↯ Adaptive engine

28 students.
28 different Mondays.

Same room. Same lesson. Coursim's engine runs 28 live state machines – one per student – choosing the next question, the next hint, the next path, every few seconds.

7B · Ratios & Rates · Lesson 7live · updated 2s ago
Aarav
Q6 · 82%
Beatriz
+extension
Carlos
Q5 · 74%
Daniela
scaffolded
Emre
Q7 · 79%
Farah
+extension
Gabe
Q6 · 77%
Hana
🌸 sage
Isaac
Q5 · 70%
Jada
scaffolded
Kai
Q7 · 85%
Lena
Q6 · 80%
Mateo
+extension
Nisha
Q5 · 76%
Oren
scaffolded
Priya
Q7 · 81%
Quinn
👧
Ravi
Q5 · 72%
Sara
Q6 · 68%
Tariq
+extension
Uma
Q6 · 78%
Viv
scaffolded
Wes
Q7 · 83%
Xiu
Q5 · 75%
Yusuf
+extension
Zara
Q6 · 79%
Amos
scaffolded
Bea
Q6 · 73%
Extension · flyingOn-level · trackingScaffolded · supportedSage intervention active

The engine in one diagram.

After every single answer, Coursim runs this loop. In ~200ms. For every student on every device.

STEP 1 · TRIGGER
Student taps an answer
every single question
STEP 2 · COLLECT
Read 12 signals
correctness, time, hints, confidence…
STEP 3 · SCORE
Update mastery model
per sub-standard, per student
STEP 4 · DECIDE
Pick next move
harder, easier, reteach, or Sage
STEP 5 · SERVE
Deliver next question
~200ms end-to-end
▼ STEP 4 BRANCHES INTO ONE OF THREE ROUTES ▼

Route A · Level up

when: confident + correct
  • Next question is 0.3 DOK levels harder
  • Skip the "you got it" interstitial
  • Track for extension-path eligibility
  • Shorten the reading passage next lesson

Route B · Scaffold

Route b when hesitant or one miss.
  • Slot in a targeted hint for the sub-skill
  • Show a worked example before retry
  • Lower difficulty by 0.2 DOK for 3 questions
  • Keep the student in the flow zone

Route C · Reteach + Sage

when: 3 misses or confusion
  • Pause the path – reteach the specific misconception
  • Route to Sage with context: "you just missed X because Y"
  • Resume path only after a variant is solved
  • Flag teacher dashboard – "Hana needs a check-in"

12 signals. Every answer.

It's not just right-or-wrong. Coursim reads the whole situation – how long, how many hints, how confident – before it decides what's next.

Signal 01

Correctness

Right, partial, wrong – with the sub-skill pinpointed.

Signal 02

Response latency

Fast-and-right ≠ slow-and-right. Different next-steps.

Signal 03

Hint depth

0 hints, 1 nudge, or full reveal? All tracked.

Signal 04

Confidence tap

Optional self-rating after each answer.

Signal 05

Streak state

Last 5 attempts at this DOK level.

Signal 06

Scratch usage

Did they show work on the whiteboard?

Signal 07

Prereq mastery

Shaky on Q6? Maybe their L5 didn't stick.

Signal 08

Time of day

Post-lunch dip is real. Engine knows.

Signal 09

Session length

After 25 min, difficulty ramp softens.

Signal 10

Device & context

Phone on the bus needs lighter cognitive load.

Signal 11

Peer variance

Only signal tracked – never revealed to student.

Signal 12

Teacher override

Your judgment always beats the model. Always.

Sara's 22 minutes.

A real student arc – one lesson, timestamped. Every adjustment the engine made, and why.

👧
Sara · Grade 7 · Ms. Torres's class
Entered 2 grade levels behind in May. Working through Ratios L7.
+1.3grade levels · 6 mo
0:00
Starts Q1 · warm-up recall
Correct in 4s. Engine notes: smooth recall, drop the modeling video.
baseline
confidence 0.72
1:12
Q2 · ratio from a table
Right, but took 38s and opened the hint. Engine: she has the concept, not the fluency.
hold level
route: more practice
3:45
Q3 · equivalent ratios
Misses. Inverts the ratio – known misconception (signal 01). Engine: the reteach fires, not the next question.
reteach
sub-skill: 7.4A.2
6:20
Reteach: 45-sec video + 1 guided example
Sara watches all of it. Retries Q3-variant and gets it.
recovered
→ resume path
9:02
Q4 · word problem, 2-step
Correct on first try. Response time 22s. Engine starts tracking extension eligibility.
+difficulty
DOK 1 → 2
13:30
Q5 · mixed ratios, multi-part
Hesitates on part (b). Opens Sage. Sage gives a re-phrase, not the answer.
sage · rephrase
hint depth: 1
17:05
Q6 · applied scenario (paint ratio)
Correct, 18s, no hint. Best performance of the session.
momentum
confidence 0.89
21:40
Exit ticket · 2 questions, both correct
Session closes. Ms. Torres's dashboard shows: L7 mastered, 7.4A.2 needs reinforcement Monday.
passed
next: L8 on track

What we're building toward: students who stay in the flow zone

  • 📐 Methodology · not magic
  • Engagement measured as median session-duration delta vs. same class's prior-year static LMS.
  • Full study, raw cohort data, and IRB methodology available on request from
  • the AI policy page Talk to us.

Want to see the engine run on your students?

Bring one lesson from your current curriculum. We'll plug it in live and show you the three differentiated paths it builds in under 5 minutes.

The student grid up top is an example of how the engine assigns paths.