The Writing Curriculum Review

These two pages are a working slice of the product the report recommends. The diagnostic places a student on a 203-skill knowledge graph in under 24 adaptive questions; the graph then paints that student's personal map: solid skills filled in, the recommended starting frontier ringed, gaps hollow. The best ten minutes of this package is taking the diagnostic yourself and clicking through to your map.

1. Executive summary

Three independent research passes now point to the same conclusion. A documented teardown of the seven nearest writing products, a separate agent-built analysis of The Writing Revolution and the Alpha School approach, and, as of this week, hands-on testing of the live products in a browser all agree: no one has built mastery-based, adaptive, credit-bearing software for writing. The two halves of the product exist separately and are individually good. Quill and NoRedInk own feedback-on-writing; IXL owns adaptive mastery mechanics. Nobody has joined them, nobody runs spaced review over writing skills, and nobody delivers recognized course credit.

Dan's framing survives contact with the evidence. The category leader in adaptive drill (IXL) contains zero sustained composition across the 197 skills of its eighth-grade ELA course, and the best free feedback engine (Quill) never asks a student for more than one sentence at a time. Meanwhile the unsolved technical problem is narrower than it looks: the skill taxonomy for writing is roughly 115 nodes against Math Academy's thousands, and LLMs now grade short convergent responses reliably. What has no prior art is the mastery-and-spacing model for a productive skill, and that is precisely the piece Jason's machinery was built to do for math.

Recommendation: proceed on Dan's plan. Build the fundraising artifact, one month of 9th-grade English as a working module, on six design principles: diagnostic assessment and review to mastery; personalization (the engine, not the teacher, picks each student's next task); spaced repetition and retention of mastered skills; writing exercises embedded in the content students are actually studying, the Hochman principle Alpha dropped and the research supports; authorship enforcement of the kind Quill just demonstrated is technically feasible; and measured transfer to unaided writing. Run curriculum documentation and Cognia accreditation concurrently: a fully documented educational program is itself accreditation evidence.

2. What was tested, and how hard the evidence is

This week I operated the live products directly, as a student, anonymously, with no accounts created, wherever the vendor allows it. That matters because marketing claims in this category routinely outrun the product. Three evidence tiers apply below: hands-on (I ran the actual product), public-deep (full public structure explored; practice is login-walled), and gated (no self-serve access exists at all; ThinkCERCA, Writable, and MY Access! cannot be touched without a sales call, which is itself a datum about the category's go-to-market).

Quill: hands-on, and genuinely impressive at its own game

I completed a full Reading for Evidence activity and stress-tested the feedback bot. A vague answer was rejected with a targeted redirect to the text. A verbatim copy of a source sentence was caught: the system highlighted the exact copied span in the passage, asked for the idea in my own words, and offered three paraphrase strategies. A genuine paraphrase then passed, with feedback naming specifically what it did right. Five attempts per prompt; because/but/so stems straight out of the Hochman playbook. Verified ceiling: one sentence at a time, no visible mastery model, no spaced review. One design flaw observed: a premature submission burns one of the five attempts.

IXL: hands-on, the adaptive benchmark with an empty core

Answering live questions confirmed the mechanics: SmartScore rose ten points on a correct answer and fell only one on a miss, with a mandatory worked explanation before continuing. The machinery is real. The content is the problem: the entire writing strand is selection and revision micro-tasks (pick the thesis, choose the evidence, remove the redundancy). A student can reach 100 without composing a paragraph. Note for us: VPA already has IXL, so the inside diagnostic view is one sign-in away.

NoRedInk: public-deep, the one to study closely

Every skill page in its public library exposes an objective and explicit, linked prerequisites ("Students can distinguish facts from opinions; try Thesis Statements"), then a Teach, Practice, Apply & Assess, Extend pipeline ending in paragraph-level Quick Writes and quizzes. This is the closest structural analog to a skill graph in the category, but it reads as teacher-facing curriculum navigation, not an engine the student experience traverses. The 2026 homepage adds an AI Grading Assistant for argumentative and literary-analysis essays and Originality Insights (paste detection plus time-spent-writing). Claimed footprint: one in two US districts. Everything behind the wall needs a free teacher login.

The Writing Pathway: identity corrected

The earlier teardown treated it as an emerging competitor. It is actually a partnership between Quill.org and Teaching Lab advised by Dr. Steve Graham: a free, open-source scope and sequence plus teacher-facing AI tools (a material generator that embeds writing practice in any subject content, and a formative-assessment tool that analyzes uploaded student work). Their pilot reports 9.3x growth in persuasive writing, effect size 0.64. Strategically this is two things at once: the open-source commoditization threat to any paid sentence-level product, and a free backbone VPA could adopt tomorrow. It is not a student-adaptive system and does not want to be.

3. Scoring the field

My scores, on a rubric built from what Math Academy actually does in math, translated to writing. Ten capabilities, zero to three points each: 0 = absent, 1 = partial or shallow, 2 = real but constrained (typically teacher-driven), 3 = genuinely delivered. Scores are evidence-weighted; a gated product cannot earn a 3 on a claim I could not verify. Khanmigo Writing Coach is added to the original seven because it is free, essay-level, and already inside our Khan World School ecosystem.

