An essay on the architectural decisions that shape what a classroom observation platform can and cannot do in Scottish secondaries, and why what we've left out matters as much as what we've built.
Every classroom observation platform is defined as much by what it leaves out as by what it includes. The things software doesn't do are rarely visible from its marketing pages — those pages naturally emphasise what the product can. But for any school considering an evidence instrument that touches teachers' professional records, the absences matter. They tell you what the builder has decided the product is for, and what it isn't. They tell you what has been considered and rejected, what has been considered and deferred, and what has never been considered at all.
This essay is about the absences in Learning Lens: what the product doesn't do, why it doesn't do those things, and why that shape is the right one for a product operating in the Scottish professional-trust tradition. The psychology of observation is a separate conversation; I have written about what goes wrong when observation systems measure the wrong things — here the focus is product shape.
The things we don't do because the profession has been clear
Some absences in the product are the direct consequence of decades of Scottish professional argument. A product of this kind that ignored them would be unusable in a Scottish school, and it should be.
The product doesn't grade lessons. It doesn't produce a score at the end of a Learning Visit, doesn't rate an observation from one to five, doesn't output an "outstanding" or "requires improvement" label. The Scottish objection to graded observation is longstanding and well-documented — it appears in SSTA policy, in EIS guidance, in GTC Scotland's Trusted Teaching frame, in the Donaldson review, in the current PRD/PU framework that explicitly separates learning-and-teaching judgement from professional review. Our reading of that argument is that it's correct. A single lesson, observed by a single observer, cannot be reduced to a grade without misrepresenting the thing being observed. Quality in teaching is a pattern that emerges across many observations. The product is built for that pattern; grades would break it.
The product doesn't rank teachers. There is no league table, no performance heatmap of named staff, no sorted list of teachers by any metric the product produces. Teacher-level views are individual and teacher-held. Aggregation happens at department, stage, and school level — never as a comparison of named teachers against each other. The same scholarly and union tradition that resists grading resists ranking, for the same reasons, and we have followed that reading.
The product doesn't integrate with performance-management systems or competence procedures. There is no export format that could be consumed by a disciplinary documentation workflow. The Terms of Use contractually prohibit use of Learning Lens records in competence procedures. This is architectural as well as contractual. We have made it difficult on purpose.
These are the absences that define Learning Lens as a Scottish product. They are not gaps; they are the shape of the product.
The things we don't do because the mechanism is the problem
A second set of absences follows from a design standard rather than from professional consensus: the structural pattern of a feature governs what it is, regardless of the copy around it.
The standard is easiest to state through the case it is usually applied to. Consider an engagement view that aggregates pupil-response data across visits, and consider extending it to alert leadership when an individual teacher's visits cross a quantified threshold of concerning pupil-response sentiment. Framed as "starting points for supportive professional conversations," it reads as developmental. Structurally it is something else: an automated trigger, a named individual as its target, and that individual's line managers as its audience. Those three properties define the feature. Warm copy does not modify them.
Learning Lens applies the standard rather than the framing. The engagement view flags concerning patterns at department level only. Individual-teacher patterns are visible to teachers on their own dashboards, as data for their own reflection. Leadership does not receive automated alerts about named teachers.
The same standard governs two adjacent decisions.
Observer calibration data is bounded the same way teacher data is. Coverage-gap prompts — observers who have not yet recorded practices in certain categories — are retained as developmental professional learning data. What the view does not surface is a measure of how far a named observer's capture habits diverge from the team mean: that is evaluative measurement of an individual, rendered visible to leadership without the observer's opt-in, and the three structural properties are the same ones.
Departmental comparison is bounded by audience. A teacher seeing their own practice profile against their department average is engaged in self-reflection. The identical chart on leadership's view of a named teacher is individual comparative judgement rendered visible to a line manager. The comparison is therefore restricted to the teacher's own dashboard.
Applied consistently, the standard rules out a recognisable class of feature: anything that flags, ranks, or compares named individuals to their line managers. That is an English-accountability-software pattern, no matter how Scottish the words around it.
The things we don't do because AI doesn't belong there
A narrower absence concerns the use of AI.
Learning Lens uses AI in one place: to generate aggregate-pattern prose using the evaluative-quantifier register of HGIOS4 or Ofsted EIF. "The majority of learning visits included evidence of retrieval practice" — that kind of sentence, generated from the data, is useful because the evaluative quantifier is the framework's own language and producing it manually across multiple reports is a grind. That sits alongside the wider cross-framework observation approach that maps evidence to inspection language without forcing a single national rubric.
What we don't do: AI-generated feedback to individual teachers. The product does not produce computer-generated advice, development suggestions, or judgements aimed at a named teacher. When a teacher sees "areas to explore" on their own dashboard, the suggestions come from the research metadata baked into the taxonomy — not from a language model's reading of their observation record.
This is also a mechanism question, not a framing question. Experience in Scottish classrooms makes it clear that teachers reject, on principle, a computer's view of their practice. They are right to reject it. The observer's professional judgement, and the subsequent professional conversation between observer and teacher, is the instrument. The product's job is to resource that conversation with evidence, not to pre-empt it with algorithmic commentary that neither party has consented to.
