Why we built mindLimits — aviation training, calibrated to thinking, not memorising

The training problem nobody quite admits

Walk into any flight school’s question-bank software and you find the same pattern: 3,000 multiple-choice questions, recycled, indexed by syllabus topic, ranked by frequency on past exams. Pass the exam — congratulations, you forgot 70% of it within 30 days. This is not a flippant claim. Hermann Ebbinghaus established it in 1885 and the forgetting curve has been replicated thousands of times since: knowledge that isn’t actively retrieved decays exponentially.

The cabin crew CRM landscape is, if anything, worse. Initial CRM training under EASA AMC1 ORO.CC.115(a) is typically delivered as 16 hours of classroom in three days. By month six, the assertiveness training, the cultural-differences module, the lithium battery thermal-runaway sequence — all of it has compressed into a vague memory of “I think we did something on this.” Then a real event happens.

Both groups — pilots facing the ATPL exam, cabin crew facing their line check — are being trained on a model that we have known is broken since the 1880s.

What mindLimits is

mindLimits is a 90-second cognitive trainer for aviation professionals. Each round is ten questions, generated live by AI, calibrated to one cognitive level of Bloom’s taxonomy, anchored to a specific clause of the EASA syllabus. There is no question bank. There is no rote learning. Every stem is fresh, every scenario is regulator-valid, and the system uses SM-2 spaced repetition to bring difficult topics back to you precisely when you’re about to forget them.

It runs in the browser. It is free. It is available in 10 languages — English, Hindi, Tamil, Telugu, Bengali, Marathi, Malayalam, Arabic, Spanish, French.

Two tracks are live now:

• Pilot — EASA Part-FCL 040 (Human Performance & Limitations). 27 topics across three chapters: TEM & SHELL, Physiology & Health, Aviation Psychology. For CPL and ATPL candidates and line pilots. Captain’s Wings rank ladder.
• Cabin Crew — EASA AMC1 ORO.CC.115(a) (initial CRM training). 16 topics across three sections: General principles & physiology, Cognition & awareness, People & performance. For CCM, CCIC, and cabin safety professionals. Cabin Safety Master rank ladder.

A third track — ATC, anchored to ICAO Doc 9683 and EASA SERA human-factors requirements — is in development.

The five teaching principles

mindLimits is not a quiz. It is a cognitive instrument designed against five teaching principles that we believe aviation training should — but rarely does — operationalise.

1. Cognitive depth (Bloom’s taxonomy)

Every question is calibrated to one level of Bloom’s taxonomy: Remember · Understand · Apply · Analyse · Evaluate. As the player’s rank rises, the cognitive demand of the questions rises with it.

A Student Pilot sees recall and recognition items. A First Officer sees scenario application and CRM judgement. A Captain sees ambiguous, compounded threats where every option is partially defensible and the player must evaluate trade-offs. Bloom 1956 established that cognitive levels are functionally distinct — a person who can recall a definition cannot necessarily apply it under pressure, and a person who can apply it cannot necessarily evaluate competing applications. mindLimits exposes the player to all five levels in proportion to their rank, and the rank ladder is gated on demonstrated performance at each level.

2. Regulator-anchored (EASA Part-FCL 040 + EASA AMC1 ORO.CC.115(a))

Every concept in mindLimits maps to a specific clause of a regulation. A pilot question on Time of Useful Consciousness lives under EASA Part-FCL 040 02 (Physiology & Health). A cabin-crew question on lithium battery fire response lives under EASA AMC1 ORO.CC.115(a) §11 (Threat & Error Management). The system enforces this server-side: pilot concepts cannot be served on the cabin track, and cabin concepts cannot be served on the pilot track. Cross-track contamination is rejected at the input gate before the AI is even called.

Why this matters: most AI-driven training tools take a generic prompt (“write a CRM question”) and accept whatever the model produces. mindLimits constrains the model to specific regulatory clauses with specific learning objectives, then validates the output against a ten-layer scenario-validity matrix. The player sees questions that an EASA examiner would recognise as on-syllabus, not LLM hallucinations dressed up in aviation vocabulary.

3. The Vedic learning cycle: Abhyāsa · Manana · Nididhyāsana

This is the philosophical foundation of mindLimits and the part we suspect most modern training products will quietly steal in the next two years.

Abhyāsa (अभ्यास) — daily, sustained practice. From the Patañjali Yoga Sūtras (1.13–1.14): abhyāso ‘sau dīrghakāla-nairantarya-satkārāsevito dṛḍhabhūmiḥ — practice becomes firmly grounded when it is done over a long time, without break, with reverence. mindLimits operationalises this as the 90-second daily round.

Manana (मनन) — reflection on what is heard. From Vedanta. Knowledge does not become useful by accumulation; it becomes useful through the second pass — the asking of “what does this mean for me, here, now?” mindLimits operationalises this as the post-question debrief: every wrong answer is shown beside the correct answer with a 1–2 sentence rationale, and the post-round summary footer reads “Manana — your reflection is the retention.”

Nididhyāsana (निदिध्यासन) — sustained meditation until knowledge becomes instinct. The third pass, where understanding integrates into reflex. mindLimits operationalises this as the SM-2 spaced repetition layer: topics return at intervals calibrated to your forgetting curve, until your recall stabilises.

The 90-second round is one revolution of a 2,000-year-old learning model. Not a metaphor — a literal implementation.

4. Memory science: SM-2 spaced repetition

Most aviation training products serve questions in the order the syllabus is laid out. mindLimits orders them by what you’re about to forget next.

This is the algorithm that turns short-term recall into long-term mastery — the same algorithm that powers Anki, Quizlet’s premium tier, and the spaced-repetition apps that medical students use to retain pharmacology over five-year programmes. The order of questions in mindLimits is not random. It is calibrated to your individual forgetting curve.