A purpose-built "Math Academy for writing" targets 30/30. The best incumbent scores 15, and, the decisive point, every product scores zero on personalized spaced review (the highlighted row). The retention engine, which Jason would tell us is where the learning actually happens, does not exist anywhere in this category.

Note on the earlier scores: the documented teardown ranked NoRedInk 19/32, IXL 17/32, ThinkCERCA 16/32 on a 25-capability rubric. The ordering is stable across both passes (NoRedInk first, IXL and Quill close behind, assessment-only and teacher-tool products trailing), which is itself a useful robustness check.

4. Incorporating the TWR analysis: what held up

The agent-built TWR summary (circulated August 19; the full document is Section 4 of this package) made five load-bearing claims. Four held up under this week's hands-on work; one needs sharpening.

• Held: TWR is a teacher-training method, not a curriculum: $825 per adult, no per-student license, sentence-up progression, and a hard requirement that exercises be embedded in course content.
• Held: Alpha's AlphaWrite kept the sentence drills and dropped the content-embedding principle. Students practice on Fortnite and Harry Potter instead of coursework, and Alpha's own parents call the app weak. Embedding is the piece to keep.
• Held: no knowledge graph, no automated assignment, no spaced repetition anywhere in the category. Hands-on testing found the same holes the desk research predicted.
• Held: two of the three hard parts are solved: LLM grading of short convergent responses (Quill's live feedback bot is an existence proof, catching plagiarism and judging paraphrase quality in real time), and a tractable taxonomy of roughly 115 skills.
• Sharpened: the whitespace is not "no one gives AI feedback on essays"; NoRedInk, Writable, and MY Access! all sell that today. The whitespace is the loop: diagnostic, skill profile, automated next-best task, embedded practice, authentic composition, verified mastery, spaced return. And one more piece of prior art matters: Quill's own published RCT (n=260 ninth graders) found sentence-combining gains that did not transfer to original paragraph writing. Sentence drill improves sentence drill. Any credible product must build composition into the mastery loop itself and measure delayed, unaided transfer. This is the single most important design constraint in the whole space.

5. Design principles for the VPA build

Pulling the threads together (Dan's email, the TWR analysis, and the hands-on evidence), the sample module should demonstrate six things. The first three are the Math Academy spine; the last three are what the writing domain adds.

1. Assess, then review to mastery, then measure progress. Diagnostic placement on entry, mastery gates on every skill, and XP-style progress that students and parents can read at a glance.
2. Personalization. The engine, not the teacher, selects each student's next task from the prerequisite graph, at the edge of their ability. This is what separates a curriculum from an adaptive system, and it is exactly what every incumbent lacks: even NoRedInk, which encodes prerequisites, leaves the sequencing to the teacher.
3. Spaced repetition and retention. The genuinely novel piece; every product in the field scores zero here. A mastered skill (appositives, evidence integration, counterclaim) resurfaces on a decay schedule inside later, harder compositions, so review compounds instead of repeating. Spacing applied to a productive skill rather than a recall item has no prior art; it is our hardest problem and our deepest moat.
4. Content-embedded exercises. Every drill built from the texts and topics students are studying that month: the Hochman principle, the piece Alpha dropped, and the natural fit with our integrated academic-athletic model. The Writing Pathway's free generator proves the authoring cost of this is now low.
5. Authorship enforcement. Quill demonstrates span-level copy detection with paraphrase coaching; NoRedInk tracks paste events and time-on-task. In the AI era, a writing product that cannot evidence authorship cannot carry credit. Ours must log the drafting process, not just the artifact.
6. Transfer measurement. Blind-scored, unaided writing at baseline, module end, and a delayed check, because the documented failure mode of the entire category is skills that live only inside the app.

6. Accreditation: concurrent, and partly the same work

Cognia accredits institutions, not software, and one of its core requirements is a fully documented educational program. Writing the curriculum down is literally accreditation evidence, so the two tracks reinforce each other rather than compete. The ASU Prep experience is the cautionary tale for the credit question: the graded-Math-Academy impasse appears to run through course-approval shells and teacher-of-record requirements (StudyForge is a named third-party beneficiary in the MSA), not economics, which means money alone does not fix it. The writing product should be designed from day one to produce the documentation an accreditor and a course-approval reviewer need: scope and sequence, mastery evidence per student, assessment integrity, and a defined teacher-of-record role. Near-term credit path: VPA's own Cognia candidacy plus a partner accredited school or umbrella; NCAA course approval follows accreditation, not the other way around.

