EMBER SPARK's Strategic Operating System
Pilot · invite-only

Real-time intelligence for every SPARK school.

EMBER is SPARK Schools' strategic operating system — four engines on a single data core, humans always above the loop, agents below. Sign in to continue.

Pilot click-through auth. Production sign-in lands once Wardensta and SentinelSta approve the consent surface.

What EMBER enables for SPARK

Five outcomes, one operating system.

Single source of truth across operations, finance, learning, and scholar data. Principal admin time halved. Fidelity drift detected in days, not terms. LaunchPad / Catalyst / Nexus / Impact captured as products with self-monitoring fidelity — packaged for any partner operator, anywhere.

Efficiently run schools

Live dashboards, real-time alerts, fewer spreadsheets. Operations and finance from descriptive → diagnostic → predictive.

Codified learning models

SPARK's pedagogy expressed as auditable, exportable artefacts — ready to extend across Africa through partnerships.

Values-aware AI & governance

Six-pillar guardrails. No high-stakes scholar decisions automated. Ethics Council reviews monthly. Trust as enrolment differentiator.

EMBER Engine · 01

Agility

Finance & sustainability — the engine that keeps every school solvent and forecast-honest. Sequenced first alongside Schools.

Agility · Finance & Sustainability

Live P&L · staffing · scenarios · risk

P1 → P3 → Ideal tier tracking · fee-collection risk · cost-per-mastery

Human supervisor · Finance lead + Principal

Outputs (per architecture)

  • P1 → P3 → Ideal tier tracking
  • Live P&L
  • Staffing optimisation
  • Scenario modelling
  • Forecasting
  • Risk & compliance alerts

Agent factory

CatchFee-collection risk · engine leadLead · hired
LedgerBudget variancePending
ScryScenario forecasterPending
RosterStaffing optimiserPending
TiltUtilisation rebalancerPending
YieldCost-per-mastery analyserPending
CompassLong-range P&L forecastPending
TetherRisk & compliancePending
EMBER Engine · 02

Schools

Operations & culture — the engine that keeps daily routines crisp, safe, and inspection-ready. Sequenced first alongside Agility.

Schools · Operations & Culture

Routines · attendance · safeguarding · inspection

Leadership effectiveness · culture health · consistency of experience

Human supervisor · Principal + Operations

Outputs (per architecture)

  • Leadership effectiveness
  • Routines & culture health
  • Consistency of experience
  • Safety & wellbeing
  • Inspection scoring
  • Risk alerts

Agent factory

StokerInspection-readiness · engine leadLead · hired
BeaconSafeguarding/wellbeing riskHired
HearthParent engagement (cross-engine)Hired
FlintRoutines & culture (cross-engine)Hired
SpotterAttendance anomalyPending
PulseRoutines & culture monitorPending
EMBER Engine · 03

Learning

Learning Efficacy (LEE) — the engine that scores SPARK's PL models (LaunchPad, Catalyst, Nexus, Impact) on fidelity and surfaces interventions.

Learning · Learning Efficacy

Fidelity · gaps · benchmarks · interventions

LaunchPad / Catalyst / Nexus / Impact fidelity scoring · internal & external benchmarks

Human supervisor · Curriculum + Master Facilitators

Outputs (per architecture)

  • LaunchPad / Catalyst / Nexus / Impact fidelity scoring
  • Internal & external benchmarks
  • Intervention recommendations

Agent factory

FlintMaster-Facilitator coach · engine leadLead · hired
TunePL Engine fidelityPending
BridgeGap detectionPending
MirrorModel fidelity scorerPending
PolishBenchmarkingPending
NudgeIntervention recommender (HITL)Pending
EMBER Engine · 04

Pathways

Scholar personalisation — sequenced last by design (highest care). Every output is reviewed by a Teacher and Phase Lead before reaching a scholar.

Pathways · Scholar Personalisation

Scholar 360 · grouping · next best step

Personalised paths · interventions · dynamic timetable

Human supervisor · Teacher + Phase Lead (always)

Outputs (per architecture)

  • Scholar 360 Profile
  • AI grouping
  • Personalised paths
  • Next best step
  • Interventions
  • Dynamic timetable

Agent factory

GlowScholar 360 maintainer · engine leadLead · hired
TrailCQ progression + persona arcHired
HearthParent narrative (cross-engine)Hired
WeaveAI groupingPending
StrideNext-best-step (HITL)Pending
Scholar 360 · the data core surface

One scholar, every signal.

