Forest Learning Stack

[00] System map

One map. Five layers. Same picture for the learner, the Guide and the parent.

Every arrow already has a tool behind it somewhere at Forest. None of them talk to each other. The stack is the wiring, not a replacement for what works.

At rest Press Play the flow to walk one real Tuesday through the stack, or click any layer to read what it owns.

Stack layer Existing tool or sub component Selected layer Data moving in the current step Playlab app, live and embedded in the layer above it

[01] Learner GPS · the hub

Everything Olivia is working on, in the order it matters.

Learner facing by default, readable by a parent in ten seconds, and nothing in it is typed twice.

Goals in flight

3

1 due Friday

Evidence this week

11

Pulled in automatically

Sources connected

5

Last sync 12 minutes ago

Hand entry required

0

Down from 6 fields per week

Goals in flight

Learner set · Guide approved

The marked row is today’s focus. The learner sets it each morning, and all three coaches push on it.

Evidence stream

Pulled from connected tools, nothing typed

This week

What the parent sees

Same record

No parent portal, no Guide writing a digest at 9pm. The parent view is a permission on the same record.

Connected sources

Khan Academy IXL NoRedInk Lexia Core5 Journey Tracker Math Academy · queued Writable · queued

Each source is an adapter into the shared record. Swap Khan for Math Academy and the goals and rubrics do not move.

[02] Gauge · self-direction, made visible

Habits you can see are habits you can coach.

Self direction is inferred from educator observation today. The Gauge turns four indicators into trended data drawn from behavior the learner already produces.

Behaviors logged this week

Source in every row. No score without an observable moment.

Every row opens the moment it came from. That is what makes a score defensible to a parent and coachable by a Guide.

How the score is computed

Playlab app

A Playlab app with Python execution reads the week’s events and returns scores plus the evidence ids behind them. The rubric is a file the Institute owns, not a model’s opinion.

# gauge_scorer · Playlab app · Python tool events = load(learner, week) score = rubric.apply(events) # Institute SDL rubric v3 return { "initiative": 3.9, "evidence": ["ev_8812", "ev_8830"], "confidence": "observed" }

If a score has no evidence

It does not post. A silent week is real information, and hiding it behind a number is how dashboards start lying.

[03] Coaches · three copilots, one script

Three coaches. Same goal, same rubric, same evidence.

What makes this different is not the chat. It is that all three coaches read the same record and push the same direction on the same day. Full app specs are on the Playlab layer screen.

Learner Coach

Steps first. Never hands over the answer.

Open in Playlab

Playlab app

Learner Coach

Loading the live app.

This viewer blocks embedded frames. The app runs normally in a tab.

Open in Playlab

Runs live in Playlab. If the panel stays blank, this host is blocking cross-site frames: open the app directly.

Shared context, all three coaches

Goal
Rubric
Gauge
Last work

Injected into every coach conversation. It is why the parent’s suggestion at 7pm matches the Guide’s Socratic block at 9am.

Alignment check

In sync

Same goal referenced

Learner, Guide and parent, past 7 days

3 of 3

Mixed messages flagged

Parent pushed speed while Guide pushed revision

1

Median time to adult response

Learner asks for help to a human reply

4h 10m

Adult alignment becomes measurable the moment all three conversations run on one platform.

Why three, not one

A learner develops faster when the signals stop contradicting each other, not because one adult got better advice. Three coaches is a coherence design, not a feature count.

[04] Assessment · authentic, and lighter to run

Keep the oral defense. Delete the paperwork around it.

Badges, portfolios, defenses and exhibitions are the part worth protecting. What the stack changes is the hours a Guide spends assembling the evidence beforehand.

Artifacts this cycle

14

Auto pulled with version history

Defenses scheduled

2

Friday, Ridge Studio

Guide prep time

38m

Was 3h 20m per defense

Rubric coverage

5 of 6

One criterion has no evidence yet

Artifacts and versions

Everything the defense will draw on

A learner who revised four times has shown something a final grade cannot, and the defense should be able to point at it.

Rubric, live

Quantitative reasoning · Ridge Studio v4

Gap before Friday

1 open

Found by reading the rubric against the evidence, before the defense rather than during it.

Assessment and Portfolio

Live

Maps evidence to criteria, names the gaps, and runs a two minute teach back before a defense. It does not score. Running live below.

RAG: rubric v4 RAG: learner artifacts Transcript visible to Guide

What stays human

The badge, the verdict and the exhibition judgement. The stack assembles and rehearses. A person decides whether a learner has mastered it.

Authentic Assessment and Portfolio

Today it works from what a learner uploads and the rubric they name. Once the record exists, artifacts and versions arrive on their own.

Playlab app Open in Playlab

Playlab app

Authentic Assessment and Portfolio

Loading the live app.

