparation for heuristic play with the very young: a non-directive supportive role, the environment, observation, and the difficult balance between not intervening and facilitating. Simoncini and Lasen (2021) name it as adulteration, the adult takeover of play, learned by preservice teachers while holding a pop-up loose-parts playground. Hu (2025) formulates it as an invitation with natural loose material inside a kindergarten program. Zeng and Ng (2025) and Tay and Kaveri (2026) place it in the question asked during play, the latter as the teacher’s own inquiry. Trawick-Smith, DeLapp, Bourdon, Flanagan, and Godina (2023) recall that teachers, peers, and materials influence play together. Hancock (2025) does not study pieces: the paper studies questions and power with families, and it is useful so that the adjective “open” is not romanticized (collective frame; design detail belongs in the mediation section).

The other line uses loose parts as a medium toward a goal: pre-mathematics (Endang et al., 2025), pattern (Dhiaulhaq et al., 2024), number (Utami and Eliza, 2022), literacy with exploration, provocation, and documentation (Arramdhani, 2025), interest (Kartini and Kurniawaty, 2023), STEAM or creativity (Dewi et al., 2024; Nurhayati, 2022; Kaenah et al., 2025; Novitasari, 2025), and free play tied to competencies the abstract does not measure (Tabi’in et al., 2024). If success is the series or the class indicator, an unused piece is read as failure. That documents classroom use, not that a score measures mediation or that a catalogue inherits legitimacy (inference).

The repertoire that comes out of both lines is provisional: prepare the material, invite, ask during play, do not adulterate it, hesitate before intervening, document the process, and withdraw so that immersion can occur. It is not a validated protocol. It is what this literature already names, and what a catalogue cannot do because it has no body in the corner. The figures that sometimes accompany the second line are read in the STEAM and creativity section, each inside its own design, with no average (frame).

3. Method: PRISMA-lite review 2021–2026

This text is a narrative review with an explicit screen, not a registered systematic review and not a meta-analysis. Lite marks a limit of honesty. No flow diagram of identified, duplicate, and excluded records is presented, because those counts were not produced by a reproducible search inside this article. Inventing them in order to look like PRISMA would fabricate a datum (method frame).

The working corpus is a curated list of twenty-eight records with Crossref-verified DOIs, from 2021 to 2026, on loose parts, open materials, kindred heuristic play, or the adjective “open” in the educational relation. Twenty-two are discussed. The other six stay out for economy of argument, not because they are declared invalid. A verified DOI is a condition of citation. Where the curated year and the online date diverge, the citation follows the curated record.

1. Object: loose parts, open materials, heuristic play in professional preparation, or a caution about “open.”
2. Population: early childhood, the first cycle, or the preparation of those who work there. A study of one-year-olds or two-year-olds is marked by age and is not merged with the three-to-five classroom.
3. Window 2021–2026, according to the curated record, and a DOI present on the verified list.
4. Use for the craft: the text allows something to be said about material, question, invitation, documentation, preparation, or a score, even if only to criticize it.

Manuals without a DOI in the corpus, apps not studied by these sources, and any figure the consulted abstract does not report stay out. If the open metadata has no abstract, the sample is not invented. If an abstract claims a difference and publishes no effect size, none is calculated. Each claim is an empirical finding —what the authors report and this review does not reanalyze—, a frame —a distinction of this article—, or an inference —a step beyond the study, said as such. Action-research percentages, means, and developmental categories are not averaged: they come from different instruments and goals. A mean effect of loose parts would be a fabricated datum. The case in the sixth section is a composite vignette, not a corpus source. Treating it as data would contradict this method.

4. Findings: mediation and questions with open materials

Where the abstract is missing, the finding about questioning stops at the published object: a situated practice, not a technique with an increment already quantified.

Zeng and Ng (2025) join, in the title, free play with loose parts, questions, and kindergarten science. It does not follow that every science question preserves openness: it follows that the question is studied during play and not as a prior bank. Tay and Kaveri (2026) make the question a teacher-inquiry project, tied to problem solving, with no increment figure in the open metadata. The shift is from the child’s score to the craft (bounded empirical finding; inference).

