SupportED Tutoring · Internal

Product Roadmap — Idea to the AI Tutor

The build sequence, the two release gates, and the sprint plan that has to hold for test-prep season. This is the team-facing version: real dates, real risks, real open decisions.

Core releaseNov 15Real students inside, finding the fringe cases
Prepped for studentsDec 15Staged release, shakeout complete
The real deadlineFebruaryExam-panic season. The revenue window
Cadence2-week sprints10 sprints from Sep 15 to Feb 1

At a glance

The whole roadmap in prose, for anyone who would rather read than trace a diagram — and for phones, where the charts below scroll sideways.

What we are building, and why it changed

Until August the platform was an operations tool: replace Acuity, then Skool, then Class Companion, and call it done when a subscription got cancelled. On 23 August that reversed. The platform is now the thing that proves the tutoring worked — every assessment question coded to a unit and a skill, so a parent can log in and see that their student actually improved. Joe calls it precision-guided tutoring. That single reversal pulled assessments forward from 2027 to the next build, parked the Skool clone, and is upstream of everything since.

The five stages

  • Idea — one platform, not four tools. Validated.
  • MVP1 · Operations — sessions book, hold and settle in-platform. Core is shipped; the admin suite, parent accounts, comms channels and self-serve signup are still open.
  • MVP2 · Proof — migrate the old test-prep platform, build the assessment engine and the skill profile, ship the parent dashboard. This is the Nov 15 release.
  • MVP3 · Membership — tiers, courses, group classes, affiliate tracking. Has to be ready before the February peak.
  • MVP4 · Cyborg — the AI tutor on top of clean assessment data, AccepteD credits, mobile.

The timing, honestly

Nov 15 is the core release — real students inside, finding the fringe cases. Dec 15 is prepped for volume. But February is the real deadline: exam-panic season is the revenue window that pays for the year, and a slip past December does not cost a month, it costs the season.

There are 4.4 sprints between now and Nov 15, and roughly three of them are migration — the old test-prep platform is 60–70% of the effort by Wenbo's own estimate. That leaves about 1.4 sprints for six admin items, parent accounts, the comms rebuild, the credit ledger and self-serve signup, on one part-time senior plus a junior. The date holds only if the scope behind it gets cut out loud.

What is actually blocking

Five decisions gate the build, and four of them are a sentence from Joe. The most expensive is D-1: what a self-serve signup actually gets — registration is still closed, an affiliate lands within the month, and there is nothing to sell them. D-2, the three-currency credit ledger (SupportED, TestED, AccepteD), blocks the data model outright: nothing can be built on a ledger nobody has defined.

On the board itself: every open issue is unassigned and unestimated, there are no sprint cycles configured, and self-serve registration has no issue at all. Which is why nothing on this page is a commitment yet — it is a plan awaiting an estimate pass.

This week's executive review

Two and a half minutes, regenerated every Monday from the live board, the week's calls, and where we actually sit against the gates. It is written to be heard once — it opens on the thing that matters and does not pad.

Sprint
Weekly briefing

Loading the latest review…

Generated Mondays. Board state, the week's decisions, what is blocking, and the ask.

The spine

Five stages. Each one is defined by a capability the business did not have before it, not by a list of features.

Stage 0
Idea
One platform, not four tools
Stage 1
MVP1 · Operations
Sessions run themselves
Stage 3
MVP3 · Membership
Revenue without a sales call
Stage 4
MVP4 · Cyborg
The data starts teaching

The delivery map

Ten two-week sprints, six workstreams, three gates. Bars are where work actually sits — not where we hope it lands. Read the migration lane first: it is 60–70% of the effort and it blocks almost everything to its right.

Lanes Foundations Migration Assessments Admin & roles Parent & comms Credits Growth Reduced capacity
Swipe to scroll — or read At a glance above
Workstream
S1Sep 15
S2Sep 29
S3Oct 13
S4Oct 27
S5Nov 10
S6Nov 24
S7Dec 8
S8Dec 22
S9Jan 5
S10Jan 19
FoundationsDecisions · specs
Ledger spec
MigrationOld test-prep platform
Audit + schema map
Tail
AssessmentsUnit + skill engine
Engine + skill profile
CS at-risk
Admin & rolesSUP-5 → SUP-8
Admin suite
Parent & commsSUP-9, SUP-10
Accounts + channels
CreditsThree currencies
TestED
Membership
GrowthSelf-serve · affiliate
Self-serve registration
Courses + groups
CapacityHoliday
Reduced
Gates
Nov 15Core release
Dec 15Prepped for students
FebSeason starts

Two lanes never run in parallel in reality — there is one part-time senior and one junior. Where bars overlap, one of them is waiting.

