The Mavromatis Acceleration Principle · MAP
Explore as
00
AI gives the answer. Your brain stops doing the work.
We call this answer collapse. It costs you the skill you came to learn - and at scale, it costs society a generation that can't think without a machine. The Mavromatis Acceleration Principle (MAP) is a 3-step structural fix for it, not a prompting trick.
Write first
→
AI critiques
→
Recall, days later
Without MAP
Fades fast
With MAP
Holds steady
Illustrative - this is what Hypothesis H2 is designed to test, not a published result.
New
MAP Research Toolkit · Open access, CC-BY 4.0
→
MAP
Research
Toolkit.
Research
Toolkit.
Everything needed to study, replicate, challenge, or extend the Mavromatis Acceleration Principle. Free. No forms. No permission required.
6
Documents
5
Hypotheses
30
Day pilot
CC‑BY
4.0 Open
MAP in the classroom · 2027
Pilot results · 2027
Research · Open
01
Prompt Scripts
Gate · Layered feedback · Spaced retrieval
Download
02
Process Log
10-field per-session record
Download
03
Micro-Outcome Worksheet
Skill decomposition in 5 steps
Download
04
Coding Rubric
0–4 reflection quality scale
Download
05
Pilot Protocol
30-day · 3-condition study
Download
06
Pre-Registration Template
AsPredicted · OSF-ready
Download
Implementation
1a · The GateBefore every session
Learner writes first
Submit this before typing anything into the AI. The AI does not respond until this is complete. This is the structural countermeasure to answer collapse.
Micro-outcome: [state the single skill you are targeting today]
My hypothesis / solution attempt: [write 2–4 sentences of your own thinking before the AI sees the problem]
Confidence level: [1–10]
My hypothesis / solution attempt: [write 2–4 sentences of your own thinking before the AI sees the problem]
Confidence level: [1–10]
1b · Layered FeedbackSubmit with 1a
Five-part response contract
Transforms the AI from an answer machine into a coach who critiques your reasoning. The structure is non-negotiable.
[Paste problem] / My attempt: [paste from 1a]
Please respond in exactly this structure:
1. CORRECTION · one sentence, first error only
2. MISCONCEPTION · the wrong belief named
3. FULL SOLUTION · complete worked answer
4. ALTERNATIVE · different method + trade-offs
5. FOLLOW-UP · harder variant to defeat your plan
Please respond in exactly this structure:
1. CORRECTION · one sentence, first error only
2. MISCONCEPTION · the wrong belief named
3. FULL SOLUTION · complete worked answer
4. ALTERNATIVE · different method + trade-offs
5. FOLLOW-UP · harder variant to defeat your plan
1c · Spaced RetrievalDay 7 and Day 21
Cold recall before feedback
What you still know at day 21 is what you actually own. Speed tells you how fast you got there. Recall tells you if you stayed.
I learned [micro-outcome] approximately [N] days ago. Do not show me the solution yet.
Ask me to solve this cold: [restate original problem]
After I attempt it, give layered feedback using the same five-part structure.
Ask me to solve this cold: [restate original problem]
After I attempt it, give layered feedback using the same five-part structure.
| Field | What to write |
|---|---|
| Date | Today's date. |
| Micro-outcome | The exact skill targeted this session - one sentence, testable. |
| Hypothesis | Your pre-AI reasoning, copied from prompt 1a. Do not edit it after the fact. |
| Confidence in | Your confidence before seeing AI feedback (1–10). |
| AI correction | Copy the AI's correction only, not the full solution. |
| Misconception named | Copy the misconception the AI identified. |
| Reflection | In your own words: what did you believe that was wrong? What is your new mental model? (2–5 sentences) |
| Speed metric | Time from starting the problem to your first solution attempt, in minutes. |
| Day 7 recall | Your unaided recall score at day 7 (0–10). |
| Day 21 recall | Your unaided recall score at day 21 (0–10). |
1
Name the domain
e.g. "Python for data analysis"
2
Name the broad skill
e.g. "Data cleaning with pandas"
3
Break into 5–8 micro-outcomes
e.g. "Load a CSV" / "Drop null rows" / "Rename columns" …
4
Write the test for each
"I can do this correctly without AI." Define what 'correctly' looks like in one sentence.
5
Sequence them
Order foundational to dependent. One micro-outcome per session. Advance only at 8/10.
-
Label
Criteria - must meet ALL to earn this score
0
Absent or copied
No reflection, OR verbatim copy of AI output with no learner restatement.
1
Surface acknowledgment
Learner states the correct answer but does not name the misconception that caused the error.
2
Misconception named
Learner names the specific wrong belief and states it was incorrect, but does not explain why.
3
Misconception explained
Learner names the wrong belief AND explains the correct principle that replaces it, in their own words.
4
Strategy generalized
All criteria for score 3, PLUS learner states a personal rule or check to apply to future problems of this type.
Research design
H1
Speed
MAP learners match or exceed unstructured AI users on time to first correct solution per micro-outcome.
Fails if: gate friction significantly slows acquisition.
