The Distance Between a Good Idea and a Useful Tool

What explicit instruction research means for building reading tools that actually work for older students

I’m very conflicted this week. So I went to the grocery store and bought myself flowers.

Anytime I get stuck and start overthinking, I look at them and it interrupts the spiral for a minute or two. Last week I was confident in the direction I was heading. This week I’m strongly considering a pivot. The flowers are earning their keep.

For those just joining: I’m documenting the process of building reading intervention tools for older students — upper elementary through high school — as a nontechnical educator using AI-assisted development. The short version: I’m building because the gap educators kept naming in classrooms wasn’t getting filled.

Two things happened this week. First, I sent the first version of Juniper Navigate out to the initial pilot cohort — the teacher-facing tool that takes existing phonics assessment data and generates a recommended starting point and instructional sequence per student. Second, I spent the rest of the week researching instructional design for the lesson component that will eventually connect to Navigate. What I found is making me reconsider the entire approach.

The research that moved the ground

Navigate tells a teacher where to start. But it can’t yet connect to the how — the actual phonics lessons that follow the recommendation. That’s always been the next piece.

So this week I started digging into the instructional design research to figure out how to approach those lessons. And the question shifted on me.

The question I thought I was answering: What format should the lessons take?

The question I’m actually answering: Who should be delivering the instruction — the app, or the teacher?

That second question changes everything.

What the research says

The CORE Sourcebook — a foundational reference for reading instruction and assessment, and the book that sits on the shelf in most reading specialists’ offices — identifies four sources of reading failure (p. 3): neurological, familial, socioeconomic, and instructional. It’s the instructional factor I’ve spent the most time thinking about this week.

The research is direct about what happens when instruction falls short. Ineffective teaching methods and strategies without sufficient evidence limit student mastery (Rosenshine, 2012; Moats, 2007). And here’s the finding that stopped me: research clearly shows that students reach higher and faster achievement with systematic, explicit instruction regardless of their learning difficulties. Yet this type of instruction is not always used (Gill & Kozloff, 2004).

That’s not a new finding. But it hit differently this week, because I was sitting in front of a screen trying to decide whether to build a tool that delivers instruction through an app — and realizing I don’t currently have the tech skills, the knowledge, or the resources to make an app capable of delivering the level of explicit, personalized instruction these students need. And building it at a lower level isn’t a neutral choice. If the instruction isn’t good enough, the practice that follows it won’t transfer.

I know from my own experience — as a classroom teacher who taught students phonics and as an instructional coach who trained teachers how to teach it — that the gradual release model is key. You model it, you do it together, you watch them try, and then you release. I’ve seen what happens when that sequence is followed, and I’ve seen what happens when it’s skipped.

The gradual release model (Pearson & Gallagher, 1983; Fisher & Frey) has over 35 years of evidence behind it. The core idea is scaffolding: you build scaffolding to support a structure while it’s going up, and you remove it as the structure becomes secure enough to stand on its own. The teacher provides the scaffold. I can’t build an app yet that can do that part. But I can build an app that can be the space where the student practices standing on their own.

All of these sources converge on the same point: for struggling readers — especially older ones — the quality of the initial instruction is what determines whether practice works. Practice without explicit instruction is just repetition. Practice after explicit instruction is where transfer happens.

The tension

The survey data was clear: teachers want something students can use independently. And that makes complete sense. A 45-minute class period is already full — content instruction, guided practice, assessment, transitions, the student who needs to talk to you about something that has nothing to do with the lesson but everything to do with whether they’ll be present for it. Finding an extra 10 to 15 minutes of explicit phonics instruction for the struggling readers in the room — while also teaching the rest of the class — isn’t a scheduling problem. It’s a physics problem. The minutes don’t exist unless something else gives.

So the original thinking came from a real place: What if the tool could carry the instructional load for the teachers who don’t have the training or the time? That’s a reasonable question. And honestly, part of me still wants that to be the answer.

But then I started asking harder questions.

Can an app deliver the level of explicit phonics instruction that struggling students require? Explicit instruction isn’t presenting information on a screen. It’s modeling. It’s watching a student’s face and listening to word pronunciations, then adjusting in real time. It’s corrective feedback that responds to how a student got something wrong, not just that they got it wrong. I can build a solid practice tool. I can build a scope and sequence engine. But the kind of phonics instruction a struggling reader needs — with the nuance and responsiveness of a trained interventionist — is a different order of complexity. Building it poorly is worse than not building it.

Are classrooms ready for it? Even if I could build it — headphones, quiet spaces, consistent device access, enough time for the app to actually teach and not just quiz? The survey told me most teachers have 10–40 minutes a week. That’s barely enough for practice, let alone instruction and practice through a screen.

And the biggest question: we've been using learning apps in classrooms for years. Why is there still a reading crisis? The National Assessment of Educational Progress showed reading scores declining between 2019 and 2022 at both 4th and 8th grade — during a period when digital tool usage was higher than ever. Current phonics apps aren't failing because they're bad. They're failing because they aren't meeting the specific real-time needs of teachers and students in actual classrooms. Currently a phonics instructional app is a practice environment. It's not a teacher.