The build sequence

Solid arrows are the happy path. Dashed arrows are fail-loops — what actually happens when a gate says no. Diamonds are binary gates: they have one answer and it is yes or no. The right rail carries who owns it, how long it takes, and what it depends on.

Read Shipped Binary gate Fail path Decision / spec — solid = happy path   ---- dashed = fail-loop
Swipe to scroll — or read At a glance above
STAGE I — MVP1 · OPERATIONS MVP1 CORE — SHIPPED book · meet · summary · credits · stripe · csv OWNER Wenbo (lead) · Jay (junior) STATE live · registration still closed OPEN 6 admin + parent + comms issues ADMIN SUITE + PARENT + COMMS SUP-5 … SUP-10 · roles · channels · linking G1 Can a stranger sign up and pay? the commercial gate — decision D-1 SELF-SERVE REGISTRATION the blocker · no Linear issue exists yet No fix, then retest DECISION D-1 — what does a signup actually get? PRESSURE Heimler lands within the month TODAY registration is closed — nothing to sell Yes STAGE II — MVP2 · PROOF — THE NOV 15 RELEASE MIGRATION — OLD TEST-PREP PLATFORM 60–70% of total effort · Wenbo's own estimate TIME sprints 2–3 · Sep 29 – Oct 26 RISK the old database had no RLS DECISION D-4 — does student history come across? G2 Schema mapped and access model written? D-2 credit ledger · D-3 unit+skill schema WRITE THE TAXONOMY FIRST credits · roles and RLS · unit+skill JSON No fix, then retest Yes ASSESSMENT ENGINE + SKILL PROFILE every question coded to a unit and a skill TIME sprints 4–5 · Oct 27 – Nov 23 EXIT fringe cases surfaced — not zero bugs NOT hundreds of students. That is December G3 Real students inside, taking assessments? NOV 15 — core release, not volume CUT SCOPE OUT LOUD admin suite and parent account slip first No fix, then retest Yes SHAKEOUT + PARENT DASHBOARD fix what real usage broke · no new scope TIME sprint 6 · Nov 24 – Dec 7 RULE repair only — scope is frozen here OWNER Wenbo + Jay G4 Ready for volume? DEC 15 — prepped for students HOLD THE RELEASE better late in Dec than broken in Feb No fix, then retest Yes STAGE III — MVP3 · MEMBERSHIP — AHEAD OF THE SEASON MEMBERSHIP + COURSES + GROUPS tiers · progression gating · group calendar TIME sprints 7–10 CONSTRAINT capacity tooling must be READY at the peak HIRE 4 tutors per subject · 6 for test prep FEBRUARY — EXAM-PANIC SEASON the revenue window. it holds or it does not

The stages

Click any stage to open it. Green bullets are shipped.

0

Idea — one platform, not four tools

Mar – Jun 2026 · the thesis that started the build

Shipped
Validated
Acuity caps out around 36 tutors. Scheduling, comms, courses and assessments each lived in a different tool, and the texts and emails leaked around all of them. One branded platform breaks the cap and stops admin work scaling with student count.

What it established

  • The four-tool problem: Acuity, Slack/Discord, Skool, Class Companion
  • Tutors run on their own free Gmail and Fathom — no per-seat cost
  • Credits as the unit of account
  • Target shape: 1,000 students on 50–100 tutors

What it got wrong

  • Defined "done" as a cancelled subscription. Reversed on 23 Aug — done is now a measurable student outcome
  • Put assessments in 2027. They are now the next build
  • Assumed Skool gets replaced. It is being deep-linked instead
1

MVP1 — Tutoring Operations

Jun – Oct 2026 · every session books, holds and settles in the platform

In progress
Sprints 1–3
A student books, a tutor shows, the credit moves, the parent gets a summary — with nobody in Slack arranging it. Acuity dies for tutoring. And a stranger can sign up without an admin creating their account.

