Lantern, a worked example · The walkthrough
Schedule a student assessment
One capability of Lantern, an invented product, written down the way Intentset asks for it: 15 records, each a Markdown file a person reads in under a minute, which together connect a reason for the product to change to the code that would deliver it and the check that would hold it.
Every record is open on this page. Read the 5 steps in order and the chain closes; open any record to see exactly what a person writes and what an agent reads. The map draws the same 15 records and every link between them in one picture.
Lantern is invented and nothing here runs. Every record but the knowledge is a draft, the code paths are planned rather than written and the check has not been executed. The knowledge is approved and reviewed, so the last step can show what publication writes for it. What the example shows is the shape of a model and what it makes visible, not a verification report.
1. Why the product is changing
A model starts with the reason rather than the feature. Lantern's intent is to take last-minute administrative work off teachers, and the outcome it should produce is less friction preparing an assignment. The outcome is not a slogan, because two measure records say how it will be judged: minutes a teacher spends preparing an assignment, read by an observed task study, and assignments scheduled a day or more ahead, read from product analytics. Each names its metric, baseline, target, window and source, so a reviewer can argue with the target. One baseline reads unknown, which is a thing a record is allowed to say and a thing a reviewer can see.
PRD-LANTERN · Lantern example
| Field | Value |
|---|---|
| ID | PRD-LANTERN |
| Type | product |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
Scope
An illustrative student assessment workflow used to review Intentset. Lantern is an invented product; nothing here describes a real system.
INT-PREPARE · Prepare learning in advance
| Field | Value |
|---|---|
| ID | INT-PREPARE |
| Type | intent |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Parent | PRD-LANTERN |
Rationale
Reduce last-minute administrative work for teachers.
OUT-PREPARE · Reduce assignment preparation friction
| Field | Value |
|---|---|
| ID | OUT-PREPARE |
| Type | outcome |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Parent | INT-PREPARE |
Measure
Success is judged two ways: teachers spend less time preparing an assignment, read by an observed task study, and more assignments are scheduled ahead of the day they open, read from product analytics. Each has a measure record under this outcome with its metric, baseline, target, window and source; both are read 90 days after the pilot.
MEAS-PREPARE-MINUTES · Teacher minutes spent preparing an assignment
| Field | Value |
|---|---|
| ID | MEAS-PREPARE-MINUTES |
| Type | measure |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Parent | OUT-PREPARE |
Method
An observed task study: a teacher prepares three assignments of the kinds the pilot schools set most, timed from opening the class to the last assignment being scheduled. The study runs once before the pilot to take the baseline and once in the window. The median over participants is the reading; the product reviewer signs off the comparison.
MEAS-PREPARE-ADOPTION · Assignments scheduled a day or more ahead
| Field | Value |
|---|---|
| ID | MEAS-PREPARE-ADOPTION |
| Type | measure |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Parent | OUT-PREPARE |
Method
Product analytics counts every assignment scheduled in the window and the share whose scheduled time was a day or more after the time it was created. A teacher who prepares in advance schedules ahead; one working at the last minute schedules for the same day. The baseline is unknown because nothing is scheduled ahead before the capability exists.
2. What the product promises
The capability is the ability that persists: teachers prepare and distribute published assessments to classes. Under it sits the one behavior this example follows, and a behavior is written as a promise including how it fails — success records the schedule, and an invalid time, an unauthorized class or an unpublished assessment is rejected without creating one. Two rules govern it, each written once rather than copied into every behavior it constrains, and the scenario is a single worked case: it shows the promise, and does not prove every path. Read the behavior's last sentence: delivery tolerance, cancellation and retry are still draft decisions. A gap written down is a gap somebody can close.
CAP-ASMT-ASSIGN · Student assessment assignment
| Field | Value |
|---|---|
| ID | CAP-ASMT-ASSIGN |
| Type | capability |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Parent | OUT-PREPARE |
Overview
Teachers prepare and distribute published assessments to classes.
BEH-ASMT-SCHEDULE · Schedule a student assessment
| Field | Value |
|---|---|
| ID | BEH-ASMT-SCHEDULE |
| Type | behavior |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Parent | CAP-ASMT-ASSIGN |
| Governed by | RULE-ASMT-FUTURE, RULE-ASMT-AUTH |
Behavior
A teacher submits a future release time for a published assessment and a class.
Preconditions
The teacher is authorized to assign to the class. The assessment is published and the requested time is in the future.