None of this makes the original constraints go away. The training gap doesn’t disappear because the research says teachers should lead instruction. The time constraint doesn’t disappear because the evidence says explicit instruction matters. Those problems are still real. The question is whether you solve them by designing around the teacher — or by building something that equips the teacher to do the part only a teacher can do.

Not app-only. Not teacher-only with no digital support. Something that gives the teacher the structure and knowledge to deliver a short, focused lesson — and then gives the student a place to practice what the teacher just taught, with feedback, at their own pace. The app doesn’t replace the teacher. The app makes the teacher’s 10 minutes count.

Where I’m leaning

Strongly toward the gradual release model. Not because it’s easier to implement — it asks more of teachers and it requires a professional development layer I haven’t built yet. But because building the easier version means designing around the instruction, and designing around the instruction is how we got here in the first place.

The whole reason these students are in intervention in middle and high school is that somewhere along the way, the system didn’t provide what the research said they needed. Building a tool that repeats that pattern — even a beautiful, well-designed tool — doesn’t actually solve the problem. It just makes the gap look more polished.

Here’s what the model looks like as it’s forming in my head.

Phase 1 — Explicit instruction. I Do / We Do. The teacher delivers initial instruction with modeling and corrective feedback. Short, structured, ten to fifteen minutes. The app provides the word lists, the sequence, the structure — but the teacher provides the teaching.

Phase 2 — Guided practice. You Do Together. The student practices on the app with built-in feedback — immediate corrective feedback, error tracking, repeated practice. The teacher is nearby but not leading.

Phase 3 — Independent practice. You Do Alone. The student works toward mastery independently. The app tracks performance and reports back to the teacher. This is where the independence teachers asked for actually lives — but it comes after the instruction, not instead of it.

The tool doesn’t try to be all three phases. It knows which phase it’s responsible for.

The reframe for teachers is this: the app isn’t replacing the 15 minutes you don’t have. It’s making the minutes you give worth more — because when students get to the app, they’ve already been taught. They’re not learning cold from a screen. They’re practicing something a human already introduced.

The complication I haven’t solved yet

If the model requires teacher-led explicit instruction in Phase 1, then someone has to equip the teachers to deliver it. And the picture is more complicated than “elementary teachers know phonics, secondary teachers don’t.”

Science of Reading training has become increasingly widespread in K–5 over the past several years — and many of the teachers I surveyed said they’d had it. But training and tools are different things. Teachers in those grades reported that the resources available to them didn’t meet their students’ needs. They had the knowledge. They didn’t have materials that fit their students’ age, their schedule, or the specific decoding patterns their struggling readers needed to practice. And by 3rd through 5th grade, the gap widens — third grade is where the system expects the transition from learning to read to reading to learn. The curriculum pivots to comprehension and text complexity. State testing begins. The message to teachers, implicit and explicit, is that phonics instruction is finished. A 3rd–5th grade teacher may have some Science of Reading training, but the skills students actually need at this level — multisyllabic decoding, syllable types, morphological patterns — are different from what K–2 curriculum covers. So a student who arrives in 4th grade still struggling with decoding often gets comprehension interventions instead, because the system assumes the phonics window has closed. It hasn’t. The skills changed, and the instruction didn’t follow.

By 6th through 12th grade, both problems compound. Teachers are content specialists — English, science, history. Most have had no Science of Reading training at all, and the tools that exist for struggling readers are almost entirely designed for younger students. The survey made this stark: where K–5 teachers had training but lacked appropriate tools, secondary teachers often had neither.

The research confirms the pattern. Sutherland (AERDF, 2024) called it a longstanding neglect of explicit instructional support for decoding in upper grades — the result of a collective assumption that decoding is an early-elementary problem. Shanahan (2025) made the same point: high-quality Tier 2 phonics instruction is often available to elementary students but tends not to exist in middle and high school. Students get assigned to all-purpose interventions that are inadequate for those below the decoding threshold.

So the professional development layer has to serve different needs at different levels. K–5 teachers may need better tools more than more training. Secondary teachers need both. All of them need something that fits the same constraints the student tool has to fit: short, structured, immediately usable.

This connects to the planned Juniper Coach module — teacher routines, mini-lessons, error-pattern guidance. The research didn’t just confirm the instructional model. It confirmed that the PD layer isn’t optional. It’s load-bearing.

Where this leaves me

Navigate is out with the pilot cohort. The lesson design is still open. I’m not going to build something fast that ignores what the research says about how students actually learn to read.

The build is slower than I’d like. But the gap doesn’t close by shipping something that looks good and sits unused. And it doesn’t close by shipping something that gets used but doesn’t work.

In the next two weeks I’ll know more — from the pilot feedback, from the instructional model taking shape, and from whatever the flowers help me think through between now and then.

Juniper Reading CoLab · Built with teachers. Improved with every release.

Responsible AI. Teacher-verified. No student PII.