Araújo et al. (2023) analyze narratives of twenty-one master’s students on heuristic play, in the Goldschmied tradition, with the very young. They value a non-directive role, the environment, and observation, and they name the difficulty of not intervening and facilitating at once. That device is not the three-to-five corner: it shows mediation as judgment. A checklist or a prompt pack leaves the judgment without language, because they score the product or the sentence before the look (empirical finding; the transfer is an inference). Simoncini and Lasen (2021) see the same judgment in practice: thirty-six Australian preservice teachers organize a pop-up loose-parts playground and learn above all to detect adulteration, the adult takeover of play; they learn less to advocate for it in public. The study names the risk of closing the piece with a prompt and does not classify questions (empirical finding; inference).

Arramdhani (2025), in a case of group A at BAIK kindergarten in Krapyak, describes four times. The teacher sets the topic and the child explores. Then there is invitation and provocation, and the child experiments. Then the child creates and the teacher documents. At the end they review meaning together. The abstract presents the sequence as a stimulus for early literacy. The finding is the choreography, not a reading gain. It resembles the schema —exploration before provocation, and documentation— and at the same time the literacy goal can tilt the ending toward a linguistic product. The case does not decide which occurred. It blocks any celebration of the word documentation that does not ask what is kept (empirical finding; inference).

Nome (2024), with participant observation in two Waldorf kindergartens, shows that open material makes failure of negotiation more present, because the piece does not bring the script. A dashboard of minutes or of product reads that failure as a bad session; a process note keeps it (empirical finding; inference). Trawick-Smith et al. (2023) block a solitary reading of the corner: adult, peers, and materials jointly influence the play of children living in more than one language. Hancock (2025), from decision-making with families, shows that the question called open can exercise power. The adjective makes neither the piece nor the question innocent (bounded empirical finding; frame).

Four review inferences, not a pooled effect, close the section. A question is a candidate for mediation only after the affordance in progress has been seen. Not adulterating was learned by holding a real device, not by receiving a script. Documentation already appears in a literacy case, which is why the document must be watched so that it does not become a rubric. Material and peers resist; a protocol that ignores them remains an individual prompt. None of the four is a validated norm.

5. Findings: STEAM, nature, and creativity with loose parts

Here the dominant risk is not the teacher’s absence but the conversion of open material into the medium of a program. The program can be valuable and still not be the same as sustaining exploration. The figures are those of the abstracts. None is added.

Hu (2025) anchors nature: action research with nature-based loose parts in the Ontario kindergarten, with no sample in the abstract. The invitation is the first step, and the device joins inside and outside. It does not prove that the outdoors outperforms the classroom, and it is not a brief for outdoor play in general (empirical finding). Kaneva (2026) shows creativity without a score in an English museum, with children aged five and under who rearrange and move pieces. The offer does not exhaust the use. The museum is not the kindergarten: it illustrates co-production; it does not guarantee it in every corner (empirical finding; bounded inference).

Park et al. (2025) and Ko and Cho (2022) show what a score of plastic originality loses. Five one-year-olds, over about nine months, with soil, wood, fruit, and water, compose visual relations, explore sound, and cross with touch the boundary between body and matter. The abstract reports 6,665 photographs and 240 videos, plus notebooks of cases and stories: process documentation, not a score. Eleven two-year-olds, with about twenty kinds of loose parts and thirty sessions of about fifty minutes, two or three times a week, show personal play —connecting experience, pleasure, immersion— and peer play —looking, feeling presence. Neither study offers a score: each describes a sensory or relational encounter at its own age (empirical finding; inference).

Dewi et al. (2024) conduct action research with twenty-one children aged five to six at Mutiara Kasih kindergarten in Padang, three cycles, loose parts loaded with STEAM. They report creativity of 6.9% before the cycles, 15.4% in cycle I, 47.1% in cycle II, and 80.42% in cycle III. The finding is that report, inside its definition and design, not a general effect. The series is the number a score would want to automate, with the limits of the device: one center and improvement cycles, without the controlled trial the abstract does not announce (empirical finding; inference). Nurhayati (2022) describes, without figures, STEAM learning in which arranging loose parts is lived as more meaningful and pushes toward making products. There is a craft of arrangement, and documenting only the finished work loses the discard (empirical finding; inference).

Kaenah et al. (2025) describe more enthusiasm, curiosity, and imagination, and they limit the study to observation, with no experimental measure of long-term creativity. That abstract does not feed a stable score (empirical finding). Novitasari (2025), in a design the abstract calls qualitative, with group B at RA Mambaul Ulum Jatirogo and materials from the surroundings, claims a 70% increase. The figure brings no visible instrument in the abstract: it is what a dashboard would display, and it is not a measure of the craft.

