The all-in-one education system that turns learner data into insights educators can act on.
This system is designed to be a complete platform for learning and business administration, scaling to meet the demands of a large organisation or the solo teacher.
It makes the running of the day-to-day administration seamless while allowing educational professionals to give learners the best support possible.
It has four aims:
Better Teaching and Learning
The system builds the bigger picture of each learner. That helps the educator make better decisions and spend less time working out the next move, and it shows learners what to work on next.
One Record for Everything
Everything about a learner lands in one place and is entered once, from the first enquiry to the latest piece of feedback. The record builds up as a side effect of the daily work, and it is only ever used to help the learner.
AI With the Educator in Charge
The AI does the marking, the drafting and the routine work. The educator approves what learners see, and learners do their own thinking before the system explains anything.
Any Subject, Any Size
Built to scale from a solo teacher to a large organisation, and from school classrooms to corporate training. English writing is the first example; the system underneath does not depend on the subject.
The Build
Five parts make up the platform: one live, one in testing and three planned. They meet in the hub: the educator and learner views.
Live
Marking and Feedback
What It Does
Before any marking appears, learners highlight their strongest and weakest passages, say why, and find each point the task asked for in their own answer. The AI then marks against the Cambridge B2 First writing rubric, a second blind pass checks every grammar error, and the educator approves the feedback before learners see it. Learners correct each mistake in their own writing before the correction is shown, and short exercises target the exact error they made. The educator sees how closely each learner's own judgement matched the marking.
Why It Exists
The best educators know where their learners will go wrong, and feedback about the task helps where feedback about the person can do harm. So the system asks learners to judge their work first and explains second. Every error is tagged to a point on a grammar map, which starts a record of how each learner goes wrong.
Built So That
the educator reviews every AI comment before a learner sees it
A CRM built around the learner. Registers, homework, family contacts and every conversation, fees and receipts, enquiries from new families and timetable changes all land on one record, typed once. Families report absences through their own link, so the person who knows does the recording. A register nobody has taken raises a reminder for the educator, then for the office. Each member of staff sees only what their permissions allow, and the owner can change any permission person by person.
Why It Exists
Running a small education business, you end up being three things: an educator, front office and back office. It's an all-encompassing role. I found no tool that was able to keep track of everything, and the various tools I did use didn't link up.
Things would come at you all the time. New enquiries. Learner or stakeholder questions. Payments. Accounting. Then you have to mark work and plan classes. Give reports. Check the general health of your business and keep track of expenses. It's a lot. And when it's not in the same place, it gets confusing.
So I automated as much as possible and built a tool that works.
It also puts the series' tracking argument into practice. The record of each learner builds up as a side effect of running the academy, so nobody has to stop and fill it in.
Built So That
nothing is typed twice
a receipted payment is corrected or cancelled, never deleted
private notes never reach the learner they are about
the classroom screen shows names only, with no leaderboards
Educators build assignments from six question types: multiple choice with an optional "justify your answer" step, open writing, reordering, spot the flaw, error correction and gap fill. Each assignment starts as a draft from the curriculum, and the educator edits it before it is set. Every question is tagged with the skill it tests, the misconception a wrong answer points to and the depth of thinking it asks for.
Why It Exists
A good learning tool is built around how learners go wrong in the subject, and that knowledge sits with educators. Time saved on preparation is only worth having if the educator still decides what is set.
Self-paced practice that picks each learner's next question from the misconception behind their last mistake. Learners try before they can ask for a hint, and say how sure they are before the answer is shown. Questions come back at spaced intervals. It starts from the existing bank of 2,243 grammar exercises.
Why It Exists
Learning styles have no evidence behind them. A tool should adapt to what the learner already knows about the topic. In maths, tools built on catalogued mistakes have trial evidence behind them, while writing has no such map yet, so the grammar map built for marking is a start.
Help for the moment a learner is stuck and the educator is busy. The agent sets a similar question at the learner's level and coaches them through it, answering with questions. Once they have cracked it, they go back and do the real task themselves. On writing, it asks guiding questions and never writes ideas or plans for them. The educator sees a summary of every conversation.
Why It Exists
An educator can help one learner at a time, so the rest wait, or reach for a chatbot that does the work for them. That shortcut costs them. In a trial with nearly 1,000 high school students in Turkey, learners given an AI that handed over answers did better in practice and worse in the exam once it was taken away. A tutor that only gave hints avoided the harm. This agent gives every learner help on demand and leaves the thinking with them.
What the Demo Shows
Help level: how much help a learner needed, compared only with their own past.
The educator view sorts learners into states such as declining, inconsistent or not submitting, each with a suggested next step. Scores are read against each learner's own average, so a 13 from a learner who usually gets 9 reads as their best work. A flag never stops a learner doing anything. The learner view shows their own work, their feedback and the exercises aimed at their mistakes, and never a running profile of them.
Why It Exists
Habits like effort and self-control shape a learner's life as much as grades do, and the fairest comparison is with their own past. A record like this gives the educator an outside check on their own judgement. Whether that check reduces bias is still to be tested.
Every part of the system is pedagogy-driven: it is built on what research says about how people learn, and the same principles hold for a 13-year-old learning English and an adult learning to sell or to work safely.
K12A learner recalls the third conditional in a gap fill a week after the lesson, instead of rereading their notes.
CorporateA sales rep answers two objection-handling questions on their phone each week, instead of rewatching the module.
The same practice spread over days and weeks is remembered far longer than practice crammed into one session.
Lives in: Assignment Builder, Adaptive Practice, Marking and Feedback (planned)
K12An error from September's essay comes back as a short exercise in October and again before the exam.
CorporateHealth and safety refreshers arrive as short spaced checks across the year, replacing the single annual session.
Mixing things that are easy to confuse teaches learners to tell them apart. Brand-new items are still best learned one block at a time.
Lives in: Assignment Builder, Adaptive Practice
K12Past simple and present perfect questions appear mixed in one set, so the learner has to decide which form fits.
CorporateA set of near-miss scenarios mixes a slip hazard, a trip hazard and a manual-handling risk, so the worker learns to tell them apart on site.
Novices need structure and worked examples. As expertise grows, the same support gets in the way and should be withdrawn.
Lives in: Assignment Builder, Adaptive Practice, Learner Help Agent
K12A weaker writer gets a paragraph frame for an FCE essay; a stronger one gets only the task.
CorporateA new hire sees a fully worked discovery call; an experienced rep goes straight to a live-style scenario with no model answer.
Mistakes in any subject follow patterns. Feedback that names the pattern, and sets the right idea against the wrong one, works better than a right-or-wrong mark.
Lives in: Quiz Blocks, Adaptive Practice, Marking and Feedback
K12The system's 2,243-item grammar bank is the error map for English; "the people is" is tagged as a plural-noun agreement error, not just "grammar".
CorporateA sales error map lists the ways reps go wrong (pitching features before finding the need, discounting too early, skipping the objection). Health and safety has its own (assuming someone else has isolated the power).
Learners who attempt, judge or explain their own work before they are told the answer learn more from the answer when it comes.
Lives in: Adaptive Practice, Marking and Feedback, Learner Help Agent
K12Before submitting an essay, the learner finds each content point in their own text and flags the sentence they are least sure of.
CorporateA rep scores their own call recording against the sales rubric before the AI and the manager give theirs. The gap between the two scores is itself an actionable insight.
Feedback works when it is about the work and what to do next. Feedback about the person (praise, blame, ranking) often does nothing, or harm.
Lives in: Marking and Feedback, Learner Help Agent
K12"Your second paragraph doesn't answer the question about cost", never "You're a weak writer".
Corporate"You offered the discount before the client raised price", never a league table of the team's close rates.
How much help a learner needs to get something right says as much about their progress as whether they get it right.
Lives in: Learner Help Agent, Educator and Learner Views
K12In October a learner needs four guiding questions to fix a tense error; by December they need one.
CorporateA new manager needs three prompts to run a return-to-work conversation in a scenario; a month later they need none.
The most useful data is direct behaviour (attempts, corrections, revisions, time on task), read against each learner's own baseline rather than the class or team.
Lives in: Educator and Learner Views
K12The dashboard flags a learner whose revisions have dropped to half their usual rate, without comparing them to the class.
CorporateA learning and development manager sees which competencies a team keeps needing help with, as an actionable insight for the next programme. Individual records are never fed into performance reviews.
People keep learning when they have real choices, see themselves getting better and can see why it matters. At work, learning only counts if it shows up on the job.
Lives in: Adaptive Practice, Educator and Learner Views
K12A learner chooses which of two essay prompts to write on, and sees their own error rate fall over the term.
CorporateAfter a module on discovery questions, the rep's next three real calls are reviewed for those questions, so the measure is behaviour on the job rather than a smile sheet.
What the system leaves out
Learning stylesNo onboarding quiz, no "visual learner" labels.
Growth-mindset scales and interventionsThe average effect on achievement is very small.
Inferred emotionThe system never guesses how a learner feels from their face, voice or typing.
Corporate folkloreThe 70-20-10 split has never been tested, and smile-sheet scores are not evidence of learning.
Gamification in feedbackLeaderboards are never used, and nothing from gamification feeds the data an educator sees.
The Decisions Behind the Build
The system is built iteratively towards the product goals:
Everything that happens with a learner ends up on one record, and the feedback makes learners reflect on their own work.
The backlog is ordered by four questions
WorthHow much is it worth to the people using it?
EvidenceHow strong is the evidence?
DependenciesWhat does it depend on?
RiskHow risky is it?
Done and LiveMarking and FeedbackLearners reflect on and fix their own writing, marked by AI the educator approvesWhy it comes here. The strongest evidence and the most value, it needs nothing else built, and it creates the first learner data
Done, in TestingAcademy OfficeEverything about a learner lands on one record, typed onceWhy it comes here. The core every other part depends on, and what solo teachers and small academies need first
BacklogAssignment BuilderTagged assignments drafted from the curriculumWhy it comes here. Adaptive practice needs its question tags
BacklogAdaptive PracticeEach next question aims at the learner's last mistakeWhy it comes here. Needs tagged questions and error data from earlier parts
Backlog, Research DoneLearner Help AgentStuck learners work through a similar question, never the set taskWhy it comes here. Builds on the data and templates before it
Why It Comes Here
The strongest evidence and the most value, it needs nothing else built, and it creates the first learner data
A clear definition of done
Every access rule passes its checks
Every feature has been used in a real browser
Nothing changes for people already using the system
Each one is then reviewed
What the review finds goes back into the backlog. Marking and Feedback went through a six-step review after its first version, and that is where learners fixing mistakes in their own writing came from.
Risk is dealt with early
The help agent is the one place AI would talk to a learner without an educator reviewing it first, so guard rails and a lot of testing will be necessary.
Demo · Part 1 of FeedForward
Marking and Feedback
In Short
The marking system is built on one idea: the student thinks first, the AI second, and the teacher has the last word. Every decision below follows from that order.
Most AI marking tools hand back a corrected text and a score. Research on feedback says that is the weakest way to use it: a student who reads a correction has not practised finding or fixing the error. This system uses the AI to locate problems, then makes the student do the fixing.
The Loop
1. The student plans the piece.
Ideas Behind the Demo
Students do the thinking before the AI does.
They plan, write and judge their own work first, then fix their own mistakes once the system shows them where to look. Students correct their own work instead of reading corrections made for them.
The teacher always has the final say.
Feedback waits for the teacher, who can change anything before releasing it. If the system is unsure about a correction, it asks the teacher instead of guessing in front of the student.
It is built to be fair and accurate.
A second, independent check reviews every grammar correction, to stop correct English being marked wrong. Alerts about a struggling student go to the teacher as a prompt to help, never as a penalty, and they are based on what the student has done, not on impressions.
It pinpoints grammar mistakes.
Each grammar error is matched to one of 113 specific points, not a vague label like "tenses", so the feedback and the practice target the same thing.
Each student is measured against their own progress.
A teacher sees every new score next to that student's own average, so a 10 out of 20 reads as a step forward for one student and a slip for another. Students are only given work on what they have been taught.
Every feature in this system is there for a reason. Each card below takes one stage of the loop and sets out the decisions behind it: what was built, how it works for the teacher and the student, and the research that backs it, listed in full under References. Where the evidence is mixed or thin, it says so.
Decision 1 of 8
Teacher and student side by side
What It Does
The demo is a thin wrapper around the same screens real teachers and students use.
Why It Is a Good Decision
The loop crosses two people. Seeing both at once shows the handover, and what a visitor sees is what a school would get.
Decision 2 of 8
A guided path of six steps
What It Does
Assign a task, write, mark your own work, AI marks, teacher approves, feedback then practice.
Why It Is a Good Decision
The unusual part is the loop, and none of it is visible on arrival. Without a path, visitors browse dashboards and leave.
Decision 3 of 8
Progress is read from the database
What It Does
A step is ticked only when it has really happened. It survives a reload and fills in even if the path is ignored.
Why It Is a Good Decision
Nothing can be ticked that did not happen.
Decision 4 of 8
The marking is real
What It Does
A visitor's piece goes through the live marker. The demo student's earlier units were marked by the same AI, not written by hand.
Why It Is a Good Decision
A demo of AI marking with scripted results proves nothing.
Decision 5 of 8
A class with real spread
What It Does
Eight students and 30 marked pieces over four units, built so each "struggling" state appears.
Why It Is a Good Decision
Reports only make sense with contrast: the same 10 out of 20 means opposite things for two students in this class.
Decision 6 of 8
No login, with firm limits
What It Does
Per demo: one marking, one Quick Mark, two photo reads. Two live demos per address per day and a $2 daily ceiling. Later visits can still look around.
Why It Is a Good Decision
A login loses visitors. Limits on the server keep an open demo from being drained.
Decision 7 of 8
Each visitor gets their own copy
What It Does
Demo sessions are private and cleaned up on a schedule.
Why It Is a Good Decision
One visitor cannot see or spoil another's work.
Decision 8 of 8
One view on a phone, both on a computer
What It Does
A Both / Teacher / Student switch; phones open on one side. The tour says which side to look at.
Why It Is a Good Decision
Two apps do not fit on a phone screen. The screens adapt to the width they are given.
Decision 1 of 7
A writing planner with no AI in it
What It Does
Three steps: find what the task asks, plan each paragraph, choose the language to use. The shape for each task type is stored as data, so a teacher's own method can be loaded.
Why It Works
Planning before writing is one of the best-supported writing interventions. Kellogg (1988) found outlining raised essay quality; Graham and Perin (2007) rank planning strategies among the strongest effects for adolescent writers.
Decision 2 of 7
The plan stays beside the writing box
What It Does
Task, plan and tips remain visible while the student writes.
Why It Works
Holding the task in mind while composing competes for working memory (Sweller, 1988). Keeping it on screen frees that capacity for the language.
Decision 3 of 7
Highlight strong and weak parts, with a reason
What It Does
Before marking, the student marks what they think is good or weak, by category, and must say why. The reason can be in their first language.
Why It Works
Self-regulated learners monitor and judge their own work (Zimmerman, 2002). Requiring a reason turns a click into a judgement. Allowing the first language keeps the task about thinking, not about English for describing English.
Decision 4 of 7
"Where is ___ in your answer?"
What It Does
One question at a time, for each point the task requires. The student answers by highlighting the place in their own text.
Why It Works
Content is one of the four Cambridge marking scales, and missed points are the cheapest marks to lose. An answer you point to can be checked; a free-text "yes, I covered it" cannot.
Decision 5 of 7
A prediction is captured before the result
What It Does
The student's judgements are saved before marking, then compared with what the marker found.
Why It Works
Weaker performers tend to overestimate their work (Kruger and Dunning, 1999). Calibration only improves when a prediction meets a real result (Andrade and Valtcheva, 2009).
Decision 6 of 7
Exam mode forces a last look
What It Does
In a timed exam, the student must re-read the piece before the final send. No prompts, and no change is required.
Why It Works
Mirrors real exam technique. The step guarantees the look without telling the student what to find.
Decision 7 of 7
The teacher decides if a plan is required
What It Does
Set per assignment. Required by default for homework and classwork, optional in an exam, where planning is the candidate's own choice.
Why It Works
Teachers keep control within a tested structure, in line with the autonomy findings of self-determination theory (Ryan and Deci, 2000).
Decision 1 of 9
Marked on the real exam scales
What It Does
Four Cambridge B2 First scales, 0 to 5 each: Content, Communicative Achievement, Organisation, Language.
Why It Works
Feedback helps most when it shows where the learner is against a clear goal (Hattie and Timperley, 2007). Students see the same criteria the examiner will use.
Decision 2 of 9
A blind second opinion on every grammar error
What It Does
A second check sees only the sentence and the words, never the first marker's comment. What it does not agree with goes to the teacher, not the student.
Why It Works
Independent double marking is how exam boards control marker error. A false correction teaches a false rule, and a trusted source makes it stick. In a test on 80 saved markings it cleared all 16 cases a teacher had ruled acceptable and removed no clear error.
Decision 3 of 9
Code catches the marker contradicting itself
What It Does
Rules drop an "error" whose own comment says "both are fine" or "if you mean...".
Why It Works
The AI decides once; code applies the same rule every time. On 130 saved markings this removed 29 false items with scores unchanged.
Decision 4 of 9
Code does what code does better
What It Does
Spelling, word-level checks and nonsense detection run in code, not in the model. Random typing is caught before any AI call.
Why It Works
Deterministic checks give the same answer twice, cost nothing, and leave the model for judgement.
Decision 5 of 9
Feedback is about the task, never the person
What It Does
A banned-phrase list and plain-language rules sit in the marker's instructions. No comment on ability or character.
Why It Works
Kluger and DeNisi (1996) found over a third of feedback interventions lowered performance, mostly when feedback drew attention to the self rather than the task.
Decision 6 of 9
Short, itemised comments
What It Does
Each point is one plain sentence on what happened, plus one action.
Why It Works
Shute (2008) recommends elaborated feedback in manageable units, on cognitive load grounds. No study has compared bullets with prose for language learners, so the claim is about load, not format.
Decision 7 of 9
Only what matters is shown
What It Does
A one-off slip is not listed as an area to work on; repeated patterns are.
Why It Works
Focused feedback on fewer error types did better than broad, shallow feedback in Lim and Renandya's (2020) meta-analysis. See also Ellis (2009) and Bitchener and Knoch (2010).
Decision 8 of 9
Explanations in the student's language, English kept as English
What It Does
Translation is on request. Quoted text, example words and grammar names stay in English; the explanation around them is translated.
Why It Works
Aldosari (2026) found corrections in English with the rule explained in the first language led to more uptake and lasting gains. One exploratory study at B1, so a direction, not proof.
Decision 9 of 9
Tested like software
What It Does
Prompt changes go through a saved set of test pieces. Every changed item is read, not just pass rates.
Why It Works
Pass rates hid real losses more than once. The agreed bar: main errors caught, score about right, a few things to work on.
Decision 1 of 8
A review queue before release
What It Does
Every marking waits for the teacher, who can change scores, edit wording, remove a comment or put it back.
Why It Works
The Eedi and Google DeepMind LearnLM tutoring study used the same design: a human reviewed each AI-drafted message before the student saw it. Human approval is the evidenced way to use AI with learners.
Decision 2 of 8
Doubtful calls go to the teacher
What It Does
Items the second opinion disagreed with appear as "Marker unsure - not shown to the student". One click shows them.
Why It Works
Doubt is routed to the person qualified to settle it. The student never has to work out whether the machine is right.
Decision 3 of 8
Removing an error removes its exercise
What It Does
Practice is built from the released feedback, so a teacher's edit carries through.
Why It Works
The teacher's judgement governs the whole loop, not just the comment text.
Decision 4 of 8
Every score read against the student's own record
What It Does
The teacher sees a mark beside that student's average, such as "6 below their average". Needs three earlier pieces, and the new score is left out of the average.
Why It Works
The same 10 out of 20 was one student's best work and another's worst. Progress against your own record is ipsative assessment (Hughes, 2011).
Decision 5 of 8
That comparison is teacher-only
What It Does
Students see their score and feedback, never a running profile of themselves.
Why It Works
Feedback aimed at the person rather than the work is the kind most likely to harm (Kluger and DeNisi, 1996).
Decision 6 of 8
The teacher can require a full rewrite
What It Does
One switch, available before release or after the student finishes the loop.
Why It Works
Ferris (2004): different errors and learners need different treatment, and weaker students may not be able to self-correct. The teacher judges when the standard loop is not enough.
Decision 7 of 8
Fixed guardrails, not a blank canvas
What It Does
Teachers adjust settings inside a tested English-writing template. They do not write the AI's rules.
Why It Works
Guardrails built by expert teams still fail differently by subject. Letting each teacher write their own multiplies that risk.
Decision 8 of 8
Quick Mark
What It Does
A teacher pastes or photographs one piece and gets a grade from the same marking engine, unsaved.
Why It Works
One marker, used everywhere, so a quick grade and a full marking cannot disagree.
Decision 1 of 11
Self-assessment comes first
What It Does
Feedback opens with "How well did you judge your own writing?", comparing the student's highlights with what the marker found.
Why It Works
The student reasons before the system explains. Learning to judge your own work is the skill that lasts after the course ends (Nicol and Macfarlane-Dick, 2006).
Decision 2 of 11
"Fix it in your own writing"
What It Does
For a taught, rule-based error the student sees the place and types the correction. The answer is not shown.
Why It Works
Ferris (2004): for rule-governed errors, indirect feedback that makes the student solve the problem is usually better. Producing the form is itself learning (Swain, 1995), and noticing the gap comes first (Schmidt, 1990).
Decision 3 of 11
Attempt before hint
What It Does
The hint and "show me the rule" appear only after a first miss. The second try is multiple choice.
Why It Works
Feedback stops helping when the answer is available before the learner has looked for it (Bangert-Drowns et al., 1991). Even a failed attempt improves later learning (Kornell, Hays and Bjork, 2009).
Decision 4 of 11
No rule, no forced correction
What It Does
Word choice, idiom and anything not yet taught are simply corrected, with advice.
Why It Works
Ferris calls these "untreatable": there is no rule to apply, so the fix is given. Extending this to "not yet taught" is this product's own step, not a published finding.
Decision 5 of 11
The place is marked, not a code
What It Does
Students see the highlighted words, not an error-type label.
Why It Works
Ferris and Roberts (2001) found no difference in self-editing between coded errors and plain underlining. The error taxonomy is used behind the scenes to pick exercises.
Decision 6 of 11
Practice follows the result
What It Does
Could not fix it: 3 exercises. Fixed on the second try: 2. Fixed first time: none, and told "you fixed this yourself".
Why It Works
Practice goes where the evidence of need is. Retrieval practice strengthens memory more than re-reading (Roediger and Karpicke, 2006).
Decision 7 of 11
Repeated failure changes the method, not the difficulty
What It Does
Two or more wrong on a point last time: the student's own correction and the rule are shown directly.
Why It Works
A ladder that gets easier until the answer is given teaches "fail and you need not think". Switching from indirect to direct follows Ferris's own exception for learners who cannot self-correct.
Decision 8 of 11
Exercises are not repeated for 90 days
What It Does
Grammar items a student has seen are held back. Vocabulary may come round again.
Why It Works
A repeated item tests memory of the item, not the rule. Vocabulary benefits from spaced return (Cepeda et al., 2006).
Decision 9 of 11
A bank first, AI second
What It Does
Exercises come from 2,243 reviewed items graded by depth of knowledge. AI writes one only when the bank has none.
Why It Works
A Washington state study found AI-led practice missed its intended thinking level about 40% of the time. Reviewed items are the safer default.
Decision 10 of 11
"Level it up"
What It Does
Shows how a correct sentence of the student's could be made stronger.
Why It Works
Feedback should also answer "where next?" (Hattie and Timperley, 2007). Strong writers get something to do.
Decision 11 of 11
Carried into the next piece
What It Does
"Last time you worked on..." sits beside the next writing box. The next feedback says whether the point came up again.
Why It Works
Feedback only works if it is used in later work (Black and Wiliam, 1998). This closes the loop across pieces.
Decision 1 of 9
A grammar profile per student
What It Does
Each taught grammar point is tracked: attempts, errors and trend across all their writing.
Why It Works
Teaching is most effective when it responds to evidence of what the learner can already do (Black and Wiliam, 1998).
Decision 2 of 9
"Struggling" is six different things
What It Does
Not submitting, declining, inconsistent, working below level, not progressing, stuck on a point. Each calls for a different response.
Why It Works
Two students with the same low average can need opposite help. A single "at risk" flag hides that.
Decision 3 of 9
No claim without enough evidence
What It Does
Each state needs a minimum number of marked pieces, from 0 to 4. Below it, the student shows as "Not enough work yet". Every flag states its evidence count.
Why It Works
Small samples mislead. Proving that nothing has changed is the hardest claim, so it needs the most evidence.
Decision 4 of 9
Rules tested on real spreads of scores
What It Does
Volatility counts changes of direction, not spread. A steadily improving student had a higher spread than an erratic one.
Why It Works
A flag that fires for half a class is noise. Each rule was checked against a real class before it shipped.
Decision 5 of 9
Flag, never gate
What It Does
A flag sends a signal to the teacher. It never locks content or holds a student back.
Why It Works
Follows the Response to Intervention model: identification should trigger support, not a penalty.
Decision 6 of 9
Objective signals shown to the teacher
What It Does
Signals appear on the teacher's screens, not only inside the software.
Why It Works
Where data is missing, impressions fill the gap, and that falls hardest on already disadvantaged students.
Decision 7 of 9
Only behaviour that can be checked
What It Does
Recorded: self-assessment accuracy, whether a hint was asked for, how fixing went. Not recorded: mindset surveys or guessed emotions.
Why It Works
Growth-mindset effects on achievement are close to zero once publication bias is corrected (Macnamara and Burgoyne, 2022). An inferred mood is not a fact about a student.
Decision 8 of 9
Class health on request
What It Does
The class report is generated when the teacher asks, showing each scale's movement since the last unit.
Why It Works
A report a teacher asks for is one they are ready to act on. An always-on panel becomes wallpaper.
Decision 9 of 9
The dashboard routes, it does not analyse
What It Does
The home screen sums up or points to a class. Detail lives in the class.
Why It Works
One job per screen keeps the first view readable (Sweller, 1988).
Decision 1 of 6
Taught points become work to do
What It Does
An error on a taught point leads to self-correction and exercises.
Why It Works
Ferris's criteria for choosing what to correct include whether the structure has been taught in class.
Decision 2 of 6
Untaught errors are noted quietly
What It Does
The student is told, with no exercise. The error is logged for the teacher.
Why It Works
Asking a student to self-correct a rule they have never met is guessing, not retrieval. The log still tells the teacher what is coming.
Decision 3 of 6
Vocabulary is never penalised for ambition
What It Does
Words outside the taught list do not count against the student. Missed chances to use taught words go to the teacher only.
Why It Works
Penalising ambitious language teaches students to write safely. Range is part of the Cambridge Language scale.
Decision 4 of 6
Each error is tied to a specific grammar point
What It Does
The marker names one of 113 points on a grammar map, not a broad label like "tenses".
Why It Works
Specific labels make the hint, the exercise and the profile line up on the same point.
Decision 5 of 6
Curriculum is data, not code
What It Does
Subject, course and unit, with grammar, vocabulary, checklists and plan shapes, are all loaded as content.
Why It Works
A school can bring its own syllabus without the software being rewritten. The marking logic stays the same.
Decision 6 of 6
Tasks follow the real exam formats
What It Does
Essays carry two fixed notes plus the student's own idea; emails carry points to answer.
Why It Works
The self-check and the Content score both read from the same required points.
Honest Limits
The design is answerable to research; the product has not yet been tested on learning outcomes. Worth saying on the page, because it is the question an informed reader will ask.
No outcome data yet. No class has used the system long enough to show that students improve. The claims here are that each decision follows the evidence, not that the product is proven.
Written corrective feedback is a debated field. Truscott (1996) argued grammar correction does not help at all. Later studies answer him, but recent meta-analyses find direct, indirect and metalinguistic feedback give similar effects (Brown, Liu and Norouzian, 2023). The fair summary: feedback type matters less than focus, fit to level, and whether the student engages with it. The system is built around engagement for that reason.
Some decisions are product choices, not findings. How many errors to show, the 3 / 2 / none exercise rule, the 90-day window and the taught / untaught split are reasoned from the research, not taken from it.
The marker is not perfect. About 1 marking in 40 offered a wrong choice that was correct English. The teacher's review exists to catch this.
Not built yet: bringing the plan back into the feedback, a whole-class report per assignment, and the wider practice engine.
References (30)
Feedback in General
Bangert-Drowns, Kulik, Kulik and Morgan (1991). The instructional effect of feedback in test-like events. Review of Educational Research, 61(2).
Black and Wiliam (1998). Assessment and classroom learning. Assessment in Education, 5(1).
Hattie and Timperley (2007). The power of feedback. Review of Educational Research, 77(1).
Hughes (2011). Towards a personal best: a case for introducing ipsative assessment in higher education. Studies in Higher Education, 36(3).
Kluger and DeNisi (1996). The effects of feedback interventions on performance. Psychological Bulletin, 119(2).
Nicol and Macfarlane-Dick (2006). Formative assessment and self-regulated learning. Studies in Higher Education, 31(2).
Shute (2008). Focus on formative feedback. Review of Educational Research, 78(1).
Correcting Second-Language Writing
Aldosari (2026). Frontiers in Education. First-language explanations with English corrections, B1 learners.
Bitchener and Knoch (2010). The contribution of written corrective feedback to language development: a ten month investigation. Applied Linguistics, 31(2).
Brown, Liu and Norouzian (2023). Meta-analysis of written corrective feedback, 52 studies. Language Teaching Research.
Ellis (2009). A typology of written corrective feedback types. ELT Journal, 63(2).
Ferris (2004). The "grammar correction" debate in L2 writing: where are we, and where do we go from here? Journal of Second Language Writing, 13.
Ferris and Roberts (2001). Error feedback in L2 writing classes: how explicit does it need to be? Journal of Second Language Writing, 10.
Lim and Renandya (2020). Efficacy of written corrective feedback in writing instruction: a meta-analysis. TESL-EJ, 24(3).
Schmidt (1990). The role of consciousness in second language learning. Applied Linguistics, 11(2).
Swain (1995). Three functions of output in second language learning. In Cook and Seidlhofer (eds.), Principle and Practice in Applied Linguistics.
Truscott (1996). The case against grammar correction in L2 writing classes. Language Learning, 46(2).
Planning, Self-Assessment and Memory
Andrade and Valtcheva (2009). Promoting learning and achievement through self-assessment. Theory Into Practice, 48(1).
Cepeda, Pashler, Vul, Wixted and Rohrer (2006). Distributed practice in verbal recall tasks. Psychological Bulletin, 132(3).
Graham and Perin (2007). Writing Next. Carnegie Corporation of New York.
Kellogg (1988). Attentional overload and writing performance: effects of rough draft and outline strategies. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(2).
Kornell, Hays and Bjork (2009). Unsuccessful retrieval attempts enhance subsequent learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35(4).
Kruger and Dunning (1999). Unskilled and unaware of it. Journal of Personality and Social Psychology, 77(6).
Roediger and Karpicke (2006). Test-enhanced learning. Psychological Science, 17(3).
Ryan and Deci (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1).
Sweller (1988). Cognitive load during problem solving. Cognitive Science, 12(2).
Zimmerman (2002). Becoming a self-regulated learner: an overview. Theory Into Practice, 41(2).
AI in Education and Measurement
LearnLM Team, Google DeepMind, and Eedi; Brazão, McKee and colleagues (2025). AI tutoring can safely and effectively support students: an exploratory RCT in UK classrooms. Report of 11 November 2025 and arXiv:2512.23633.
Liu, Sun, Esbenshade, Tian, Zhang and He (2025). Teacher-authored prompts for configuring student–AI dialogue: K–12 classroom implementation. arXiv:2604.16738.
Macnamara and Burgoyne (2022). Do growth mindset interventions impact students' academic achievement? Psychological Bulletin.
Joe Allport