This viewer blocks embedded frames. The app runs normally in a tab.

Open in Playlab

Runs live in Playlab. If the panel stays blank, this host is blocking cross-site frames: open the app directly.

[05] Pathways · aptitude to opportunity

Open the window years before a transcript does.

A Forest learner cannot reach YouScience until middle school or PrinciplesYou until high school, but the aptitudes are visible long before the tools are. Pathways opens age appropriate doors on evidence the GPS and Gauge already hold.

Connections surfaced

Filtered by studio age gate

Outcomes Pathways will actually track

Outcome	Measured today	With the stack
Apprenticeships	Anecdotal	Placement, duration, completion
Dual enrollment	Counted at the end of the year	Eligibility surfaced when the evidence appears
Mentor contact	Not recorded	Sessions, follow through, learner rating
Offers and aid	Self reported by families	Recorded against the learner’s pathway
Gainful employment	Not tracked past graduation	Alumni check in at 6 and 24 months

Aptitude signals in use

Drawn from evidence, not a survey

Pathways and Opportunity Matching

Live

Retrieval over the mentor network, programs, dual enrollment and aid. The age gate is set per studio by the school, not by a vendor’s grade band. Running live below.

If a pathway is gated

It says who can open it. A third grader sees “ask your Guide about visiting a design studio,” not a padlock and a sales page.

Pathways and Opportunity Matching

It explains what evidence produced each suggestion, and it widens rather than narrows.

Playlab app Open in Playlab

Playlab app

Pathways and Opportunity Matching

Loading the live app.

This viewer blocks embedded frames. The app runs normally in a tab.

Open in Playlab

Runs live in Playlab. If the panel stays blank, this host is blocking cross-site frames: open the app directly.

In practice · three real learners

The same stack, three very different Tuesdays.

The outlined card follows the learner selected in the top bar. Each story runs today, what changes, and the signal that would tell us it worked.

Elin · grade 3 · Meadow Studio

Wants to write a series of stories set in worlds she designs.

Writing

Today

Journey Tracker entries are typed by hand and hard for a parent to parse. She brainstorms in ChatGPT, but nothing there knows the storytelling rubric, so the feedback is not aligned to what good looks like. Career and mentor windows stay locked until middle school.

With the stack

Her GPS holds the character bible, drafts, deadlines and rubric targets in one view her mother can read. The Learner Coach helps her sharpen dialogue and run a revision checklist without writing a line for her, the Guide Coach proposes a critique circle on the same rubric, and the Parent Coach offers one move when she stalls.

Signals we would watch

Rubric scores rise on voice and structure, reflections get more specific, and portfolio evidence grows in coherence rather than count.

Olivia · grade 7 · Ridge Studio

“I don’t want answers. I want it to teach me.”

Mathematics

Today

She moves between Khan, ChatGPT and Khanmigo, and the hints often do the thinking for her. Khan’s dashboard does not know her Journey Tracker goals, so three people reconcile two pictures every week. Without a Gauge, progress can quietly mean guessing or outside help.

With the stack

One GPS holds Khan progress and her goals together. The Learner Coach compares two strategies and asks her to pick one, the Guide Coach queues an error analysis for tomorrow’s Socratic block, and the Parent Coach sends one line rather than a lesson.

Signals we would watch

Fewer hint chasing loops, stronger explanations at Exhibition, better performance on cold problems, and measurable CCRA+ growth.

Luke · grade 11 · Summit Studio

Wants to learn faster across every subject at once.

Study systems

Today

He stitches YouTube, PDFs, slide decks, voice memos and flashcard apps into one place by hand. Most of the effort goes into hunting and copying, and none of it becomes portfolio evidence without a second round of work.

With the stack

The GPS ingests his videos, PDFs, notes and audio. The Learner Coach builds summaries, flashcards and quizzes grounded in his own materials and pushes reflect, plan, act, iterate instead of passive consumption. Assessment surfaces the gaps before the Exhibition rather than during it.

Signals we would watch

Less tool wrangling, better recall on cold questions, faster exhibition preparation, and rising rubric scores on explain your steps.

Measurement · assertion to evidence

We judge this the way clinicians judge UpToDate: time saved and better decisions.

Forest already measures more than most schools. The honest gap: self direction is inferred from observation, and platform data captures procedural gain better than understanding.