Dhiaulhaq et al. (2024) bead patterns with children aged four to five, in two action-research cycles. In the categories the abstract names BSH and BSB, recognition —especially the ABC-ABC pattern— moves from 14.28% in the pre-cycle to 50% in cycle I and 78.57% in cycle II. The finding lives inside that task. A compulsory pattern closes openness: combining otherwise, or leaving the piece, is no longer equivalent. Calling the beads loose parts does not preserve openness if success is the series (empirical finding; inference). Endang et al. (2025) observe ten children aged four to five and describe patterns, classification, number, shape, exploration, and collaboration, with no percentage. They recommend daily pre-mathematical use and do not show how they keep each combination from becoming a scored item (empirical finding; inference).

Sumarseh and Yaswinda (2023) compare science plus loose parts with a nonequivalent control, in twenty children aged four to five, and report more critical thinking in the experimental group, without means or an effect size. None is calculated here: the science in the study is an activity, not a property of the piece (empirical finding; inference of abstention). Kartini and Kurniawaty (2023), with ten children aged five to six at Kuntum Mekar and two cycles, report 90% on the class interest indicator. That is interest under that definition, not mediation: the abstract does not separate novelty of the material from ease of the rubric (empirical finding; inference).

Utami and Eliza (2022) require a pause. Over four weeks, at TK Mutiara Ceria in West Pasaman, with children aged five to six, the experimental mean moves from 17.7 to 23.1 and the control mean from 18.6 to 22.1. The declared differences are 5.4 and 3.5. The abstract says that the two-tailed significance of 0.06 is less than 0.05 and speaks of an effect of 1.4. The sentence does not square with itself: 0.06 is not less than 0.05. The contrast is not redone. What is recorded is that “very high” efficacy is not supported by the inequality the same abstract states. Even if the contrast had been clear, recognizing numbers is a closed goal, not evidence of open exploration (empirical finding of the abstract; methodological inference).

Tabi’in et al. (2024) describe, at KB Labschool UIN K.H. Abdurrahman Wahid in Pekalongan, a sequence without those figures: the teacher prepares the loose parts and the children explore. The link to four competencies is not measured. Preparing and leaving room to explore fits the schema; calling it an already measured acquisition does not (empirical finding; inference). Invitation, arrangement, and attention inform the craft; percentage series stay inside their device. Exporting 80.42%, 70%, 90%, or 78.57% into a universal score is rejected: they do not measure mediation and, in several designs, they measure the closing of openness.

6. Anonymized classroom case (ages 3–5, materials corner)

The vignette is composite and anonymized. It does not come from an identified center, it is not a sample, and it is not an empirical finding. It serves to see the criteria in a classroom of three- to five-year-olds and, in the same movement, to say what is not claimed. Citing it as efficacy would misuse the section.

A mixed group of three- to five-year-olds, in a center this text does not name, keeps for several weeks a permanent corner. It is not a one-morning workshop or a module opened from an app. The boxes are low and they stay: spools, cardboard tubes, lids, small cloths, pinecones, smooth stones, ribbons, and rings. There is no finished model and no card assigning each piece to a letter of STEAM. The quantity allows combining and also allows a box to remain full. Material brought out only for the photograph is not this corner (frame of the vignette).

At the start of the week the teacher brings a tray level with the floor and withdraws to observe. She announces no product. In the first minutes she does not ask: she watches what moves, what stays still, who carries, and who watches without touching. Only if a child holds a tube and looks for more material does she offer a continuity question —whether another piece fits— and then she is quiet. She does not correct the form or name a machine. When a cloth slips and the child holds it out, she holds it and lets go as soon as he takes it back. That is the conditional co-explorer. There is no log of minutes. There are sequence notes —combination, undoing, return the next day, a piece that migrated— and a few photographs of the process, not of a centered finished work. There is no originality rubric (frame of the vignette).

How many children were present is not stated: a number would turn the vignette into a false sample. Duration in counted weeks or mean minutes is not stated, nor a rise in creativity, nor a difference in critical thinking, nor that the question caused the next combination. There is no interobserver agreement, no baseline, no app switched on as a comparison, and no generalization. Pinecones and tubes are an example of pieces without a single instruction, not a shopping list. The vignette tries the criteria and does not confirm them. If a minutes dashboard entered that classroom, the morning when a child only looks —an echo of what Ko and Cho (2022) describe at age two, not a datum of this center— could be counted as low exploration. If a score entered, the cloth released without becoming a product would fall outside. The loss is shown. It is not measured.