7. People and deals on the table

• Jason Roberts (Math Academy), Tuesday. Two distinct asks worth separating: (a) the wholesale/enrollment deal and the ASU-P graded-math elements Dan raised; (b) the deeper conversation, whether Jason would license, advise on, or co-develop the mastery-and-spacing engine applied to writing. His system's diagnostic, knowledge-graph, and XP machinery is exactly the unsolved 20% of this product. Suggested Tuesday agenda: wholesale terms; what ASU-P needs from MA to unlock a letter grade; and whether "Math Academy for writing" is something he wants to be inside of or is happy to see us build alongside.
• Kyle Cureau (Scrivium / Crone AI). Scrivium is a mobile app teaching history through three-week "apprenticeships" with primary sources, daily 5-10 minute lessons, and a student commonplace book. It is evidence that the adjacent lane Dan flagged (history, behind writing) is already being explored by a builder in our own valley; Kyle is a Vail Symposium speaker. Worth a conversation on two fronts: his read on content-embedded skill-building at the app layer, and whether Crone's authoring pipeline is relevant to our module build.
• Michael Resnick. Northeastern grad leaving Alterra for a masters in education, the profile that worked in summer programming (young guides out-connecting the teachers). Curriculum development on the sample module is a real, bounded first project: drafting embedded exercises for the 9th-grade English month under a defined scope and sequence. If the work excites him, it also opens the broader family conversation Dan raised with Eric.
• Free assets to adopt rather than build. The Writing Pathway's open-source scope and sequence (Graham-advised) as the taxonomy starting point; Quill's open-source codebase (GitHub: empirical-org) as a reference implementation of writing feedback; Khanmigo Writing Coach as the interim essay-revision layer inside Khan World School while our module is built.

8. Recommendation and next steps

Build the month-of-9th-grade-English module as the fundraising artifact, and treat it as a concierge pilot rather than software: a human-operated skill graph, LLM-graded exercises embedded in that month's texts, mastery gates, one spaced-return cycle, and blind-scored before/after/delayed writing samples. That artifact simultaneously (a) proves the loop for investors, (b) generates the documented-program evidence Cognia wants, (c) gives ASU-P and any partner school something concrete to evaluate for credit, and (d) tells us cheaply whether the transfer problem yields. Decision gates stay as previously framed: stop if an incumbent ships the full loop, or if gains vanish when students write unaided.

Immediate steps: (1) Tuesday: put writing on the Jason agenda alongside wholesale; (2) sign into NoRedInk (free teacher account) and our own IXL so the inside teardown can complete; (3) a call with Kyle Cureau; (4) an exploratory conversation with Michael Resnick scoped to the sample module; (5) draft the module's scope and sequence from the Writing Pathway / TWR spine, embedded in the chosen 9th-grade texts.

9. Conclusions: the opportunity

The market is crowded and the whitespace is real, which is the best possible combination. Crowded, because forty years of products (from IntelliMetric's essay scoring to this year's AI grading assistants) prove schools pay for writing instruction and assessment. Real, because after three research passes and a week of hands-on testing, the complete learning loop still does not exist anywhere: not the personalization, not the spaced retention, not the credit. Every incumbent stopped at the piece its origin story made easy: Quill at the sentence, IXL at the multiple-choice item, NoRedInk at the teacher dashboard, the essay scorers at the rubric. Nobody has done for writing what Math Academy did for math, and the reason is not that nobody noticed; it is that the retention-and-transfer model for a productive skill is hard, and the incumbents' business models do not force them to solve it. Ours does, because our students need the results, not the seat time.

VPA's specific advantages are unusual for a school this size: a live student body to pilot on; an existing Math Academy deployment that models the target mechanics daily; direct access to Jason Roberts for the engine, to a local app builder for the authoring layer, and to motivated curriculum talent; free, open-source raw material (the Writing Pathway sequence, Quill's codebase) covering most of what should not be built from scratch; and an accreditation track that turns curriculum documentation into a dual-purpose asset. The downside is bounded: the concierge pilot costs a month of teaching effort and produces the fundraising artifact even in partial success. The upside, as Dan put it, is larger than everything else we are doing: a credit-bearing, mastery-based writing curriculum would be the first of its kind, the beginning of the full-stack academic model (writing first, then history and reading), and the difference between VPA assembling other people's apps and VPA building something. The evidence says the lane is open. It will not stay open indefinitely; NoRedInk has the pieces closest at hand, and the Writing Pathway is commoditizing the bottom of the stack. Move while the loop is still unclaimed.

1. What this is

A machine-readable prerequisite graph of 120 writing skills spanning K-12, with 178 hard prerequisite edges and 63 encompassing (implicit-credit) edges, designed to power a Math Academy-style mastery engine for writing. It ships as three coordinated artifacts: the JSON data file, an interactive visual map, and this specification, which records the design rules and the evidence behind them. The graph was synthesized by research agents from the primary sources below and is a v0.1 hypothesis to be refined against student data, not a finished truth.

Sources for the skill inventory: The Writing Revolution (Hochman) strategy sequence; The Writing Pathway open-source scope and sequence (Quill.org and Teaching Lab, advised by Dr. Steve Graham); Quill Connect's seventeen sentence-combining concept areas; the Common Core Writing and Language standards progression with its grade-by-grade deltas; SRSD (self-regulated strategy development) planning strategies; and Berninger's transcription-automaticity research. Sources for the mechanics: Math Academy's published methodology (knowledge graph, diagnostics, mastery gates, FIRe spaced repetition with repetition compression), the spacing and interleaving literature (Cepeda, Rohrer, Soderstrom and Bjork), mastery-learning meta-analyses (Bloom, Kulik), and the writing-instruction effect-size literature (Graham and Perin's Writing Next; SRSD meta-analyses; Kellogg's working-memory model).

