For associations · societies · firms

Mastery you can trace to the syllabus.

Pinnacle runs on a knowledge graph of 386 skill nodes across seven certifications — each one verified against the awarding body's own published outline — with Bayesian mastery tracking on top and an AI voice tutor whose every session traces back to named learning outcomes. This page is the engine, opened up for the people who steward those syllabi.

No account needed, no model spend per visitor — the same 25-question diagnostic your members would take.

386skill nodes, verified against official outlines
7certifications: CFA L1–L3 · FRM I–II · CPA · ACCA
100%of tutoring sessions traceable to learning outcomes
4,596automated tests behind the platform, run on every change
01 · The knowledge graph

Verified against the documents your candidates are examined on.

Most prep providers write a table of contents and call it coverage. We model each certification as a graph of skill nodes with prerequisite edges — and in August 2026 we verified every node's name and scope against the awarding body's own published curriculum documents, reading the primary sources in full rather than summaries of them. Nodes teaching retired material were retired. Modules with no node were added. Descriptions that had drifted from their outcomes were rewritten from the outline's own words.

The result is a curriculum spine that says, for every concept, exactly which published learning outcome it serves — and refuses to teach anything that serves none.

CertificationVerified againstNodes
CFA® Level ICFA Institute 2026 topic outlines & learning outcome statements104
CFA® Level IICFA Institute 2026 topic outlines, incl. the 2027 Ethics restructure61
CFA® Level IIIAll three 2026 pathway outlines — Portfolio Management, Private Wealth, Private Markets52
FRM® Part IGARP 2026 FRM Study Guide — all 62 Part I readings mapped32
FRM® Part IIGARP 2026 Learning Objectives document, incl. all six live Current Issues topics54
US CPAUniform CPA Examination Blueprints, January 2026 edition33
ACCAPaper-by-paper 2026/27 syllabus & study guides, incl. the 2027 restructure flagged on every node50
Verification pass completed August 2026. Every verdict carries quoted evidence and a source reference in the graph itself — not in a marketing document.
02 · The engine

Mastery tracking that refuses confidence it hasn't earned.

On top of the graph sits a knowledge-state engine: every answer a candidate gives updates a Bayesian estimate of mastery on the nodes that question actually tests — and only those nodes. Question-to-node mappings are curated and certification-scoped, so practice on one exam never contaminates the estimates of another.

P(mastery | evidence)

Bayesian, per node

Item response theory and Bayesian knowledge tracing give each skill node a posterior — not a percentage of questions right, but an estimate of what the candidate actually knows, with the item's measured difficulty and discrimination in the model.

Z = n / (n + 5)

Credibility-gated

The engine tracks how much evidence sits behind every estimate, and the interface is required to say so. Two answers on a topic produce low credibility — shown as exactly that, never dressed up as a confident mastery bar. Candidates fail exams on what they were sure about; we refuse to manufacture that certainty.

argmax I(θ)

Information-gain selection

The next question is chosen to learn the most about the candidate — not to pad a streak. Combined with FSRS spaced repetition for retention scheduling, study time goes where the posterior is weakest and the exam weight is highest.

"Three models agreeing is not the same as being correct — and a confident bar built on one answer is a lie."
— from the engine's own design notes. The honesty is enforced in code, not aspired to in copy.
03 · Traceability

Every tutoring session, traceable to learning outcomes.

Amber — our voice AI tutor with a live mathematical whiteboard — is bound to the same graph. She opens each session from the candidate's actual mastery state, anchors every thread she teaches to a named outcome, is forbidden from teaching material that serves none, and closes with a concrete practice commitment she will follow up on next session. The session summary opens with the learning outcomes covered — which means a member's study history is an auditable record against the syllabus, not a chat log.

OPEN

Picks up where they left off

Last session's material, retrieval-checked by name. The practice commitment they made — met, partial, or missed — raised once, with the real numbers.

TEACH

Anchored to outcomes

Each thread names the outcome it serves. Depth follows the outcome's own verb — "describe" is taught differently from "calculate", in both directions.

CHECK

Calibration before confirmation

Candidates commit to a confidence level before hearing whether they're right. Confident-and-wrong is treated as the most important event in the session.

CLOSE

A commitment, on the record

A countable practice target on a named topic by a named day — recorded, and reopened at the start of the next session.

04 · Working with us

Conversations we're built for.

Pinnacle is an independent platform — we are not affiliated with or endorsed by any awarding body. That independence is exactly what makes the verification honest. Here is what we bring to an institutional conversation:

Member benefit programs

Adaptive preparation as a society or association member benefit — priced for cohorts, with the free diagnostic as a zero-friction entry point your members can take today.

  • Cohort licensing & society pricing
  • Branded entry via the public diagnostic
  • Privacy-first: no member data resold, ever

Continuing professional development

The same engine that tracks mastery for candidates is built for maintaining it across a career — spaced retention scheduling, annually refreshed current-issues content, and outcome-traceable session records that fit CPD evidence requirements.

  • Outcome-level activity records per member
  • Annually refreshed current-topic coverage
  • Retention scheduling measured in years, not weeks

Firms & universities

Exam-track cohorts with real visibility: which outcomes a cohort is weak on, who is on track for their sitting, and where teaching time actually moves the needle — from the engine's posteriors, not self-reports.

  • Cohort mastery analytics by outcome
  • Exam-date readiness tracking
  • Candidate-level calibration reporting
05 · Next step

Open the engine. Then talk to us.

The fastest way to evaluate Pinnacle is the way a candidate would meet it: 25 questions, a diagnosis of the root cause rather than a score, and the graph doing its work in front of you.

Take the free diagnostic Read the science
Institutional enquiries — support@pinnacle-prep.org