Differentiated SKU
You're the only LMS that measures live-class behavioral signal at scale. Not as a feature ticked off in a comparison chart — as a primary capability nobody else has shipped.
Cadence for schools & LMS platforms
Every live class has behavioral signal running through it. Most LMS platforms can't see any of it.
Cadence is the on-device behavioral measurement layer that runs in every student's browser during live Zoom, Meet, Teams, BigBlueButton, and LMS-embedded classes. It measures attention, engagement, and cross-modal convergence — without recording the call, without uploading media, without putting a third party in the network path. Runs above the meeting layer — works inside Canvas, Blackboard, Brightspace, and Moodle with no LMS-side registration, no admin gauntlet, no integration plumbing required to start measuring.
Drop-out early warning
Students showing convergence signals across modalities surface to instructors weeks before grades reflect it. Class-wide aggregate only — never per-student exposure to the instructor.
FERPA / COPPA / GDPR aligned by construction
Nothing leaves the student's device. No recording, no media upload, no third-party processor in the path. Privacy isn't a policy — it's the architecture.
The headline metric
A composite attention score, recomputed every 30 seconds. Personal baseline, not population norm.
Attention isn't one signal. Cadence decomposes it into six independently-measurable components — gaze on task, fixation duration, head-pose stability, blink-rate deviation, vocal engagement when speaking, and response latency on turn-taking — and computes a single 0–100 index from them.
Every component is normalized against the student's own baseline, not against a population. This is the single most important design decision in the product. Neurodivergent students don't get penalized for non-standard gaze patterns. Instructors don't see "this student is paying less attention than that student" — they see "the class has dropped below its own average."
Per-student attention is never shown to the instructor. The classroom dashboard is a single class-wide curve. Individual self-reflection happens in the student's own Cadence app, on the student's own Mac.
Class avg right now
73
62% of students above their personal baseline
- Gaze on task 25%
- Fixation duration 15%
- Head-pose stability 15%
- Blink-rate deviation 15%
- Vocal engagement 15%
- Response latency 15%
How it's different from your current analytics
Attendance counts heads. Cadence measures presence.
Every LMS already tracks logins, clicks, time-in-page, video completion percentage. None of those tell you whether the student was actually engaging with the material — only that their account was online. Cadence sits one layer deeper, where the behavioral signal actually lives.
| Login & click analytics | Recording-based engagement | Cadence | |
|---|---|---|---|
| Measures actual behavioral signal | No — proxy only | Yes — recorded | Yes — live |
| Records video / audio | No | Yes (server-side) | Never |
| Data leaves the student's device | Yes (server-side) | Yes (server-side) | Never |
| Per-student data exposed to instructor | Yes | Yes | No — class aggregate only |
| Works inside iframed LMS sessions (Canvas + BigBlueButton, etc.) | N/A | No | Yes |
| Personal-baseline (not population) normalization | No | No | Yes |
Privacy by construction
The privacy story is the architecture, not a policy.
Cadence's perception pipeline runs entirely on the student's own device — microphone audio, webcam frames, and transcript fragments never leave the Mac. There is no server-side video processing path, no "anonymized recording" upload, no third-party processor between the student and the measurement. The on-device app derives 32 stable scalar features per 30-second window, and only those scalars — never the raw streams — are available to the classroom aggregator.
The classroom aggregator itself uses k-anonymized federation: a student's component scores are only included in the class average if at least k = 5 other students contributed to the same window. Below that threshold, the value is suppressed. Individual identifiability is structurally impossible from the aggregate.
Cadence holds a HIPAA-eligible BAA with AWS for any cloud-side workloads (citation lookups, software updates, the partner license server). The behavioral data path doesn't touch any of it.
FERPA-aligned
COPPA-friendly
GDPR-compatible
No media recording
No server-side audio/video
AWS HIPAA BAA
Security — defense in depth
Privacy is the architecture (we don't send data worth contracting over). Security is the second layer — every cloud hop, every credential, every privileged surface gated.
Every cloud-reasoner call is signed with Apple's App Attest — proves the request came from a verified build of the Cadence app on a verified Apple device. The server rejects anything that isn't.
Eleplan ⇄ AWS + Eleplan ⇄ Anthropic BAAs signed; OpenAI ZDR in force. The cloud reasoner proxies through our infrastructure — provider API keys never touch the client.