2. Shape of the graph

By grade band of introduction: K-2 has 18 nodes, grades 3-5 have 50, grades 6-8 have 41, grades 9-10 have 9, grades 11-12 have 2. The taper is intentional and mirrors the research: nearly everything structurally new in writing happens by grade 8; high school deepens, combines and matures existing skills rather than introducing new machinery.

Each node carries: a definition, grade band, strand, node class, hard and soft prerequisites, encompassing edges with weights, a mastery criterion, an exercise-type ladder, a spacing class, sources, and an embedding hook tying its exercises to the content students are currently studying (the Hochman principle).

3. The four node classes

• Transcription-convention (TC). Behaves like math facts: drill to automaticity, fast-stabilizing review intervals. Gate: three consecutive sessions at 90 percent or better, spaced across days.
• Construction (CN). The generative trunk: one composable move per node (combine with an appositive; write a forecasting topic sentence). Five exercise types per node on a recognition-to-generation ladder: identify, repair, constrained generation, cued-in-context, free generation. Only free generation counts toward mastery.
• Process-strategy (PS). SRSD planning and self-regulation routines, taught by model-support-fade. These carry the largest effect sizes in the writing literature (strategy instruction 0.82; SRSD at or above 1.0) and are first-class nodes with their own reviews, a genuine extension beyond Math Academy, which has no analog.
• Composition (CP). Milestone nodes (paragraph, essay, research piece). Mastery requires two rubric-pass compositions on distinct contents plus a delayed, proctored cold prompt. Compositions are also the review medium for everything below them.

4. Edge semantics and the implicit-credit rule

A hard prerequisite edge means: failure on the downstream node cannot be interpreted without knowing the student's state on the upstream node, so assignment is blocked until the prerequisite is mastered. Soft prerequisites warn and trigger parallel remediation without blocking, cloning Math Academy's practice of fixing foundational gaps alongside frontier work rather than sending students backward.

Encompassing edges carry Math Academy's most valuable trick, adapted for writing. In their system, doing advanced work sends fractional implicit review credit down to component skills. Writing needs one modification, because an essay, unlike a multiplication, does not necessarily contain an appositive: implicit credit flows only when the component skill is actually detected and judged adequate in the graded artifact, at the edge's weight. A due skill that keeps failing to appear where it is warranted is treated as decaying, and the scheduler responds with an explicit or cued review. This detected-credit rule is the single most important adaptation in the whole design.

5. Mastery, spacing and transfer

Mastery of a generative node: three rubric-pass productions across at least two content domains and two task frames, at least one of them uncued inside a composition, and no more than one qualifying success per day, which forces spacing into the criterion itself. The bar is deliberately stricter than Math Academy's two-in-a-row because a rubric pass is a noisier signal than a correct math answer. Convergent items are auto-graded; generative items are LLM-graded criterion by criterion against three-to-four-point micro-rubrics; milestone compositions and all cold prompts get human sign-off in proctored, keystroke-logged sessions, which is also the authorship evidence an accreditor will want.

Spacing: per-student, per-node memory decay in the FIRe style, with a base interval ladder of 1, 3, 7, 21, 60 and 180 days. An explicit drill counts as a full review rep; a detected use inside a composition counts at the edge weight; a cued addition to the student's own draft counts at 0.7. The scheduler's preferred move is repetition compression: when several skills fall due, it assigns one short composition engineered to elicit them all rather than separate drills. One fifteen-minute paragraph can legitimately clear six to ten reviews, and this is also the transfer mechanism, because the review economy itself forces composition several times a week.

Transfer safeguards, designed against the documented failure mode (Quill's own RCT: sentence-combining gains with no transfer to paragraph quality): no terminal drill nodes, since every construction node is encompassed by at least one composition node and cannot reach mastered status without an uncued in-composition success; strategy nodes sit above construction nodes; and every six weeks a proctored cold prompt is scored holistically and analytically, with authority to demote any mastered node that fails to show up, overriding the memory model. If node masteries climb while cold-prompt quality does not, the graph is lying, and the cold-prompt trend is the truth.

6. Honesty clause and next steps

The interval ladder, the 0.7 cue discount, the mastery thresholds and the edge weights are defensible priors drawn from the nearest evidence, not findings. There is no validated knowledge graph for writing anywhere; that is precisely the whitespace. Instrument everything from day one, compare delayed performance against predicted memory, and re-fit parameters within two semesters, pooling by node class given VPA's small cohort.

Next steps: (1) review the graph with Jason Roberts against Math Academy's mechanics, especially the detected-credit rule and the compression scheduler; (2) select the 9th-grade English month and mark its module slice through the graph (roughly the grades 6-8 and 9-10 frontier of strands 2 through 6); (3) author micro-rubrics and embedded exercise templates for that slice; (4) run the concierge pilot with a human executing the scheduler's rules; (5) blind-score baseline, end and delayed samples.

Companion artifacts: the graph data (JSON), the interactive map at elevateedwards.com/graph, and the competitor teardown in Section 3 of this package.

Context. VPA is evaluating whether to build a "Math Academy for writing": a mastery-based, prerequisite-aware writing curriculum app that could carry accredited course credit. An earlier documented teardown scored seven competitors from public materials. This document adds the hands-on phase: live product testing in Chrome, done anonymously with no accounts created. Evidence level is labeled per competitor.

• HANDS-ON  I operated the live product as a student would.
• PUBLIC-DEEP  Full public curriculum and product structure explored; the practice itself is login-walled.
• GATED  No self-serve access exists; a sales demo is required. Claims rest on marketing and documentation only.

1. Quill.org hands-on

Played the full "Reading for Evidence" activity loop (Should Schools Have Grade Requirements for Student Athletes?, grades 8-12) anonymously; no account needed. What the loop actually does, verified live:

1. Read and highlight: the student must scroll the entire text (completion is gated on reading) and highlight exactly two sentences matching an evidence goal, with both checkboxes tracked live.
2. Write: sentence-stem expansion ("Critics have opposed No Pass No Play laws because..."), five attempts maximum, AI feedback per attempt.
3. Feedback quality, tested adversarially: a vague answer ("because they are bad") was rejected with a targeted redirect ("Why do some critics disagree...? Check that your response only uses information from the text."). A verbatim copy from the passage was caught: the system highlighted the exact copied span in the source text, said "You have the right idea! Now rewrite the idea using your own words," and opened a hint panel teaching three paraphrase strategies. This is real plagiarism-aware feedback, not string matching theater. A genuine paraphrase then passed, with specific positive feedback naming what was done right. One flaw: a premature submission burns an attempt.
4. Sequence per activity: because, but, so (Cause / Contrast / Consequence, Hochman-style).

Verified limits: everything is one sentence at a time. No paragraph or essay composition, no cross-activity mastery model visible to the student, no spaced review surfaced in anonymous play. The feedback bot is labeled Beta and self-disclaims accuracy.

Verdict: the best free sentence-level feedback loop tested, strongly aligned with TWR's sentence spine. Not a curriculum, not a mastery system.

2. NoRedInk public-deep

Explored the full public Assignment Library. Verified structure: collections for Skill Building, Standards & Tests, Daily Writing, Writing Genres, Text-Based Writing, and a Grading Assistant (AI scores and comments on student essays). Skill Building has eight categories (Parts of an Essay; Evidence, Citations and Plagiarism; Clarity and Style; sentence and grammar strands), filterable by grade 2-12.

Key find: each skill page exposes an objective ("Students can use strong evidence to support a claim") and explicit prerequisites with links ("Students can distinguish facts from opinions. Try Thesis Statements."), plus a pipeline per skill: Teach (tutorial), Practice (Core and Scaffolding, login-locked), Apply & Assess (Quick Writes, a guided short-response paragraph task, a quiz), Extend. So NoRedInk does encode prerequisite relationships, but as teacher-facing curriculum navigation, not as an automated graph the engine traverses for the student.

From the current homepage (2026): AI Grading Assistant on argumentative and literary-analysis essays ("50% less grading time, 5x more feedback," more genres rolling out); Originality Insights (tracks time spent writing and flags pasted text); a Language Support Suite for English learners; 1,000+ skills and 10+ guided writing genres in Premium; new Read-Write-Reason integrated reading and writing activities; claimed use in one of two US districts. Grades 3-12. Every practice item requires a login; free teacher accounts exist, so an inside teardown is one sign-up away.

3. IXL ELA hands-on

Answered live questions in the grade 8 skill "Choose evidence to support a claim" without an account. Verified mechanics: instant next-question serving; SmartScore started at 0, rose 10 for a correct answer, fell 1 on a miss (asymmetric scoring confirmed); a wrong answer triggers a full explanation screen with the item re-shown and reviewed before continuing. Question format: two-option multiple choice.

Verified scope: grade 8 ELA alone is 197 skills and 87 videos across four strands (reading strategies, writing strategies, vocabulary, grammar and mechanics), with videos per skill and skill plans aligned to textbooks and state tests. The entire writing strand is selection and revision micro-tasks: identify the thesis, choose the evidence, remove the redundant phrase, rewrite in active voice. Zero sustained composition anywhere.

Relevant to us: VPA already licenses IXL and students are enrolled, so the inside view (Real-Time Diagnostic, the Recommendations wall) is one login away.

4. The Writing Pathway public-deep

Correction to the prior teardown: The Writing Pathway is a Quill.org and Teaching Lab partnership, advised by Dr. Steve Graham. It is a free, open-source scope and sequence plus a teacher-facing AI assistant, explicitly not a student-adaptive app. The Material Generator lets a teacher pick a writing skill and topic and produces curriculum-tied practice in any subject; the Formative Assessment Tool analyzes uploaded student work for class-wide trends and recommended next instruction. Pilot evidence they cite: middle-schoolers made 9.3x growth in persuasive writing, effect size 0.64.

Strategic read: this is the open-source commoditization threat to any paid sentence-level product, and simultaneously a free asset VPA could adopt as its scope-and-sequence backbone. It does not do student-facing adaptivity, mastery tracking, spaced review, or credit.

5-7. ThinkCERCA, Writable (HMH), MY Access! (Vantage) gated

Checked all three for self-serve access on August 24, 2026: none offers any trial, sandbox, or sample a person can operate without a sales call. A true hands-on teardown of these requires booking demos.

• ThinkCERCA: Core ELAR Suite 6-12, Writing Across Disciplines 3-12, supplemental writing; "patent-pending AI-powered automated feedback" (launched March 2024); free downloadable lessons only.
• Writable: rubric-aligned AI comments and scores, revision cycles, proficiency grouping, customizable rubrics; the teacher assigns everything.
• MY Access!: IntelliMetric automated essay scoring plus MY Tutor "adaptive, scaffolded revision directives"; login or sales form only.

Capability table, hands-on evidence tier

Scale: ✓ verified live · ~ exists but shallow or teacher-driven (verified or documented) · ✗ no evidence it exists · 🔒 unverifiable without an account or demo.

Unchanged conclusion, now with live evidence: nobody does prerequisite-aware automated sequencing plus spaced review plus mastery verification over authentic composition. The two halves exist separately (Quill and NoRedInk own feedback-on-writing, IXL owns adaptive-drill mechanics) and no one has joined them. Credit-bearing status remains unclaimed by all seven.

Also now confirmed: the whitespace is narrower than "no one gives AI essay feedback" (three products claim it) and wider than the earlier document assumed on one axis: even the best free tools cap at one sentence of student writing at a time, and the strongest adaptive engine (IXL) contains literally zero composition.

What login access would unlock next

1. A free NoRedInk teacher account: adaptive practice behavior, Quick Write feedback quality, quiz and mastery mechanics, the Grading Assistant.
2. VPA's existing IXL account: the Real-Time Diagnostic, the Recommendations wall, and the SmartScore-to-mastery flow across the writing strand.
3. A free Quill teacher and student account: the Quill Diagnostic to recommended-practice loop, dashboards, and whether any review or retention mechanism exists.
4. ThinkCERCA, Writable and MY Access! require booked sales demos; decide whether they are worth it after 1-3.

Tested August 24, 2026. No accounts created; no logins performed.

The headline finding, before anything else

Alpha School does not use The Writing Revolution. They built an in-house clone of it called AlphaWrite, and they stripped out the one principle the method's own authors consider load-bearing.

This is not speculation. Natalie Wexler, co-author of The Writing Revolution, investigated AlphaWrite and published her findings on August 17, 2026: "AlphaWrite is a copy of the Hochman Method... [but] the app's creators have overlooked a crucial component of the Hochman Method: it's designed to be embedded in curriculum content."

AlphaWrite's student topic menu is Fortnite, Harry Potter, Social Media, Video Games, Basketball. Hochman's method requires students to write about the history, science, and literature they are currently studying, because the writing is the retrieval-practice mechanism that makes the content stick. Alpha kept the sentence drills and threw away the reason they work. Wexler also quotes an Alpha parent calling AlphaWrite "weak" and in need of redesign. AlphaWrite is proprietary and not licensable. There is no relationship, license, or endorsement between Alpha / 2 Hour Learning and The Writing Revolution, Inc.

What this means for VPA: the thing we thought we were copying from Alpha isn't what Alpha is doing. That's good news; Alpha's version is the compromised one. We can do the real thing more cheaply than they built the knockoff.

1. What TWR actually is (and isn't)

TWR is a $6-7M/yr New York 501(c)(3) that sells teacher professional development. It is emphatically not a curriculum. Their own language: "The Hochman Method is not a separate writing curriculum but rather an approach designed to be adapted to and embedded in the content being taught in any subject area and at any grade level."

That single sentence is the whole strategic picture. TWR sells a method for teaching writing inside whatever you're already teaching. It has no scope and sequence of its own, no texts, no assignments. It attaches to a content-rich curriculum, which for VPA means ASU Prep's course shells, Levitt Lab, and whatever history and science the students are doing.

The strategy sequence (grades 3-12)

Sentences: fragments, scrambled sentences, sentence expansion, sentence types, developing questions, because/but/so, subordinating conjunctions, transitions, appositives, sentence combining. Note-taking: underlining key words, converting notes to and from sentences, taking notes. Paragraphs: the Single-Paragraph Outline (SPO) and its scaffolded variants. Revision: vary vocabulary, improve topic and closing sentences, elaborate. Summaries: Summary SPO, summary sentence. Composition: Pre-Transition Outline, Transition Outline, Multi-Paragraph Outline (MPO), introductions (general to specific to thesis), conclusions.

Roughly 115 distinct activity types across the full sequence, entirely expository and analytic. There is no narrative or creative strand. If VPA wants students who can write a college essay with a voice, TWR is not that; it's the layer underneath.

Current pricing (verified on TWR's site)

Group discounts at 10+ and 25+; one free administrator seat per organization. All training memberships include 12 months of MyTWR Tools. For a school VPA's size, the honest cost is roughly $825-1,250 per adult, once. That is the entire barrier to entry: no per-student license, no annual curriculum fee, no minimum cohort.

TWR now has a digital product, but it's teacher-facing only. MyTWR Tools launched July 2025; "Judy," an AI activity generator built with NIIT, launched November 2025. Judy generates Hochman-approved activities, worksheets, teacher guides, and anticipated student responses. TWR is explicit: "a collaborator for educators... not student-facing software." They built the content-generation half of the product and deliberately stopped short of the student half. That is the gap in the market.

2. Who actually uses it

Verified state-level: Louisiana is the flagship (TWR strategies are embedded in the state's revised ELA Guidebooks, 2022). Arkansas runs free statewide TWR cohorts, funded through the 2026-27 school year. Ohio has a state-funded four-district pilot (Athens, St. Marys, Campbell, Ironton). In England, Judith Hochman is formally acknowledged as a contributor to the DfE Writing Framework (July 2025), whose core claim is that "the best way to teach pupils to write is by teaching them to master sentences"; the framework endorses the principle, not the program. Australia is the strongest non-US market, with Ballarat Clarendon College (Victoria) as the flagship: its cohort topped state results in four of the last five years.

Verified district-level (US): LAUSD, Newark, White Plains, Rockville Centre, four NYC districts, Pueblo District 60 (Colorado), Charles County MD, Wayne County KY, and roughly twenty others.

Corrections to common assumptions: NYC has not adopted TWR citywide (NYC Reads mandates reading curricula; TWR's NYC presence is four districts buying PD). The elite charter networks are unverified: no sourced evidence for Success Academy, KIPP, Uncommon, Achievement First, IDEA, BASIS, or Great Hearts; the only Uncommon connection is that Doug Lemov wrote the book's foreword. Hillsdale-affiliated classical schools use IEW, not Hochman.

Read on the pattern: TWR's real customer is the science-of-reading public district, not the innovator-school sector. Nobody in VPA's peer group (Alpha, Khan World School, the microschool world) is running actual TWR. Alpha cloned it. That's the market.

3. Does it work? An honest evidence grade

As a packaged program: D+ / Emerging

No What Works Clearinghouse review exists. ERIC returns exactly one result, a practitioner essay. Zero RCTs. The entire direct evidence base is one TWR-commissioned quasi-experiment: NORC at the University of Chicago, April 2025, 957 students in grades 4-5 in Monroe City, Louisiana. Result: effect size 0.25 on ELA writing performance, p = 0.039, significant on one of four outcomes with no multiplicity adjustment; the other three were positive but null. The comparison group was not a control group: it was lower-implementation TWR schools, so the study measures a dose-response gradient inside TWR adopters, not TWR against anything else. A 2020 Metis Associates evaluation of 16 NYC schools found mixed-to-null Regents results; the report is not public, and TWR's marketing does not surface it.

The most telling quote is from TWR's own co-author, Natalie Wexler in Forbes: "We don't yet have studies of the impact of an approach that, like The Writing Revolution, starts with knowledge-building sentence-level activities."

As a set of instructional moves: B / B+

The components are individually well-supported, mostly by Steve Graham's meta-analyses:

So: not teaching grammar in isolation is solidly right. Writing embedded in content is the best-evidenced thing TWR does. Sentence combining is real but rests on a thin, old base, and it did not clear the bar for inclusion in the strictest recent meta-analysis (grades 6-12, 22,838 students).

The theory of action: C, unfalsified rather than supported

The specific claim that starting with sentences and building up beats starting with whole texts has never been tested against a fair counterfactual. Two facts cut against the framing TWR uses to sell it: the IES elementary practice guide's only STRONG recommendation is teaching the writing process, the thing TWR positions itself against; and the strictest 2024/25 meta-analysis puts process writing at 0.75, second only to SRSD. Hillocks' famous finding is routinely misread: he found unstructured free writing weak and structured problem-solving strong. TWR is on the right side of that, but "process writing doesn't work" is not what the data say.

The one intervention with better evidence than TWR: SRSD

Self-Regulated Strategy Development posts 0.84-1.17 across every major meta-analysis, the highest numbers in writing research, and survived the strict 2024/25 screen that eliminated sentence combining. Its active ingredient is self-regulation (goal-setting, self-monitoring, self-talk): generic strategy instruction is 0.62, SRSD is 1.14, and that delta is the self-regulation layer. TWR is silent on self-regulation; running TWR alone leaves the single largest documented effect in writing research on the table. But they are complementary, not competing: TWR builds sentence-level capacity that SRSD assumes, and SRSD supplies whole-composition genre strategies and metacognition that TWR doesn't touch. Given VPA's population of student-athletes managing training loads, race calendars, and asynchronous coursework, the self-regulation layer may matter more here than at a conventional school.

4. Is there a Math Academy for writing?

No. The category is empty. That is the finding. Nothing on the market has what makes Math Academy Math Academy: diagnostic placement, a knowledge graph with real prerequisites, automated task selection, spaced repetition, and progression with no adult in the loop.