7. Discussion: criteria for genuine mediation

The four criteria are design hypotheses. They are not standards, not a supervision rubric, and not items in an app. They ask whether a corner has a craft, and they judge an artifact by what it omits.

1. Observation of affordances before intervening. The teacher looks at what actions the material is already offering —roll, contain, join, stay still, migrate— before speaking. It is not the formality that legitimates a prompt written in advance.
2. Provocation without hijacking play. The invitation continues a visible direction or opens a compatible possibility. It does not replace that direction with the program goal. The teacher enters as co-explorer only when she helps, and leaves when play resumes.
3. Documentation of processes, not a product score. Combinations, pauses, undoings, returns, resistances, and negotiations are kept. The morning is not translated into a score of creativity, interest, or correct pattern.
4. Technology in the service of material access. A digital resource fits if it helps to obtain, arrange, lend, or remember real materials. It ceases to fit when the catalogue, the prompt, the minute, or the score takes the place of the corner and of mediation.

The first rests on Araújo et al. (2023): observing and organizing the environment are the role, and the balance between not intervening and facilitating is its difficulty. Arramdhani (2025) states exploration before expansion. Park et al. (2025) show, at another age, that attention to the material is already an event. Intervening earlier imports an objective the situation has not revealed. Zeng and Ng (2025) place the question during play. A prompt pack reverses the order: the sentence exists before the situation (inference).

The second takes adulteration from Simoncini and Lasen (2021) and the invitation as a first step, not a full script, from Hu (2025). Kaneva (2026) shows an offer that does not exhaust use. Always entering is a polite hijacking. Never entering can abandon access, if nobody holds the cloth when the child asks. The criterion does not fix a number of seconds: the judgment is the craft, and the generative text does not see the cloth fall. Tay and Kaveri (2026) recall that the question can be studied inside play and not downloaded as a correct list (criterion inference).

The third does not deny the percentages: it leaves them inside their design. Dewi et al. (2024), Novitasari (2025), Dhiaulhaq et al. (2024), and Kartini and Kurniawaty (2023) offer series a score would want to inherit. Utami and Eliza (2022) state efficacy against the significance figure in their own abstract. Park et al. (2025) stand at the other pole, with a large corpus for five infants. Nome (2024) asks that the failure of negotiation also be kept. Documentation is faithful if someone who was not there can reconstruct what was done, what was undone, and what was not touched. It is unfaithful if it can reconstruct only a percentage (inference).

The fourth separates access from substitution. Tabi’in et al. (2024) begin by preparing the material. Ko and Cho (2022) leave it within reach across thirty sessions: availability over time is part of the play observed. Access, as an inference, includes the child’s height, quantity, permission not to produce, and continuity across days. An app fits if it remembers the boxes, lends pieces between classrooms, or keeps the process photograph. It does not fit if it labels pieces as STEM areas before use, writes the day’s provocation without a look, sums creative minutes, or demands a score to count the morning as done. None of the twenty-two studies trialed a commercial app. The incompatibility is one of craft. To claim a technological effect would fabricate the experiment that is not there.

Without the four together the protocol comes apart: observation without pieces, provocation without a look, documentation of a pattern already closed, or a score that justifies the purchase return the device to the product. Hancock (2025) blocks using the criteria as a supervision sentence uttered without having been in the corner. Trawick-Smith et al. (2023) recall that the peer is already part of what must be observed (frame).

8. Contrast: catalogue apps versus co-exploration

On one side is conditional co-exploration: a body in the corner, real material, judgment about when to enter. On the other are four genres that represent the corner without inhabiting it. They may serve a meeting or a materials order. They are not support for play while they occupy the place of the craft. No brands are named, and no effects unmeasured in this corpus are attributed to them.

The “STEM loose parts” catalogue classifies in advance: the tube is engineering, the stone is science, the lid is mathematics, the cloth is art. Utami and Eliza (2022), Dhiaulhaq et al. (2024), Sumarseh and Yaswinda (2023), Dewi et al. (2024), and Nurhayati (2022) show that closure by number, pattern, science, or STEAM already happens in classrooms, with teachers present. The catalogue industrializes that closure and moves it ahead of use. Co-exploration can discover that the tube was, today, a container for pinecones. If the app cannot admit the discovery, it serves the label and not access (inference).