CCRA+ improved, middle school

94%

2023–24 external check

CCRA+ improved, high school

63%

2023–24 external check

Parent and student NPS

4.87

Out of 5

Schools designed or launched

60+

Through the Institute

Today at Forest, and with the stack

The marked row is the one that changes most

Domain	How we prove it today	How we would prove it with the stack
Mastery	Badges, oral defenses, performance tasks and portfolios against clear rubrics. Time intensive to assemble.	Same evidence, assembled automatically. Quality and pass rates stay the measure, preparation stops being the cost.
Academics	Triangulated platform data plus CCRA+. Platform data favours short term procedural gain.	Platform mastery plus transfer tasks and error analyses, with CCRA+ as the external check.
Self direction	Inferred from educator observation. Real, but anecdotal and uneven between Guides.	Logged and trended from observed behavior: clear why, daily plan, if–then moves, feedback before final, iterate.
Experience	Parent and student NPS around 4.87 out of 5.	NPS plus reduced friction: fewer reconciliations, fewer places to look, less hand entry.
Pathways	Strongest for older learners through PrinciplesYou and YouScience.	Real outcomes tracked: apprenticeships, dual enrollment, offers and aid, employment, linked earlier to aptitude.
Equity	Reviewed periodically, by hand.	Usage and outcomes monitored by learner group, with a triggered support when a gap opens.
Adult alignment	Not measured. There is no instrument for it.	Mixed messages across learner, Guide and parent, and time to a consistent response. Only possible once all three run on one platform.

Leading indicator

Reconciliation events per week

How often a Guide or parent has to compare two systems to answer one question. This should fall to zero in the first term, and it is the cheapest proof the GPS is working.

Leading indicator

Hint chasing ratio

Requests for the answer divided by requests for a next step, per learner per week. The Learner Coach either moves this or it is not doing its job.

Leading indicator

Guide hours returned

Hours moved out of assembly and reporting and into mentoring and relationship. Measured by Guide time log, not by a survey.

Build · the Playlab layer

Playlab is the runtime under the Coaches, the Gauge and Pathways.

The GPS, the Gauge dashboard and the Assessment record are surfaces an engineer builds. Everything that talks, retrieves or scores is a published Playlab app, which is what makes the coaching layer inspectable, versionable and shippable to a partner school in an afternoon.

Six apps, and where each one is embedded

Select a row to run it below

Learner Coach

Running live in the page. Pick any row above to load it here.

Playlab app Open in Playlab

Playlab app

Learner Coach

Loading the live app.

This viewer blocks embedded frames. The app runs normally in a tab.

Not built yet. It needs Python execution rather than a chat surface, so it is the one app in the stack still to make.

Open in Playlab

Runs live in Playlab. If the panel stays blank, this host is blocking cross-site frames: open the app directly.

Why an app and not a raw model call

[01]

Every conversation is visible to the school

A Guide can see how the Learner Coach actually answered, not how we hoped it would. Over hinting shows up as data instead of as a suspicion.

[02]

Guardrails are a published prompt, not folklore

Steps first and the never write it for them rule live in the app the Institute owns. When a rule changes it changes once, with a version, for every school running the pack.

[03]

Retrieval keeps the coach on the school’s own material

The rubric, the SDL definitions, the family systems playbook and the learner’s own work are the corpus. That is what makes feedback aligned to what good looks like at Forest rather than generically reasonable.

[04]

Publishing is the scaling mechanism

The Institute publishes one app pack. Indiana, the ten Georgia districts and the UPenn Institute install it and start from the same coaching language on day one.

Playlab back end, this week

Ridge Studio

Example figures. The shape is the point: usage and outcome by learner group, in one place, without a data request.

The app pack

One installable set the Institute maintains: the apps, the SDL rubric, the coach prompt library and the guardrail configuration. A new microschool gets the coaching layer before it gets a data layer, which is the right order to pilot in.

Forest School Indiana microschools Georgia · 10 districts UPenn Institute · 2026

Honest limit

Playlab covers the conversation and scoring layer. It does not give you the GPS. The unified record, the adapters into Khan and IXL, and the parent readable view are an engineering build, and that is what the grant funds.

Build · interoperability and guardrails

Five layers, one record, handoffs in both directions.

Parts of this stack exist today and are siloed. What does not exist is the API first layer that lets each piece talk to the others. This is the smallest model that makes the five cohere.

Core entities

The marked row is the one every layer depends on

Entity	Owned by	Read by	Notes
learner	Roster and identity	All five layers	One id across every tool. No shadow rosters.
goal	GPS	Coaches, Assessment, Gauge	Learner set, Guide approved, dated, evidence linked.
rubric	Institute	Coaches, Assessment	Versioned. A coach cites the version it was reasoning against.
evidence	Assessment	GPS, Gauge, Pathways	Artifact plus version history plus who reviewed it. The currency of the whole stack.
event	Gauge	GPS, Coaches	An observed behavior with a timestamp and a source. Never a bare score.
consent	Family	All five layers	Per learner, per surface. A parent can see what a coach can see.
pathway	Pathways	GPS	Aptitude, interest, opportunity, outcome. Age gated per studio.

One handoff, in full

The Learner Coach finishes a session and writes back. Every layer that cares is current, and nobody typed anything.