The Scholar 360 profile is the spine of EMBER. Every engine writes into the same record, every agent reads from the same record, every human override is logged. Glow maintains the canonical view. Wardensta watches the consent gate.

Scholar 360
Academic · Behaviour · CQ · Engagement · Pathways
Academic Data
Assessments · Grades · Curriculum · Progress
Operational Data
Routines · Attendance · Observations · Safety
Financial Data
P&L · Budgets · Payroll · Procurement · Fees
People Data
Staff profiles · HR · Performance · Tacit knowledge
Engagement Data
Parent · Scholar · Community interactions
Behavioural Data
CQ trajectories · Sparks Fly · Persona-arc
Decision Logs
Master Facilitator decisions · Principal interventions
External Data
Benchmarks · Assessments · Socio-economic context
Cross-cutting

Governance & orchestration.

The conductor and the watchdog. The orchestration layer routes work to the right model with the right context, guardrails, and cost ceiling — selectively in the loop. Governance agents keep every prediction explainable and every override studied.

Orchestration layer

Context · routing · guardrails · HITL · governance

Five layers, one conductor — humans above the loop, agents below.

Reports to · Stace + Adam (supporter)

Orchestration layers

  • Context layer — SPARK semantic ontology
  • Agent router — multi-model, cost ceilings
  • Values guardrails — six pillars, phase-aware
  • Human-in-the-loop — decision rights, explainability
  • Governance — cost, quality, drift, bias, audit logs

Cross-cutting agents

AnvilSemantic ontology · Context layerHired
CinderExplainability + Decision LogsHired
BellowsNetwork-level rollup · governanceHired
WardenstaChild-data safety · POPIAHired
SentinelStaCode & infra securityHired
ScoutstaAlternatives researchHired
LibrarianStaTenant library + glossaryHired
Adjacent · top of funnel

Marketing.

Adjacent to the four engines — not in the architecture roster, but essential. Tinder owns top-of-funnel attention, channel mix, and the partners-adopt-deliver surface that brings new families to SPARK.

Adjacent

Tinder · marketing & channel mix

Top-of-funnel attention · partners-adopt-deliver surface

Reports to · Stace

Surfaces

  • Channel mix & attribution
  • Brand awareness signals
  • Top-of-funnel inquiry volume
  • Partner / referral pipelines

Agent

TinderMarketing & top-of-funnelHired
Adjacent · funnel conversion

Admissions.

Adjacent to Schools. Kindling owns the inquiry-to-enrolled funnel: from first contact through application, assessment, offer, and acceptance.

Adjacent

Kindling · admissions funnel

Inquiry → application → offer → enrolled

Reports to · Stace

Surfaces

  • Inquiry → enrolled conversion
  • Stage-by-stage drop-off diagnostics
  • Cohort intake forecasting
  • Waitlist health

Agent

KindlingAdmissions inquiry-to-enrolledHired
About EMBER

Co-branded with SPARK · powered by EMBER.

EMBEREducational Model & Behavioural Engine for Real-time Intelligence. Built for SPARK Schools, the South African independent network of ~17,000 scholars whose values spell its name: Service · Persistence · Achievement · Resilience · Kindness.

EMBER is SPARK's strategic operating system — distinct in identity but rooted in the same warmth. We don't mirror sparkschools.co.za; we sit alongside it, pointed inward at the people who run the network.

SPARK 5 Ps Framework

Reviewed monthly by the AI Ethics Council.

Policy
Public AI principles
Protection
POPIA-first design
Professionalism
Human always above the loop
Practice
Embedded with non-negotiables
Potential
Scholars' AI literacy & responsible use

Maturity ladder

Every metric is labelled with its current rung. Most ship at Descriptive; we earn the right to move up.

1 · Descriptive  what happened 2 · Diagnostic  why it happened 3 · Predictive  what's likely ahead 4 · Pre-emptive  next best action, with approval

Sequencing

Schools & Agility ship first → Learning → Pathways last (highest care). Scholar-touching agents only land after every guardrail upstream is in place.