Collaboration is just the first human skill we’re learning to make visible; dispositions, well-being and psychological safety ride the same listening infrastructure. MagpieMatrix is a research prototype that surfaces those skills from a classroom’s team talk — live, during the lesson, not weeks later. We’re building it with the teachers who trial it.
Adelaide University · CSIRO On Prime · Microsoft AFMR · In SA classroom pilots
In an AI-shaped world, the most valuable skills are the most human ones — and the Australian Curriculum now mandates them. Schools must grow and assess collaboration, learning dispositions, well-being and psychological safety. Yet there’s no practical way to see any of it as it happens, so it goes ungraded, unimproved, and unfairly judged.
One teacher, six conversations
Collaboration happens at every table at once. Nobody can hear it all.
Opinions, weeks late
Surveys collect polite recollections long after the moment has passed.
No standard to stand on
No reliable data, no validation, no way to moderate one teacher’s judgement against another’s.
So the most human skills get opinion-based grades: observation stacked onto a 50-hour median week (AITSL 2025), with zero auditable evidence. The signal was always there, in what students say to each other. (And the world is betting big on the gap: education spend heads past US$10T by 2030, in a A$4B+ EdTech sector that still can’t see these skills live.)
We interviewed 42 stakeholders: teachers, parents, regulators, policymakers. Every single one pointed to the same thing: the burden on teachers.
CSIRO On Prime · 2025 · read about MagpieMatrix on CSIRO ↗
No laptops, no screens for students. A passive listener sits with each team. Here’s what that does, in two moves: first what changes in the room, then how the talk becomes evidence a teacher can act on.
A teacher can only be at one table at a time. Watch what stays invisible today, and what changes the moment the magpies wake.
Beat 01 · One group at a time
The table she stands at glows, its skills rising in colour. That one window is all a teacher ever gets.
Beat 02 · Everywhere else
Skills keep rising at every other table: grey, unheard, unrecorded.
Beat 03 · The magpies wake
The devices open their eyes: a quiet, consented witness at every table. No screens, no interruption.
Beat 04 · Visible at last
Every table lights up at once. Evidence replaces opinion — the teacher finally sees the whole room.
No screens, humans in the loop, and it works for any lesson. Four steps from a device on the table to a skill profile a teacher can act on.
Step 01 · Data capture
An IoT device sits with each team, no laptops. Every voice arcs into the magpie eye and streams into secure, school-segmented storage.
Step 02 · Speech to text
The audio becomes a running record of who said what, when: collaboration’s raw material, timestamped and attributable.
Step 03 · The agent reads
The magpie agent runs its lens along the transcript, checking each moment against a collaborative problem-solving framework. Every match drops into the evidence bin.
Step 04 · The skill profile
The skills resolve into a radar profile (collaboration and human skills alike) that the magpie hands to the teacher, a shape they can read at a glance and map to curriculum frameworks.
Data minimisation Australian Privacy Principles
School-segmented cloud In-region storage · APPs & EU GDPR
Human oversight, always AU Framework for Generative AI in Schools
Consent & ethics first University HREC-approved research
The mechanism works anywhere. Here’s why you’d reach for it. Jigsaw is a structure you might already run: students split a problem, become experts, then teach each other. You only ever see the finished product. The magpies show you the exchange in between: who explained, who built on whom, where understanding actually landed.
Phase 01 · Home groups
Four tables, four subsystems, every group hands-on while a magpie eye listens. The ribbons circling each table are turn-taking, live.
Phase 02 · The exchange
One specialist carries their piece to the next group. The first offer gets pushed back (questioned, reshaped), then it docks. Knowledge is negotiated, not handed over.
Phase 03 · The return
Home again, the mission clicks together above every table, and at each one a piece goes in wrong, gets caught, gets fixed. That correction is the learning.
Phase 04 · The magpies
The magpies replay how the design was really built (offered, pushed back, rebuilt in dialogue), and Mars appears. The how can map to frameworks like ACARA’s capabilities.
You’re already expected to grow and assess these skills. MagpieMatrix is an early research prototype working to surface them from real dialogue — and the teachers who trial it help shape what it becomes.
Aligned to your curriculum
Maps to ACARA general capabilities, SACE and learning dispositions: the skills you’re already required to assess, working toward evidence you can stand behind.
A brand-new lens
See what was invisible: well-being, psychological safety, leadership, cohesion and problem-solving, surfaced live from real dialogue.
Shape what gets built
You’re not buying a finished tool, you’re helping build one. Your lessons and feedback steer the research, and your students benefit along the way.
Backed & validated by
Several years of peer-reviewed work on making collaboration observable (as first author and co-author alongside wonderful colleagues), including one of the first LLM-validation studies of collaborative problem-solving assessment from real classroom dialogue.