// POST /events { "learner": "lr_0192", "source": "coach.learner", "goal": "gl_ratios_7", "rubric": "qr_v4", "indicator": "persistence", "observed": "wrote why-it-works before hint", "artifact": "ar_4471#v3", "visible_to": ["learner", "guide", "parent"] }

Guardrails, in from the start

Not a policy document. Four things configured in the apps and checked in the pilot review every week.

Steps first default

Every interaction opens with a next step or a question, never the finished answer. Answer requests are logged as their own event type, so over hinting is visible.

Citation checks

A coach that references a rubric, standard or source names it and its version. Unsourced claims are flagged in the transcript.

Educator oversight log

Every conversation is reviewable by the Guide of record. Sampling is scheduled rather than incidental, and recorded.

Bias review

Usage and outcomes read by learner group every cycle. A gap triggers a support, as part of the pilot protocol rather than a later analysis.

Build · what is next

Coaching layer first. Data layer second. Network third.

The apps ship inside Studios without waiting on interoperability. That order also gives the engineer real transcripts to design the data layer against instead of a guess.

Phase 01 · now

MVPs inside Studios

In flight

Five apps live in Playlab against rubrics that already exist, plus playbooks and simple dashboards. No data layer required.

Learner CoachGuide CoachParent CoachGauge ScorerWeekly improvement cycle

Phase 02 · next term

The data layer and the GPS

Grant funded

A full time engineer builds the API first record, the adapters into Khan, IXL, NoRedInk and Lexia, the GPS dashboard and the parent readable view.

Entity modelPlatform adaptersGPS dashboardParent viewConsent and roster

Phase 03

Assessment and Pathways close the loop

Planned

Artifacts, versions and feedback flow into the record automatically, and Pathways starts tracking real outcomes.

Artifact ingestDefense rehearsalMentor networkOutcome tracking

Phase 04

Validate across a public pilot network

Planned

Forest plus partners on shared metrics and weekly improvement cycles, scaling through the Institute’s adult formation and state partnerships.

Indiana microschoolsGeorgia districtsUPenn Institute · 2026

What we already have

The Institute’s research base and learning frameworks.
Rubrics, SDL indicators and assessment criteria to feed the AI.
A coach prompt library written by practitioners.
Human structures that already work: mixed age Studios, Socratic Guides, mastery evidence, family partnership.

What we most need now

Interoperability. The API first layer between the five parts.
A unified dashboard that is genuinely the single source of truth.
A clean interface a learner, a Guide and a parent can each use without training.
A full time engineer to build all three.

Overview · the case

The human architecture already works. What is missing is the operating system.

Forest was founded in 2018 to grow interdependent, self directed learners. Seven years in, the model holds. The tools around it do not.

The macro problem

We are graduating generations of dependent learners. Creativity, academics and critical thinking are sliding, anxiety and disconnection are rising, and employers report graduates who struggle to problem solve and self start.

Why

The system rewards compliance.
Well intentioned adults overfunction where learners should own the work.
We confuse coverage with learning.
Seat time and grades still outrank demonstrated mastery.
Families opt out because progress is opaque: letters and averages instead of evidence.

The local problem, at a school that gets this right

Even in a well designed self directed school the tools are fragmented. Goals live in one place, progress data in another, parent updates somewhere else. Rubrics do not talk to the learning platform, and coach notes do not connect to assessment records.

A learner, a Guide and a parent each work from a different picture of what is happening. We have the philosophy. We do not have the coherence.

What the stack is supposed to change, per audience

Four claims, each with an instrument on the Measurement screen. A claim without one is a brochure.

Learners

Faster mastery, stronger self management, clearer next steps, and purpose that comes from work connected to something beyond the studio.

Educators

More time mentoring and building relationships, less time assembling evidence and writing updates that a record could produce.

Families

Progress they can read, and just in time support that preserves the learner’s agency instead of quietly transferring the work to a parent.

Leaders

Seed a learning community anywhere, with coherence from day one rather than a folder of disconnected tools.

The human foundation this sits on

Mixed age Studios replace classrooms.
Guides coach with Socratic questions instead of lectures.
Learners set meaningful goals and work at their own pace.
Mastery is shown through badges, portfolios, oral defenses, performance tasks and public exhibitions.

Core skills are practised daily.
Transdisciplinary projects connect required knowledge to real problems.
Progress is based on evidence, not seat time.
Families and mentors partner through regular check ins.

The stack sits on top of these structures. If it ever starts replacing one, that is the signal we built the wrong thing.

Not just a private school

Indiana

Consulting with the Department of Education on the nation’s first network of public microschools featuring self directed learning.

Georgia

Partnering with the Department of Education on models in up to ten districts.

University of Pennsylvania

Launching the Microschool Institute in early 2026.

A few core layers run school wide. Most other tools stay situational. The stack is the few, not the many.