The Admin panel is gated by a password tied to the user's Eleplan account, verified server-side. Same password across every Mac the user signs in on; not resettable by physical access. Grants expire automatically.
Sensitive credentials (Cognito tokens, admin-access grant, bridge token) live in the macOS Keychain with WhenUnlockedThisDeviceOnly — strictest tier. Never backed up, never migrated, locked while the screen is locked.
Cohort aggregates suppress any value below k = 5 contributors. The math forces anonymity — it's not a policy you have to trust us on. Opt-in regardless.
Every release is Developer-ID signed + notarized; in-app updates run through Sparkle with an EdDSA signature verified against a key bundled in the .app. A tampered appcast can't ship to the user's machine.
How it deploys
Students install once. The Bridge detects live classes in any LMS automatically. Instructors see a standalone dashboard — deeper LMS embedding when partners want it.
Cadence is two pieces: a small browser extension (Cadence Bridge) that students install from the Chrome Web Store, and a native macOS app that runs on their Mac during a class. The Bridge sits above the meeting layer — it detects live calls in Google Meet, Zoom Web, Teams Web, and any meeting embedded in Canvas, Blackboard, Brightspace, or Moodle pages, automatically, with no LMS-side registration or admin approval required to start measuring.
Students install once, semester-wide
One-time install of the Cadence Bridge extension + Mac app. Works in every live class the student attends that semester, across every LMS and every video tool — no per-course setup, no LTI registration, no per-class onboarding.
Classroom aggregator on your infra
K-anonymized window aggregates flow to a single endpoint you operate. Cadence ships a reference Lambda; you can re-implement on your own stack. POST /aggregate takes the window scalars; GET /class/{id} returns the live curve.
Instructor dashboard — standalone, embedding optional
The live class curve, historical session strip, and "engagement dropping" alerts run as a standalone web view or inside the Cadence Mac app. Deeper LMS integration — instructor dashboards inside Canvas / Brightspace / etc. via LTI 1.3 — co-developed with launch partners who want it. Not required to start measuring.
Apple Intelligence layer
When the class's behavior signals start to converge, the instructor sees a soft prompt. On every student's device. From an on-device model.
Every student's Mac runs Apple FoundationModels — the on-device language model that ships with macOS. Cadence orchestrates it. When three or more behavioral signals across modalities converge for a stretch of the class, FoundationModels reads the indicator deltas, the canonical metric definitions, and the relevant on-device research excerpts, then produces a one-sentence observation grounded in the citation library.
The instructor sees a class-aggregate version of that observation as a quiet banner — "Cross-modal engagement dropped across the cohort over the last 4 minutes, consistent with cognitive overload" — surfaced only when the aggregate is large enough that no individual student is identifiable. The student's own copy lives in their own Cadence app on their own Mac. Never uploaded, never logged centrally.
The model is grounded in a 47-study seed shipped with the app + 12 indicator kinds computed from 26 canonical measurements in the research metrics spec — and every cited DOI is validated against OpenAlex's 250M-paper scholarly graph for retraction and topic grounding before it persists. Every observation cites its sources by DOI; the cloud reasoner (post-class, with consent) can re-evaluate any fire against deeper analysis. This is the patent's "cross-modal convergence + citation-grounded reasoning" claim — live, on every student's device, with no cloud round-trip.
Why now
The behavioral signal layer is what the LMS layer was missing.
✓
Apple FoundationModels ships on every recent Mac
The on-device reasoning layer is no longer a Cadence-only stack — every student already has the model running in their OS. Cadence orchestrates it; nothing new to install for the AI piece.
✓
Online + hybrid is permanent
Live remote classes are now ~30% of higher-ed credit-hours and rising. The LMS that owns the "is this working?" question for live classes owns the next decade.
✓
Recording-based analytics has hit a wall
Engagement SKUs based on uploaded recordings are now blocked by FERPA escalations, COPPA tightening, and the EU AI Act. Cadence's on-device-only architecture sidesteps every one of those flags.
✓
Patent-pending
The cross-modal convergence detection method is the subject of a provisional patent filed by Eleplan. Licensees inherit a defensible position — not "another engagement tool," a structurally different measurement layer.
Want to see it running in a class?
30-minute walkthrough — live behavioral signal from a real Mac, live aggregator curve, the deployment brief.