Shipped

  • Session lifecycle: credits → book → Google Meet → Fathom summary → refund path
  • Tutor availability with Google Calendar busy-blocking
  • AI session summaries emailed to student, parent and tutor
  • Admin credit management, no-show and refund queue
  • Stripe payments live (1 Sep)
  • CSV student import, 7-day invite window (1 Sep)
  • Tutor can see a student's credit balance from the message thread

Still open

  • Admin: runtime subject management, AP and non-AP SUP-5
  • Admin: role filtering and search SUP-6
  • Admin: user deletion + remove the demo tutor accounts SUP-7
  • Admin: tutor profile builder and editor SUP-8
  • Parent accounts: role, secure student linking, credit purchase SUP-9
  • Comms: one channel per student, tutors auto-added, admin god-mode SUP-10
  • Self-serve — the commercial blocker. No issue on the board yet
  • Bind app.supporteduniversity.com — still does not resolve
Exit criteria
Real students booking with zero Acuity bookings and zero scheduling traffic in Slack · admin scheduling hours down measurably · a person who has never met us can create an account and pay.
3

MVP3 — The Membership Product

Jan – Apr 2027 · revenue that does not need a sales call

Later
Sprints 8–10+
A parent finds us through an affiliate, signs up without talking to anyone, and pays monthly. The platform becomes the destination for the low-ticket funnel instead of a fulfilment tool for high-ticket sales.

The build

  • Membership tiers and entitlement gating
  • Courses: open access and progression-gated
  • Group classes + a community calendar carrying the Zoom links
  • Affiliate and referral tracking in-product
  • Capacity tooling — tutor onboarding at volume, utilisation view
  • Joe's two existing apps brought in: automated session notes, the initial diagnostic prompt

The timing constraint

  • Feb–Apr is exam-panic season and the revenue window that pays for the year
  • Capacity tooling has to be ready before the peak, not built during it
  • Tutor recruiting must front-run January: 4 per subject, 6 for test prep
Exit criteria
A membership tier sells itself end to end with no human in the loop, and an affiliate can be paid from tracked in-product attribution.
4

MVP4 — Cyborg Tutoring at Scale

2027+ · the data starts teaching

Horizon
Gated on MVP2 data
Precision-guided tutoring, running on its own. The AI reads the skill profile and tells the parent and the tutor exactly what to work on next — and the accumulated assessment data becomes an asset in its own right.

The build

  • AI tutor connected to the assessment data, not bolted on
  • AccepteD credits — college coaching moves off Acuity at its own exchange rate
  • Mobile app with two-way parent ↔ student link invites
  • Outcome modelling: this much practice reliably produces this result

Hard gate

  • Clean, diagnosed assessment data has to exist first. A disconnected prototype is allowed before that; an integrated one is not
  • Three credit currencies must already be live and behaving

Sprint plan

Two-week sprints from 15 September. Nov 15 lands inside Sprint 5 — which means four full sprints exist before the core release, and roughly three of them are migration.

Sprint 1Sep 15 – 28
FoundationsRule the credit ledger and the assessment schema. Write the access model. Linear hygiene: owners, estimates, cycles. Jay's first two small PRs inside the main repo.
Sprint 2Sep 29 – Oct 12
Migration 1Audit the old test-prep platform. Map its schema onto the new model. Admin suite: subjects, filters, delete.
Sprint 3Oct 13 – 26
Migration 2Assessment content and data move across. Parent accounts and secure student linking.
Sprint 4Oct 27 – Nov 9
BuildAssessment engine live in the hub. Student skill profile. Comms channels.
Sprint 5Nov 10 – 23
Nov 15 — core releaseParent dashboard v1. TestED credits. Real users inside; start collecting the fringe cases.
Sprint 6Nov 24 – Dec 7
RepairFix what real usage broke. CS at-risk view. No new scope.
Sprint 7Dec 8 – 21
Dec 15 — prepped for studentsSelf-serve registration. Membership v1. Staged release to volume.
Sprint 8Dec 22 – Jan 4
Holiday bufferAssume reduced capacity. Do not plan load-bearing work here.
Sprint 9Jan 5 – 18
Scale prepCourses and group classes. Capacity tooling. Tutor onboarding at volume.
Sprint 10Jan 19 – Feb 1
February — the wallFreeze and load-readiness. Exam-panic season starts and the platform has to hold.