The distinction matters because the ed-tech industry is increasingly pitching AI-assisted teacher feedback as a feature, from Scottish vendors and from English vendors hoping for Scottish adoption. The research points the other way: work like Bastani and colleagues' 2025 PNAS paper on AI in educational contexts finds that AI without pedagogical guardrails undermines the thing it is meant to support. The design choice matters more than the presence or absence of the technology. National AI guidance for schools reaches the same position on human judgement; this boundary is that position implemented in a product.
Learning Lens uses AI where it is useful and safe (pattern-summary prose at the aggregate level). It doesn't use AI where it would be unsafe (individual teacher feedback). That boundary is deliberate and will not shift in response to market pressure.
The things we don't do yet
A third category is deferred rather than rejected: built later, and stated plainly now so that a school evaluating the product is working from its current shape rather than its intended one.
Teacher-first sequencing of records. Observation records are currently visible simultaneously to the teacher, the observer, and senior leadership. A teacher-first sequencing model — where records become visible to leadership only after the teacher has acknowledged and annotated them — is a stronger implementation of the teacher-held-records principle. It is a substantial change to the data architecture and is scheduled for 2027. Where a local agreement treats teacher-first sequencing as a precondition for observation infrastructure, that is a material difference and is better established before a pilot than during one.
Full export audit visibility to teachers. Every PDF export, bulk report, and staff-meeting text generated from Learning Lens is logged server-side with its generator, timestamp, and scope. What that log does not yet have is a teacher-facing view: a record on the teacher's own dashboard of which exports included their data. It is on the near-term roadmap, and it moves the log from something held by the vendor to something visible to the person the data describes.
Anonymised follow-up tracker in bulk PDF exports. The Term Summary PDF currently names teachers in the Follow-up Tracker table. That is leadership planning data and in context it is generally appropriate, but named individuals in a bulk leadership export is precisely the pattern the previous section describes. The default is under review and may move to anonymised (Teacher A, Teacher B) with names held separately.
Stating these publicly makes them checkable against a date, which is the point of stating them.
The things we don't do because they would make the product worse
Finally, there are things the product doesn't do because doing them would make it worse rather than better. Completeness is not always a virtue in software.
No real-time observation dashboard. There is no view showing which classrooms are currently being observed, which teachers are live on the system, or any other moment-to-moment feed. This is a decision rather than a backlog item. The real-time-dashboard pattern is the shape of surveillance software, not of professional-learning software: it signals to teachers that their line managers are watching, and it makes the product's use feel like being watched rather than like professional dialogue being recorded. It is not on the roadmap and will not be added on request.
No pupil-level identification. Pupil response data is at classroom-aggregate level. No pupil is named, tracked, or identified. There is no pupil-facing component to the product. We considered, at one point, a pupil-voice module that would let pupils respond to lightweight descriptive statements about their classroom experience. We may still build that as a separate module, in a form that is anonymous, aggregate, and teacher-initiated. We will not build pupil-level identification. The product has no business identifying pupils; that is properly held by the school.
No "coaching methodology" prescription. The product does not require schools to adopt a particular developmental model — no instructional coaching, no specific peer-observation framework, no proprietary methodology. Learning Lens is an evidence instrument. Schools using instructional coaching, practitioner enquiry, professional learning communities, or any other developmental approach can use the product within their existing practice. The evidence layer is neutral about the pedagogy layer. This is deliberate: imposing a methodology would make us a competitor to specific coaching platforms and commit schools to a particular approach they may have good reasons not to adopt. Staying neutral lets us serve the full range of Scottish school practice.
No teacher progress-against-expectation tracking. The product does not place teachers on developmental trajectories and does not measure their movement along those trajectories. There is no concept of "developmental stage" for individual teachers in the product. Progression-against-expectation models are, in our reading, a particular way of operationalising performance management that the Scottish system has not adopted and does not want. We have not built the infrastructure for it.
Why the absences are the shape
When someone asks what Learning Lens is, the easiest answer is to describe what it does — observation capture, framework-mapping, aggregated analytics, PDF reports. That answer is accurate but incomplete. What Learning Lens is, as much as anything, is a particular set of absences. It doesn't grade. It doesn't rank. It doesn't flag. It doesn't generate AI feedback. It doesn't surveil. It doesn't prescribe methodology. It doesn't track teachers against a developmental ladder.
Each of those absences is an active decision, not a missing feature. Each has been reasoned from the Scottish scholarly and procedural tradition the product is built to serve. Each is maintained by design constraints that make re-introducing the behaviour harder rather than easier — architectural decisions, contractual commitments, data-structure choices.
For a school considering Learning Lens, the absences are what make the product fit the Scottish system rather than work against it. They are what a professional association representative reads the product against, and they map onto the procedural floor that tradition has defended for decades. They are what allows the product to sit with teachers as a resource for their own professional learning rather than as an instrument held by a line manager. They are what makes inspection-readiness work possible without distorting a professional-trust culture.
A product that did all the things Learning Lens deliberately doesn't do could not work in Scotland. The shape is the argument.
Jamie Scobie writes from extensive experience in Scottish secondary education, including pastoral care, data for improvement, and school self-evaluation. This blog is an independent publication: he writes in a personal capacity as the creator of Learning Lens, writing about classroom observation, teaching evidence, and education policy. He speaks here only for himself and for Learning Lens, not for any employer or other organisation. He holds an MSt from Cambridge (Distinction) and a Masters from Stirling.