The prompt pack speaks the language of invitation and therefore seems close to the second criterion. Hu (2025) and Arramdhani (2025) place the invitation in an environment already under way, not on the previous Sunday. Simoncini and Lasen (2021) teach that takeover of play is seen in practice. A chance hit by the prompt is not mediation, because mediation includes staying silent. The generator produces. The co-explorer may not speak. What is claimed is that difference of structure, not a harm in points (inference).

The minutes dashboard says how much time counted as exploration, not what happened. Immersion, looking without entering, and failed negotiation —Ko and Cho (2022), Nome (2024)— are not homogeneous minutes, and Park et al. (2025) needed months for five infants. A clock that only says whether the corner was open fits the fourth criterion; one that declares creativity by a threshold does not. The score looks like evidence because it inherits percentages from small devices, sometimes with the goal that closes the material and, in Utami and Eliza (2022), with an internal inconsistency. Discard and an unnamed piece lower that score and can be the play itself (inference).

The practical test, a hypothesis and not a result, is one of substitution. If, when the app is switched off, the corner is still supplied and the teacher can still observe, provoke, be silent, and document, the app was logistical. If the provocation, the success criterion, or the very idea of what to do with the boxes disappears, it had occupied the craft. None of the twenty-two studies ran that test. Anticipating its figure would fabricate the future study.

9. Limits

Twenty-two texts from a list of twenty-eight are not the universe of the topic, and there is no flow diagram to reconstruct. The set is enough to see craft and score. It is not enough for a mean effect. Several readings depend on the abstract, not on microdata. Without an abstract —Zeng and Ng (2025), Tay and Kaveri (2026), Trawick-Smith et al. (2023), Hancock (2025)— use stayed at the object of the title. The full article may qualify what was not seen here. The percentages cited are in their abstracts; their reach is one center, one cycle, and a local definition.

Ages are not added. Park et al. (2025) study one-year-olds; Ko and Cho (2022), two-year-olds; Araújo et al. (2023), preparation for heuristic play with the very young; the vignette thinks of ages three to five. Kaneva (2026) takes place in a museum, Simoncini and Lasen (2021) in a pop-up playground for teacher preparation, Hu (2025) in an Ontario program that joins inside and outside. The inferences cross gestures, not populations. Some pieces circulated online the year before the curated record: the citation follows the record —2025 for Zeng and Ng, Hu, and Hancock; 2026 for Tay and Kaveri— without changing the DOIs.

The vignette is a dramatized hypothesis. It does not enter a quantitative synthesis and should not be taught as a real morning; if it is too tidy, the tidiness is a defect of the illustration. No commercial app was evaluated. A future tool that only lent boxes might fit the fourth criterion: this text has not seen it and neither absolves nor condemns it. Turning the four criteria into a verbal supervision list, a risk already marked with Hancock (2025), destroys them as design hypotheses.

10. Conclusions

A catalogue of “STEM loose parts,” generative prompts for provocations, a dashboard of creative-exploration minutes, and a creativity-score checklist do not constitute support for play with open materials when they do without teacher mediation. The craft is that mediation in preschool: observe the affordance before intervening, provoke without hijacking play, document processes instead of scoring products, and use technology only if it leaves the pieces truly within reach, with an adult who enters as co-explorer when she helps and withdraws when play goes on. That sentence is the thesis, not an estimated effect.

Verified literature from 2021 to 2026 sustains it as a hypothesis, not as a norm. It sustains it by showing question, invitation, non-adulteration, observation, and documentation learned in situation —Tay and Kaveri (2026), Hu (2025), Simoncini and Lasen (2021), Araújo et al. (2023), Arramdhani (2025)— and by showing material that resists, is attended to with the body, and is played among peers —Nome (2024), Park et al. (2025), Ko and Cho (2022), Kaneva (2026)—. It sustains it by contrast when percentages of creativity, pattern, interest, or number remain inside small devices —Dewi et al. (2024), Novitasari (2025), Dhiaulhaq et al. (2024), Kartini and Kurniawaty (2023), Utami and Eliza (2022)—. It does not sustain it as a trial against digital products: that trial is not in the corpus. Until descriptions of the teacher’s gesture are as careful as this corpus’s categories of child development, the responsible conduct is not to wait for the algorithm of the perfect provocation. It is to keep the corner supplied and to exercise the judgment that none of the four criticized pieces can exercise in the teacher’s place.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre