The Build

Lessons, Checks & Milestones

The interactive system as it is built: the activities inside a Lead Designer lesson, the lesson's check that writes evidence, and the milestone assessment experience that produces the grade. The whole adaptive engine runs on one fuel — trustworthy evidence of what each learner actually knows — and everything here exists to keep that fuel clean.

Three kinds of interaction

It is critical not to confuse these — they have different purposes, different rules and different security needs.

Kind	Where	What it is & its rules
Try it	Inside each part of a lesson	Short practice with instant feedback. It helps the learner learn; it is retryable, asks for no typing, and is not how the system decides mastery.
The lesson's check	At the end of every lesson	Response, reasoning in the learner's own words, and confidence, with "I'm not sure" always available. It produces one of five results — Solid · Nearly there · Not yet · Not sure · A mix-up — and is never graded.
Milestone assessment	At each milestone on the course road	Substantial work in a chosen context, then a spoken defense, marked against the course's weighted rubric. The design law: lesson results shape the journey; only milestones produce the grade.

Try it activities

Every lesson follows one Lesson Page Standard: an opening case with one choice, then parts — each with Watch (narrated video) or Read instead, and a Try it — with Pause moments, the check, a close and a carry-forward. A Storyline ties the lesson together, and the page is designed phone-first. The Lead Designer chooses each Try it from six mechanics:

Try it	What the learner does
Sequence and order	Puts steps in order for a fresh situation that shares no wording with the lesson's case.
Classify and categorise	Sorts items into the lesson's categories.
Match pairs	Matches each item to its partner.
Quick pick	Picks the best option from a short set.
Find the error	Spots what is wrong in a short piece of work.
Compare and decide (practice)	Judges two attempts side by side — a warm-up for the check's own format.

Each reply's feedback names the card line it comes from, and each tempting wrong move carries a "why not there" written by the designer. A Try it never asks the learner to type and never asks for confidence: that belongs to the check.

The check types

The lesson's check comes in three shapes, each a typed schema:

Check	Shape
Compare and decide	Two attempts at the same task; the learner chooses one of four equal verdicts — A is sound, B is sound, both, neither — and justifies it.
Select and justify	Choose the best option and give reasons.
Misconception confrontation	Predict the outcome of a case from three or four options, one of which embodies a named mix-up; then explain.

Every check tests specific lines of the lesson's cards, carries a hidden rubric for judging reasoning, and has one feedback block per result plus one per bound misconception. Each check has a variant pool — same idea and rubric, different story — reviewed and approved in Studio; fresh checks and Revision always draw a variant the learner hasn't seen. More in Mastery & the Lesson's Check.

How the evidence stays honest

Any on-screen task whose answer is visible can be outsourced in seconds. That matters twice: it would corrupt marks, and — the deeper problem — a faked check writes a false "Solid" into the learner model and routes the learner past the support they needed. Maestro doesn't fight this with lockdown or detection. It layers the design so real understanding is the cheapest thing to have:

Layer	Built in Studio	Run in Journey
Honesty economics	Feedback written per result, including Not sure	No grade on checks; no correctness in-flow; "I'm not sure" defers the belief write
Your own thinking	Required reasoning; hidden rubric; declared tempting errors	Reasoning judged; "Look again" quotes the learner's own earlier answer
Per-learner variants	Approved variant pools; milestone worlds with per-learner details	Unseen variants for every re-check
Locked kept work	—	First answer kept; milestone work locked but readable at submission
Spoken defense	Defense criterion in the rubric and weights	Live voice questions on the learner's own locked work, every milestone
Belief over time	—	Mastery settles only after an idea survives a later return
Signals support, never accuse	—	Process signals lower evidence weight against the learner's own baseline; never shown as flags

The integrity reasoning behind each layer is in Academic Integrity.

The Milestone Guide

Every milestone is entered through its Milestone Guide — never directly. The guide is readable at any time, even while the milestone is still ahead. It sets out the outcomes covered, the facts (weight, pass mark, no time limit), "How it goes" in five steps, what the learner will make, the "What counts" table of criteria with their % weights, what AI may and may not be used for, and what the spoken defense asks. Its main button follows the learner's progress:

Moment	Main button
Milestone still ahead	"Opens after Topic N" (greyed, with the reason)
Required Revision items untried	"Start opens after Revision · N left", with a link to Revision
Ready	Start assessment
Work under way	Continue your assessment — back to the task they were on
Submitted, defense not finished	Continue your defense — resumes at the same question

Once results are out, the milestone's doors open the finished assessment page instead; the guide stays readable from its own button.

Choose your world

Starting an assessment, the learner chooses a world — a context for the brief, from a small set approved in Studio with an equivalence check across worlds. The world gives them their brief (role, audience, situation, limits) and their files, and some tasks carry per-learner material. The rubric and standard are identical across worlds; only the context differs.

The assessment blocks

Every milestone uses one page frame; only the tasks and their blocks differ. Each task has a title, a one-sentence purpose, the criteria it counts toward ("Counts toward: … %"), an estimate in minutes, and an AI posture shown as a neutral label. Tasks are built from a fixed registry of 13 blocks:

Block	The learner…	Block	The learner…
Write	writes prose to a word range	Prompt Bench	runs prompts on the course tools; runs are saved
Fill in	completes labelled fields, a grid or a canvas	Calculate	builds a small worked calculation
Choose and justify	picks, then gives reasons	Upload	adds a file, within type and size limits
Compare	sets things side by side against criteria	Sources	records each source and where it is used
Grid	fills rows and columns	Field notes	records anonymised conversations, with consent
Mark up	highlights, tags and fixes parts of a text	Arrange in order	sequences items and explains each
Plan	lays out steps over time

An assessment that needs something outside the registry is flagged for the subject expert rather than forced into the wrong block.

Building the work

• Autosave with version history. Every block saves as the learner works, and earlier versions are kept — also serving as process evidence.
• File uploads with accepted types, size limits and a per-block count.
• A daily work log and a step tracker beside the tasks.
• No course navigation and no Atlas inside the assessment; the only way out is the page's own back link.

Before submitting

Submission is also the start of the defense, so the page prepares the learner first: a microphone and speaker test, a reminder to find a quiet place, and a plain note that the work becomes locked but readable. The button says exactly what happens: "I am ready: submit and start my defense".

The spoken defense

The defense is a live voice conversation. The page splits: the learner's locked work on the left, the questions on the right. The learner can cut in at any time and the examiner stops to listen. If the learner leaves, their place is kept; "Continue your defense" picks up at the same question.

Report & results

The weighted report marks each criterion on four levels and combines them with the % weights from the course's Rubric and Weighting review — including the defense criterion, marked from the transcript. A finished assessment opens on its own page with two tabs: Your results (the mark, what each criterion found, feedback per part, the defense transcript, and the pieces to strengthen when below the pass mark) and Your submitted work, locked and read-only. The course page's Your results tab shows that course's assessments; My results lists them across all the learner's courses. Each approved milestone also has its own gradebook column, with its weight, in the institution's LMS. Released course versions are frozen, and each learner stays on the version they took.

The binding contracts

Contract	What it binds
Lesson publishing	A lesson publishes live only with its required pieces: the check and its approved variants, the mix-up message, the second way in, the firm-up prompt, the worked example and the look-again probe. Publishing refuses an incomplete lesson rather than letting a learner page fail.
The milestone pack	Every rubric criterion is served by a task; every task serves a criterion or named process evidence; every block's configuration is complete; AI postures match the course's AI rule; no block outside the registry. The pack is frozen per build, so a later edit never reaches work already under way.
Typed answers	Every Try it, check and block has a schema for its configuration and its answers, and the server validates every answer.

Design laws — do not violate these

• Lesson checks are never graded. Only milestones produce the grade, and a grade never re-decides a lesson's result.
• No correctness feedback in-flow during a check. The first answer is kept; the result card comes after submission.
• A check is never a simple right/wrong quiz. Response + reasoning + confidence, always; "I'm not sure" always available and penalty-free.
• Learners see only learner words. Solid · Nearly there · Not yet · Not sure · A mix-up — never band names, engine words or scores on lesson results.
• Every milestone goes through its guide and ends in a spoken defense.
• Nothing reaches a learner unreviewed. Variants, worlds and remediation pieces are approved in Studio first.
• Behavioural signals support, never accuse. They weight evidence and prompt support; never surfaced as cheating flags or used as standalone proof.
• Accessibility is first-class. Keyboard operable, screen-reader supported, working at phone width, with a text alternative ("Read instead") wherever content is a video.