5. Scenario-true AI generation

Every question in mindLimits is generated live by Google Gemini, in the moment you click Play. There is no question bank, no recycled stems, no “I’ve seen this question before” tell. You will never see the same item twice.

But — and this is the part most AI-training tools don’t do — the model is constrained by a ten-layer validator stack that rejects any item violating EASA principles, scenario plausibility, option-length parity, or the aviation validity matrix. Forbidden combinations like “rapid decompression below FL100” or “lithium battery fire suppressed by water as primary” are caught and the item is rejected; the system retries up to six times, with the previous failure reason fed back into the next prompt, before giving up. The validator gauntlet refuses bad items rather than ship them. The result: every question the player sees is fresh and scenario-valid.

Why two tracks (pilot and cabin crew)

A pilot does not double as a cabin crew member. A cabin crew member does not hold a flight licence. The regulators recognise this — pilot Human Performance is governed by EASA Part-FCL 040, cabin crew Crew Resource Management by EASA AMC1 ORO.CC.115(a). The two syllabi share core human-factors science (fatigue, attention, decision-making) but differ in their scenarios, their rank ladders, their evaluation milestones.

mindLimits enforces this separation at every layer:

• Per-track concept allowlists — pilot concepts on the cabin track are server-rejected, and vice versa.
• Per-track localStorage — your pilot stats live in one bucket, your cabin crew stats in another. Switching tracks does not contaminate the other side’s progress.
• Per-track Firestore leaderboards — mindlimits_leaderboard_pilot and mindlimits_leaderboard_cabin_crew are separate collections. Neither track can overwrite the other’s row.
• Per-track AI prompts — the system prompt for cabin crew has its own ROLE CONTEXT block, its own validity-matrix additions (no decompression sub-FL100, no evacuation with engines running, no water as primary suppressant on lithium fires), and its own worked example.
• Account lock after first round — once a user completes their first round, the track is committed to that account. Playing the other track requires signing up with a different email. This mirrors aviation role identity: one CPL, one CCM, two different documents, two different identities.

Why ATC is next, and what’s after

Air traffic controllers are subject to the same human-factors training requirement under ICAO Doc 9683 — situational awareness, communication, fatigue, threat and error management — but no equivalent micro-learning tool exists today. The ATC track is the obvious third leg of the aviation triangle, and the cabin track’s content engineering will accelerate it.

Beyond aviation, four roadmap candidates are in scoping:

• Healthcare — surgical and emergency-medicine teams have the same human-factors science as cockpit crews; the mindLimits engine generalises with a different syllabus.
• Students & Academics — cognitive training as a study skill, applied to whichever syllabus the student is preparing for.
• Safety Leadership & Industry — operations managers, plant safety officers, the supervisory layer that’s responsible for systemic safety.
• General professional cognitive training — broader, role-agnostic.

The shared engine is the asset. The track is just the syllabus.

Who built it

Capt. Amit Singh FRAeS — line captain, Fellow of the Royal Aeronautical Society, founder of Safety Matters Foundation. Author of the HPL Workbook v1.5 that mindLimits is built on. Decades in commercial aviation, decades in safety advocacy. The cognitive science underneath mindLimits is his professional position; the workbook is his curriculum.

Safety Matters Foundation is a not-for-profit advancing aviation safety in the Indian subcontinent. SMF owns the curriculum copyright. The HPL Workbook v1.5 — 27 topics across the EASA Part-FCL 040 syllabus, with learning objectives, scenarios, and references — is the source of truth for the pilot track.

mindFly Technologies is the technology vehicle that operationalises the SMF curriculum into the mindLimits product. mindFly Technologies licenses the curriculum from SMF, builds and runs the AI generation pipeline and validator stack, hosts the application, and ships the user-facing product. The two-entity structure is deliberate: it separates the academic owner of the syllabus from the technology operator of the platform, and it lets the curriculum continue to belong to a non-profit even as the technology scales.

BETA: how to participate

mindLimits is in BETA as of May 2026. The core gameplay loop, the AI generation, the validators, the leaderboards, the multilingual layer, the track lock, and the rank-first scoring are all production-ready and tested. What we are still tuning:

• The cabin-crew prompt. Pilot content has been validated against thousands of practice items over years; the cabin AMC1 prompt is new and benefits from real cabin-crew feedback on question realism.
• Multi-language quality at scale. The Hindi/Tamil/Bengali/etc. layer works correctly but the model occasionally produces a stilted phrasing that a native reader would polish. We want those reports.
• Edge-case scenarios. The validator gauntlet catches the obvious mistakes; subtle ones still slip through. Tell us when one does.

If you train aviation — pilot, cabin crew, instructor, examiner, training department head — please play a round and email Admin@safetymatters.co.in with your reaction. Specifically helpful:

• Cabin-crew tester reports on question realism (we have very few cabin-crew testers right now).
• Hindi-language polish notes (which questions read awkwardly?).
• Specific scenarios that should never have shipped (so we can add them to the validity matrix).

We read every email. The first 100 BETA testers will be acknowledged on the eventual production launch page.

Play the 90-second challenge

The fastest way to evaluate mindLimits is to play one round. It will take 90 seconds. Free. No download. Sign in with Google or email.

mindLimits™ is a product of mindFly Technologies. Curriculum licensed by the Safety Matters Foundation. HPL Workbook v1.5 by Capt. Amit Singh FRAeS. EASA Part-FCL 040 + EASA AMC1 ORO.CC.115(a). Abhyāsa · Manana · Nididhyāsana — the Vedic learning cycle, in 90 seconds.