Quill gets ~60% of the way there, and it's free

Quill is the single most important find in this research: a 501(c)(3), 12M students served, 42,000 schools, roughly 10% of US schools. All activities are free forever for unlimited students (Premium, $115/yr per teacher, buys reporting, not content), and homeschools and microschools are explicitly eligible. Quill Connect is a sentence-combining engine with 500+ exercises across 17 skill areas including appositive phrases, participial phrases, relative clauses, and parallel structure. Quill Reading for Evidence (grades 8-12) has students read nonfiction and complete because/but/so stems with up to four AI-coached revisions. Free College Board Pre-AP English packs include sentence-expansion and combining activities keyed to real texts.

The lineage: Quill's own About page credits Peg Tyre's 2012 Atlantic article "The Writing Revolution," which profiled Judith Hochman at New Dorp High School, as its founding inspiration. Quill and TWR descend from the same source event; convergent, not derived.

Quill's evidence is better than TWR's, and more honest about its limits. Two RCTs with College Board: n=82 showed d > 0.80 on sentence combining, sustained at two months, with the caveat that the outcome measure mirrored the intervention; and n=260 ninth graders showed sentence combining improved (TOWL d = 0.30) but no significant effect on original paragraph writing or revision. That second result is the most important number in this entire report: sentence drill improves sentence drill, and transfer to real writing is the documented failure mode. Quill published it about their own product.

What Quill lacks: no knowledge graph, no auto-assignment (a human pushes the diagnostic's recommended packs, ~10 min/week), no spaced repetition, no full 3-12 placement spine.

The rest of the landscape

NoRedInk has the best mastery mechanics in the category (four mastery levels, two-correct-to-advance, teacher alert after seven wrong) but diagnostics don't auto-generate paths. Khanmigo Writing Coach is free for US teachers and students, coaching outline, draft, and revision at the essay level with a teacher dashboard; it does not grade, and it is already in-ecosystem via Khan World School, covering exactly the transfer gap Quill's own RCT exposed. Writable is $12/student/yr of AI feedback on assignments. Lexia and Achieve3000 are reading products. Grammarly is anti-pedagogical here: it corrects for the student. Brisk, MagicSchool, Diffit, SchoolAI, and CoGrader are teacher productivity tools. Among 2024-26 AI-native entrants (YC cohorts and funding trackers searched), no student-facing, sentence-level, adaptive writing tutor exists. The lane is open.

5. What I'd actually do

Recommended stack: roughly $1,000-1,500 total, year one

Three things that materially improve the bet

1. Protect the content embedding. It is the best-evidenced element (0.40 to comprehension) and the exact thing Alpha discarded. At VPA this means students write about their physics, their history, their Levitt Lab projects, never about Fortnite. Writing is how the content gets consolidated, not a separate subject.
2. Protect sentence combining specifically. In the NYC evaluation, 81% of teachers regularly taught the easy because/but/so activity but only 22% taught sentence combining, the one component with independent research support. Schools adopt the front end and drop the part that works. Quill automates this one, which is a good reason to lean on Quill for it.
3. Instrument for transfer from day one. Quill's own RCT found sentence gains did not transfer to independent writing. Collect extended writing samples on a common rubric from the first month. With a cohort this small, our own data will be better evidence than anything a vendor can hand us.

On building our own

The lane is genuinely open, and more tractable than it looks. Auto-grading is solved for this task type: because/but/so completions, sentence expansion, and appositive insertion are convergent, narrowly-rubricked tasks, exactly the regime where LLM graders hit QWK 0.8+, versus holistic essays where human-AI agreement is still QWK 0.42-0.59. The taxonomy is off the shelf: ~115 activity types published by TWR, so a writing knowledge graph is low hundreds of nodes versus Math Academy's thousands. The corpora are public (PERSUADE 2.0, ELLIPSE, the Kaggle AES 2.0 solutions), and Quill's own NLP tools and full platform are open source on GitHub. Read the Project Evident case study on Quill's AI build first: it documents their 95-99% correct-feedback bar, 300+ hand-graded responses per prompt, and a 300-teacher review council. It is the closest thing to a build manual that exists.

The genuinely unsolved part is mastery modeling for writing. No prior art exists for knowledge-graph prerequisites plus spaced repetition on writing skills. Three real obstacles: writing skills aren't cleanly prerequisite (appositives don't gate relative clauses the way factoring gates quadratics); "mastery" is fuzzy (correct in a scaffolded stem is not produced unprompted); and spacing intervals were tuned on recall, not generation. Scope it as 12-18 months, not a semester. But if VPA wants a defensible claim to being a next-generation academy rather than an assembler of other people's apps, this is the one open lane found.

6. Open questions

1. Who teaches writing at VPA? The TWR cost model is per-adult, which is cheap at this scale, but the method is feedback-heavy and NORC's effect only showed up for full-implementation teachers. Fidelity is the binding constraint, not price.
2. Does ASU Prep's English leave room for this? TWR attaches to content we control. If English is fully outsourced to ASU Prep shells, the natural home for embedded writing is physics, Levitt Lab, and history instead, which is arguably more true to Hochman anyway.
3. Is Alpha's AlphaWrite worth a conversation? Proprietary, not licensed out, and its own parents call it weak. Probably not. But if there is a channel to 2 Hour Learning, the interesting question is whether they'd fix the content-embedding gap for a partner school.
4. SRSD or not? It has the best evidence in the field and costs nothing to adopt from the research literature. Worth a separate look.