Outcomes
Success: the schedule is recorded and students gain access at the release time under the eventual implementation tolerance. Failure: invalid time, unauthorized class, or unpublished assessment is rejected without creating a schedule. Delivery tolerance, cancellation, and retry semantics remain draft decisions.
RULE-ASMT-AUTH · Require assignment permission
| Field | Value |
|---|---|
| ID | RULE-ASMT-AUTH |
| Type | rule |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
Constraint
The server must reject scheduling unless the assessment is published and the actor is authorized to assign it to the requested class.
RULE-ASMT-FUTURE · Require a future release time
| Field | Value |
|---|---|
| ID | RULE-ASMT-FUTURE |
| Type | rule |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
Constraint
The authoritative server time must be earlier than the requested release time when the schedule is accepted.
SCN-ASMT-SCHEDULE · Accept an eligible schedule
| Field | Value |
|---|---|
| ID | SCN-ASMT-SCHEDULE |
| Type | scenario |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Illustrates | BEH-ASMT-SCHEDULE |
Given
An authorized teacher, a published assessment, and a future release time.
When
The teacher schedules the assessment for a class they manage.
Then
The system records the schedule and does not expose the assessment to students before its release time.
3. Where the promise is delivered
One slice is accountable for the behavior, which is what makes a change to it reviewable: an agent handed this slice's context knows what it may touch. The contract is what another part of the product may rely on, and it says what a breaking change is, so a consumer is reviewed rather than surprised. The decision record carries the reasoning that would otherwise be lost in a pull request, and joined the model by gaining frontmatter, its text unchanged. The slice says plainly that its code paths are planned.
SLICE-ASMT-SCHEDULE · Student assessment scheduling slice
| Field | Value |
|---|---|
| ID | SLICE-ASMT-SCHEDULE |
| Type | slice |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Implements | BEH-ASMT-SCHEDULE |
| Exposes | CONTRACT-ASMT-SCHEDULE |
| Informed by | ADR-ASMT-SEAM |
Responsibility
Accountable implementation owner for scheduled assessment delivery. Code paths are planned; this example is L1 only and makes no implementation-conformance claim.
Public contract
ScheduleAssessment accepts a class, assessment, and release time and returns a stable schedule identifier or a typed rejection.
Verification
TEST-ASMT-SCHEDULE defines a manual pilot review, with no execution evidence yet.
CONTRACT-ASMT-SCHEDULE · ScheduleAssessment contract
| Field | Value |
|---|---|
| ID | CONTRACT-ASMT-SCHEDULE |
| Type | contract |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
Interface
Input: assessmentId, classId, releaseAt (UTC timestamp). Success: scheduleId. Rejections: unauthorized, unpublished, invalid-time. Draft transport representation is not yet selected.
Compatibility
Removing a field or tightening accepted times is breaking. Consumers must be reviewed when preconditions change.
ADR-ASMT-SEAM · Keep backend access behind the slice seam
| Field | Value |
|---|---|
| ID | ADR-ASMT-SEAM |
| Type | decision |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
Context
Frontend and backend resources may live in different directories while serving one product behavior.
Decision
The scheduling client adapts transport into domain results. The backend handler enforces authorization and rules; shared data composition has a technical owner.
Consequences
Ownership stays explicit without moving all cloud resources into a frontend feature. Imports and bundle contents require checking.
4. How it is checked
One verification record defines the check: a manual pilot review of five cases, recording inputs, observed outputs, environment, commit, graph hash and reviewer. It states that the procedure has not been executed, and its expected result goes further — it cannot pass until the release-time tolerance is defined. That is the model reporting a hole in itself, which is the whole reason for writing checks down beside the promise they hold.
TEST-ASMT-SCHEDULE · Review student assessment scheduling behavior
| Field | Value |
|---|---|
| ID | TEST-ASMT-SCHEDULE |
| Type | verification |
| Status | draft |
| Owner | team-assessment |
| Visibility | internal |
| Audiences | engineering, product |
| Verifies | BEH-ASMT-SCHEDULE, RULE-ASMT-FUTURE, RULE-ASMT-AUTH, SCN-ASMT-SCHEDULE |
Procedure
In a pilot environment, run: authorized published/future request; past-time rejection; unauthorized-class rejection; unpublished rejection; student access before and after release. Record inputs, observed outputs, environment, commit, graph hash, and reviewer. This procedure has not been executed.