H2
Fidelity
MAP learners significantly exceed unstructured AI users on delayed-recall accuracy at 7 and 21 days.
Fails if: no significant difference at either interval.
H3
Transfer
MAP learners outperform both comparison conditions on structurally novel problems requiring the target skill.
Fails if: transfer scores do not differ significantly.
H4
Metacognition
MAP increases metacognitive awareness; coded log quality predicts individual fidelity gains.
Fails if: MAI scores do not differ; log quality unrelated to retention.
H5
Durability
After scaffold removal, former MAP learners show less answer collapse than former unstructured users.
Fails if: no behavioral difference in unscaffolded phase.
"MAP will only matter if it is true. We think it is true. We have not run the trials ourselves. You are. If it fails under rigorous testing, we want to know now, not after adoption. Either way, the learners win."
Pavlos Mavromatis · Mavromatis Institute · July 2026
Create a free OSF account
At osf.io. Takes two minutes.
Start a new pre-registration
Use the AsPredicted template - nine questions, under 30 minutes.
State your hypotheses
Copy the exact wording of whichever H1–H5 you are testing. Paste verbatim. Do not paraphrase. This creates an unambiguous audit trail.
Specify your analysis plan
Before collecting data: statistical test, significance threshold (p < .05 standard), effect size (Cohen's d recommended). The pilot protocol includes analysis language you can paste directly.
Submit and send us the link
OSF locks the document and issues a timestamp. Email the@mavromatisinstitute.org. We add your study to the MAP registry.
Pre-Registration Template · AsPredicted Format · Copy into OSF
Have any data been collected?
No, data collection will begin after this pre-registration is submitted.
Main hypothesis being tested?
[Copy the exact wording of H1–H5 from the MAP Researcher's Invitation.]
Key dependent variable(s)?
[State your outcome measure from the pilot protocol corresponding to your hypothesis.]
How many observations?
[Feasibility: 10–20 per condition. Justify with power analysis for a powered RCT.]
Stopping rule?
Target N per condition reached or 35 days elapsed. No interim analyses.
Statistical analysis?
One-way ANOVA, planned pairwise contrasts (A vs. B; A vs. C). Cohen's d with 95% CI. p < .05. Intention-to-treat.
Framework citation
Mavromatis, P. (2026). Learn faster, think deeper: The Mavromatis Acceleration Principle (MAP) [White paper]. Mavromatis Institute. https://doi.org/[to be assigned]
Mavromatis, P. (2026). Learn faster, think deeper: The Mavromatis Acceleration Principle (MAP): An AI-native framework for self-directed learning that preserves the learner's thinking [White paper]. Mavromatis Institute. https://doi.org/[to be assigned on preprint upload]
Bloom, B. S. (1984). The 2 Sigma Problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6), 4-16.
The tutoring-effectiveness gap MAP is trying to close without a human tutor for every learner.
Roediger, H. L., & Butler, A. C. (2011). The critical role of retrieval practice in long-term retention. Trends in Cognitive Sciences, 15(1), 20-27.
The basis for the day 7 and day 21 recall test in the Process Log.
Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19(11), 1095-1102.
Why the recall test is spaced days apart, not asked immediately.
Zimmerman, B. J. (2000). Attaining self-regulation: A social cognitive perspective. In Handbook of Self-Regulation (pp. 13-39). Academic Press.
The self-regulated learning theory behind writing a hypothesis before the AI responds.
Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489-530.
Direct empirical evidence for what we call answer collapse.
Download the white paper for the complete reference list and full argument.
08
Frequently Asked
The questions that arrive most often, answered directly.
Is MAP proven?
MAP is a framework paper, not an outcomes study. The building blocks are proven across decades of research. The specific combination awaits its own controlled trial. That is why the Researcher's Invitation exists - and why every hypothesis is stated so it can fail.
How is this different from just using ChatGPT?
Unstructured AI use is what MAP calls the comparison condition. The difference is structural: MAP requires you to externalize a hypothesis before the AI responds. The AI then critiques your reasoning instead of replacing it. MAP is not a prompting style; it is a workflow with a non-optional step - the gate.
Can schools and companies use MAP?
Yes, without asking permission. All materials are CC-BY 4.0. Download, adapt, translate, build on. Attribution only. To run a formal pilot and contribute to the evidence base, see the Researcher's Invitation.
What does MAP cost?
Nothing. Framework, materials, and pilot protocol are all free. Running a pilot requires only the AI tools you already use. Consulting and training on implementation are available for organizations that want supported deployment.
How do I join the MAP Pilot Consortium?
Contact form below or email the@mavromatisinstitute.org. Founding consortium partners run the shared 30-day protocol, contribute anonymized data, and receive pre-publication results. Named in resulting research. No fee. The contribution is the data.
09
Get in touch.
Consortium enquiries, pilot partnerships, research collaboration. Responses within 48 hours.
Message received. We will be in touch within 48 hours.
Stay connected
Pilot results.
First to know.
One email when the first MAP studies are published. No noise, no spam.
Free. Unsubscribe anytime.