AI answers instantly, and the answer is where the learning stops. Trailhead runs it the other way round: you invent the rule, Scout argues with you, then the lesson lands — and you prove it on real work a reviewer scores. Students and working teams, same method, under a semester.
Now piloting with El Segundo Unified School District.
Employer brief · Aerospace
Reduce ground-test data review time without losing anomaly detection
Compare
Probe
Learn
Apply
Teach
Try it new
Your rule says “downsample to save time.” Case B did that and missed the spike. What does your rule need to say about the spike?
A pressure spike under 200ms. So the rule is: never downsample below the shortest anomaly you can’t afford to miss.
Rule revised once · Lesson unlocked
Transfer probe: passed
Work samples predict job performance better than credentials
The full course, self-paced, with scheduled check-ins — never dragged out
Minimum reviewer rating (of 10) to earn the AI-Ready mark
Work-sample validity r=.54 vs. credentials r=.10 — Schmidt & Hunter, 1998.
Tool-specific AI training expires in months. What lasts is judgment: decomposing the work, checking the confident-but-wrong number, knowing when to overrule the model. Trailhead builds that the way the research says it is built.
Invent before you are told, wherever there is hidden structure to find; models and critique where it is craft. A coach that problematizes instead of explaining. Learning by teaching. Transfer measured without the coach, not task completion. Built on Stanford learning science and Cognitively Guided Instruction.
Students solve briefs posted by local employers. Teams solve their own recurring workflow, scoped by a sponsor. When learners submit, Scout checks the work against criteria a person approved and asks what to look at, never giving the answer; a teacher or sponsor makes the call. Every artifact is scored on the same six-dimension rubric by the people who actually use the work. Scout never scores.
The course finishes in under a semester. Modules are self-paced; check-ins are scheduled — monthly for schools, weekly for teams or fully self-paced. Nothing waits on a calendar to unlock.
The Trailhead Loop. The first three are the lesson; the last three are the project. The lesson cannot be reached by giving up — only by inventing.
Two cases set in your own brief that differ on one deep feature. Pick the better one, say why, and invent the rule.
Scout pokes holes with a counter-case. You revise the rule. It never tells you the answer here.
The lesson is built from your own rule and your cohort’s. It unlocks once you have invented a rule and answered the counter-case — not once you give up.
Produce the real artifact: the employer brief, or your team’s own workflow.
Write the rules for Scout. It runs only your rules on a fresh case and says where they fall silent. Fix what broke.
Two problems you have never seen, without Scout: one cold, one with a new resource to learn from. That is what we measure.
Meet Scout
“Typing ‘I don’t know’ three times unlocks nothing. Inventing a rule and defending it once unlocks the lesson.”
Scout does not count attempts. It reads what you actually wrote. Until your rule names a real difference between the cases, it offers counter-cases, not explanations. After that, it teaches — and then it becomes your student, running your rules on a case it has never seen so you can see where they break.
Argues, then teaches
Never writes the artifact
Facilitator-visible
Every thread, every time
idk. Just tell me which one is right.
Not yet — the lesson opens when you’ve named one difference. Look at the two spec lines. What is different about the units?
One converts N·m to lb·ft and gets a number ten times too big.
That is a rule you can defend. Write it down as “I’d flag a conversion when…” and I’ll try to break it.
A cross-grade team works a real employer brief through seven self-paced modules, finished inside a semester with monthly check-ins the teacher leads. Employers post and assign the tasks; the teacher sees the learning.
COPPA / FERPA by design
Parent consent flows, role-scoped access, and audit trails are part of the schema — not a bolt-on.
Teachers see everything
Every Scout conversation is visible to staff. Crisis language routes to a human immediately.
Your team lead runs it and sees the metrics. Any team you designate as sponsor — the same team or a different one — posts the projects and scores the playbook. Weekly check-ins, or fully self-paced.
Your text is never trained on
Every model call runs through a no-train gateway. Workspaces are isolated and every access is audit-logged.
Client data stays out
Every outside-AI step carries a do-not-share boundary and PII redaction. Sponsors score artifacts; they never read transcripts.
Task completion does not predict transfer, so it is not the headline. These four are, for every learner, shown to the teacher or to your organization.
Transfer
Can they solve a problem they have never seen, with the same deep structure — and learn from a new resource on their own? Measured before and after, without Scout, in a different domain each time.
Response to pushback
When Scout argues, do they revise the rule or defend it with evidence — and does it get better? Persistence after failure, not time on task.
Teach-back accuracy
When their rules run on a fresh case, how often are they right? And how much better after the fix?
Discernment calibration
On audits: real errors caught versus good values wrongly flagged. Confidence that matches the evidence.
Deliberately not a score:
Minutes on taskStreaks and pointsLessons completed as a scoreAnswers accepted from the AIUnder a semester, self-paced, with check-ins you choose and reviewers who commit to scoring the work. We set it up with you.