Expected result
The valid request is recorded, rejected requests create no schedule, and student access follows the reviewed release-time tolerance. The procedure cannot pass until that tolerance is defined.
5. What a customer may be told
The knowledge record is the guidance a teacher would read, kept beside the behavior it explains rather than in a separate system that goes quietly out of date. It is the one approved record in the example, and approval has a shape: a reviewer, a date, and a pin on each of the three sources it was reviewed against, so a change to the behavior or to either rule sends it back for review before it is published again. It also carries two tips, one sentence each, for the product to show beside the control that delivers the behavior and where the rule about the release time applies; the permission rule has none, because a tip is optional. The last fold is what intentset publish writes for it: the document with its provenance, and the help file a product's interface reads.
KB-ASMT-SCHEDULE · Prepare a scheduled student assessment
| Field | Value |
|---|---|
| ID | KB-ASMT-SCHEDULE |
| Type | knowledge |
| Status | approved |
| Owner | team-assessment |
| Visibility | customer |
| Audiences | teacher |
| Explains | BEH-ASMT-SCHEDULE, RULE-ASMT-FUTURE, RULE-ASMT-AUTH |
| Reviewed | team-assessment, 2026-10-11, pinned to BEH-ASMT-SCHEDULE, RULE-ASMT-AUTH, RULE-ASMT-FUTURE |
| Tip | BEH-ASMT-SCHEDULE · Pick a published assessment, a class you manage and a future release time. |
| Tip | RULE-ASMT-FUTURE · The release time must be in the future. |
Guidance
Choose a published assessment, a class you can manage, and a future release time. Review the schedule before confirming. If scheduling is rejected, check your class permissions, publication state, and chosen time.
intentset publish · The document and the help file, as written
What intentset publish writes for the customer projection, audience teacher, PRD-LANTERN pilot-1, role teacher, edition standard, from the example as it stands. The document keeps the record's body and replaces its frontmatter with provenance: the three sources it was reviewed against are internal, so they are counted and not named. The help file carries the two tips, keyed by the ID a control in the product carries, and names the document each comes from.
---
markset: 0
intentset-publication:
profile: intentset/publication/0.2
derived: true
projection: customer
source:
id: KB-ASMT-SCHEDULE
revision: 1
sourceHash: '8b56db9e7810bd2d0fea8cca87a43463188775945831d313a4ee01b68720182e'
sources: []
withheldSources: 3
audience: teacher
availability:
product: PRD-LANTERN
release: pilot-1
role: teacher
edition: standard
flags: []
reviewer: team-assessment
reviewedAt: '2026-10-11'
publishedAt: '2026-01-01T00:00:00Z'
---
<!-- Generated by intentset publish from KB-ASMT-SCHEDULE; do not edit. -->
# Prepare a scheduled student assessment
## Guidance
Choose a published assessment, a class you can manage, and a future release time. Review the schedule before confirming. If scheduling is rejected, check your class permissions, publication state, and chosen time.
{
"profile": "intentset/help/0.2",
"derived": true,
"projection": "customer",
"audience": "teacher",
"availability": {
"product": "PRD-LANTERN",
"release": "pilot-1",
"role": "teacher",
"edition": "standard",
"flags": []
},
"publishedAt": "2026-01-01T00:00:00Z",
"knowledge": [
{
"id": "KB-ASMT-SCHEDULE",
"title": "Prepare a scheduled student assessment",
"path": "KB-ASMT-SCHEDULE.md"
}
],
"tips": {
"BEH-ASMT-SCHEDULE": {
"text": "Pick a published assessment, a class you manage and a future release time.",
"knowledge": "KB-ASMT-SCHEDULE"
},
"RULE-ASMT-FUTURE": {
"text": "The release time must be in the future.",
"knowledge": "KB-ASMT-SCHEDULE"
}
}
}
The same records, for review
The records above are what an agent writes and a person approves. The map and the review pages are the same records as intentset serve draws them for the people who review a model: the capability and its behavior, who owns what, verification shown as missing because the check has not been executed, and what publication would release. A team's own Atlas stays inside the team; Lantern's is public because Lantern is invented.
Try the same thing on your own code
Nothing above was written by hand for this site. It is the shape an agent produces when it reads a capability out of a repository that already has the code, the tests and the decisions: the first capability modelled in Streamlane came out this way, and listed two rules with no test and three behaviors checked only in the browser.