In preparation · preprint coming
The next step: agents that follow a team across a task and read how their problem-solving unfolds, moving from measuring skills to evaluating the AI agents that assess them.
In preparation · preprint coming
Can how a team talks tell you how its students are doing? Early work on hearing well-being in ordinary classroom dialogue.
03 / 10
Under review: IJAIED (Q1) · preprint on EdArxiv
Puts large language models up against trained human coders on real classroom talk, and asks how far they can be trusted to assess collaboration.
04 / 10
Artificial Intelligence Review · 2026
How an AI teammate is built shapes whether the people working with it end up trusting it. A review of 34 studies connecting the two.
05 / 10
Learning and Instruction · 2026
Reads how tightly a team is pulling together from the traces of an ordinary lesson: no surveys, no interruption.
06 / 10
Int. J. of Computer-Supported Collaborative Learning · 2024
What students say about pulling together on a task, set against what their team actually did.
07 / 10
Technology, Knowledge and Learning · 2024
When the AI stops being a tool and joins the group: what students expect of it, and where their trust runs out.
08 / 10
Journal of Computer Assisted Learning · 2023
Cohesion recovered from the digital footprints teams leave behind them in an online learning platform.
09 / 10
Computers & Education · 2022
What teams actually do once you hand them a dashboard about themselves, and what makes them act on it.
10 / 10
Educational Research Review · 2022
Two decades of research on what holds a small group together when the learning runs through technology.
Where the work reaches
Red marks where the magpies are on the ground: Adelaide, Wollongong, the Netherlands. White traces the research reach: talks and workshops from Kyoto to Dublin, and co-authors from Valladolid to Medellín. Drag the globe; tap a marker.
Devices on the ground: Adelaide: multiple public and private schools across South Australia, STEM-girls research, and a dedicated R&D school · Wollongong: Sarah Digan, University of Wollongong · Heerlen, Netherlands: Marcel Schmitz, Zuyd University of Applied Sciences.
Research reach: talks and workshops at LAK 2024 (Kyoto), EC-TEL 2024 (Krems), the Human-AI Teaming workshop at AIED 2025 (Palermo), CrossMMLA at LAK 2026 (Dublin) and AIED 2026 (Seoul); a systematic literature review with Valladolid; a listening-skills study with TU Munich; and co-authors at the University of Michigan and ITM Medellín.
The road ahead: a validity chain
We don’t assume collaborative-learning evidence naturally becomes useful in practice. The research forges a staged evidence chain: measurement, interpretation, instruction, durable learning. The first link is forged; the magpies make the next three possible.
Assessment-of-learning science
Are the measures valid, generalisable and equitable?
Delivered: LLM-measured CPS from real classroom dialogue, under review (Q1).
Assessment-for-learning science
How do teachers read the evidence, and what do they do with it?
Observable: what teachers read, ask and decide when the evidence arrives.
Assessment-as-learning science
Do teacher-mediated interventions change collaboration?
Teachable: does CPS-informed teaching move groups measurably?
Learning-outcome science: beyond assessment
Does it stick, and transfer?
Durable: what remains when the scaffolds come away.
One link forged, three to go — and none of it possible without the magpies. MagpieMatrix wasn’t born of the AI moment: it has been in the making since the earliest days of Andrew’s PhD.
MagpieMatrix wasn’t born yesterday
Built by one pair of hands, inside the centre that made it possible: C3L, Adelaide University.
Close to home, we’re in the booths at the state’s biggest gathering of public secondary school leaders.
6–7 August 2026 · Adelaide
The South Australian Secondary Principals Association brings the state’s public secondary leaders together at Morphettville Event Centre, under the theme Equitable Excellence: Empowered by Collective Agency. We’ll be there with a booth, showcasing the magpies to the leaders of the schools they’re built for.
saspa.com.au ↗
MagpieMatrix is on the program at The Adelaide School of Learning Sciences, Data & AI, a free, hands-on two-day school hosted by C3L at Adelaide University.
12–13 November 2026 · Adelaide
As part of the school, Andrew Zamecnik is leading a panel on learning dispositions and MagpieMatrix, joined by teacher Jarrod Johnson and other brilliant colleagues. See the device and the evidence up close, and ask the people who run it in a real classroom.
adelaide-school.c3l.ai ↗
Further afield, work from MagpieMatrix will also be presented in North America.
14–18 April 2027 · Toronto
MagpieMatrix research is making its way to the American Educational Research Association (AERA) Annual Meeting, one of the largest gatherings in education research. The 2027 theme: Transcending Borders and Boundaries in Education Research.
aera.net ↗
This is early R&D, not a finished product. We’re inviting SA schools and teachers to trial the prototype in real lessons, with real support, and help shape what it surfaces. Your students benefit now; you help build something made for teachers. Researchers and industry partners: let’s talk too.