What decides whether this holds

The arithmetic

61 days from 15 Sep to 15 Nov — 4.4 sprints, and the first has not started.
60–70% of effort is migrating the old platform, per Wenbo's own estimate. Against 4.4 sprints, that is ~3 sprints of migration before a new feature ships.
That leaves ~1.4 sprints for six admin items, parent accounts, the comms rebuild, the credit ledger and self-serve signup.
Capacity is one part-time senior plus a junior who has not yet worked in the main repo.
And this team's measured calendar-to-reality gap runs ~4×.

Read: Nov 15 is real for the migration plus the assessment core. The admin suite and the parent account are the two most likely to slide. Cut them out loud now rather than discover it in November.

Open decisions — these block the build

D-1 What does a self-serve signup actually get?Blocks the affiliate motion entirely. Heimler is expected within the month and registration is still closed. Highest commercial cost.
D-2 The credit ledger — three currencies, and what each buysSupportED / TestED / AccepteD at different exchange rates. $50 a credit against $100 an hour still implies a credit is 30 minutes, against an MVP1 built on one-credit-one-session. The schema cannot be written without this.
D-3 The assessment schema, unit + skill, written downJoe has given the spec sentence. It needs to become a document. Everything in MVP2 reads from it.
D-4 Does the old platform's student history migrate?Decides whether legacy rows inherit the new access model or get left behind. It is 60–70% of the effort either way.
D-5 Ownership and IP signature eventOverdue since July. Wenbo is now on a monthly retainer with no assignment on record.

Feature intake

Ideas are welcome. Unpriced ideas are not. Every request runs the same route: grilled against the job it serves, vetted against what already exists, staged, and only then allowed near a sprint — and nothing enters a running sprint without naming what leaves.

Swipe to scroll — or read At a glance above
THE ROUTE — EVERY REQUEST, NO EXCEPTIONS REQUEST IN slack · loom · call · bot GRILL — JTBD who · job · today · reach · cost I-1 A real job with a painful workaround? yes VET dupe? already staged? STAGE IT MVP1–4 or unranked I-2 Into the current sprint? no — say so plainly PARKED recorded, with the reason no — the default BACKLOG staged, unestimated CURRENT SPRINT only if you name what leaves
Intake assistantJTBD intake
Grills against jobs-to-be-done, reads Loom links automatically, and files to Linear Backlog only. Never into a sprint, never estimated — a human still ranks it.

What it asks, and why

Who is it for?Student, parent, tutor, admin, customer success or founder. A request with no role is a preference.
What is the job?"When ___, I want to ___, so I can ___." Solutions get turned back into jobs.
What happens today?Every real request has a painful workaround. No workaround and no pain usually means an idea, not a job.
How often, how many?Once for one admin is not daily for every parent.
What if we never build it?"It would be nice" is a legitimate answer — and it lands unranked.
What leaves to make room?Only asked if someone wants it in the running sprint. Nothing enters free.

The board, live

Read straight from Linear when this page loads — so it cannot drift from the board.

Loading the board…

How new requests get in

Ideas are welcome. Untraded ideas are not — every addition after a sprint is agreed costs something, and the cost gets named before the yes.

The intake rule

  • Every request lands in Linear, not in a DM or a Loom
  • It gets a stage (MVP1–4) and a sprint, or it goes to the backlog — never straight into the current sprint
  • Anything entering the current sprint must name what leaves it
  • Jay runs first-pass intake and vetting; John rules the trade-off; Wenbo estimates
  • Nothing is a deadline until it has survived an estimate pass

Who does what

  • Joe — vision, priority calls, the five open decisions
  • John — product, roadmap, scope protection, trade-off rulings
  • Wenbo — lead engineer, architecture, estimates, PR approval
  • Jay — junior engineer inside the main repo, small incremental PRs, intake first pass
  • Rikki — student operations, assessment content
  • Erika — customer success, training, at-risk students