How EntrenAI calculates your plan
The plan is not written by a language model: it is calculated by an in-house rules engine built on training science, with formulas you can read, question and check. This page explains them, what evidence they rest on and where their limits are.
Updated
At a glance
| Who calculates the plan | A rules engine, not a language model |
|---|---|
| How it progresses | One week at a time, in blocks with a deload week, looking at how the previous week went |
| Daily adjustment | Sleep, stress and fatigue check-in → Readiness Score from 0 to 100 |
| Readiness Score | 40% accumulated volume + 30% sleep + 30% stress |
| References | Daniels (VDOT), Seiler (80/20), Schoenfeld (volume and frequency), Coggan (power zones), Friel (periodization) |
| Data it reads | Your check-in, your sets and RIR/RPE, and Strava, Apple Health or Health Connect if you connect them |
| Made by | Iván López Martín · more about EntrenAI |
How this differs from a plan written by a language model
A language model writes plans that sound good, but nothing makes them consistent from one week to the next: it does not know the load you have accumulated or remember how you responded on Monday. EntrenAI splits the two jobs. The engine decides what to do, how much and how hard, with fixed, reproducible rules: same data, same plan. Natural language is kept for what it does well: reading your notes, explaining the session and answering questions (see "Where generative AI is used" below).
What goes into the engine
- Your initial answers: goal, level, available days and hours, equipment, injuries and, if you are preparing for an event, its date.
- Your daily check-in: sleep, stress and fatigue.
- What you do: every set with its weight and RIR, each session's RPE, real paces and power.
- Your external data, if you connect Strava, Apple Health or Health Connect (watches that sync with them arrive that way).
From that, the engine generates one week at a time, in blocks with a deload week, taking into account how the previous one went, how far away your race is and when a test is due. If you miss a session, the week is rearranged without breaking the block.
The Readiness Score
Every morning, the check-in becomes a number from 0 to 100 using a simple formula:
40% accumulated volume + 30% sleep quality + 30% stress level
The engine uses that number to decide whether the day's session stays as planned, gets lighter or gets heavier:
- If fatigue is critical, the plan cuts the day's volume by 40% and caps RPE at 7.
- If you are recovering better than expected, it adds 15% load.
- Otherwise, the session stays as it is.
It is an EntrenAI rule, tuned to decide the day's load. It is not a validated clinical index, and it does not replace a heart-rate monitor, heart-rate-variability analysis or a doctor's opinion.
How paces, loads and zones are calculated
Running. Paces come from your VDOT (Daniels), estimated from a recent result. If you outperform the plan in a key session and your VDOT goes up, the engine recalculates all zones for the following week. If you consistently run your easy sessions faster than prescribed, a very common mistake, the plan lowers the target pace instead of letting it slide.
Cycling. Everything hangs on your FTP, which the plan calculates from a 20-minute test (average power × 0.95) and repeats every 6 weeks. The 7 Coggan zones come from it, and the block type (polarized, pyramid or sweet spot) changes with your goal and level.
Strength and hypertrophy. Volume and frequency follow Schoenfeld's meta-analyses, with ranges by level (for example 12-20 or more sets per muscle group per week for hypertrophy, and frequency 2 on key muscle groups). Progression runs on RIR: after each set you log how many reps you had left, and if your average RIR on an exercise is 4 or more, the engine raises your base weight by 10% next time.
Endurance. In cycling, running, swimming and triathlon, intensity distribution follows Seiler's polarized 80/20 model: most of the time at low intensity and a small share on quality work.
Several sports at once. If you train, say, strength and running, the engine spreads the week so the leg session does not ruin the long run. That is what no single-sport plan can do.
Where generative AI is used, and where it is not
- Yes: reading free-text check-in notes ("my calf hurts", "bad day") and turning them into data the engine understands; writing the comment after each session; writing your progress analysis; answering questions in the in-app assistant.
- No: deciding your loads, paces, volume or calendar. That always belongs to the engine.
Limits
- It is not medical advice. If something hurts persistently, see a healthcare professional. The plan can lighten a session, but it does not diagnose.
- Predictions are estimates. The HYROX race-time prediction, for example, has a ±3% margin only if you have completed at least one medium-level simulation.
- The plan is only as good as the data. A check-in filled in at random produces a random Readiness Score.
- It is a new system. EntrenAI launched in May 2026. The rules are reviewed and improved, and this page is updated when they change.
Sources
The sources for each sport are cited on its plan page. The main ones:
- Daniels, J. Daniels' Running Formula: VDOT and pace zones.
- Seiler, S. (2010). What is best practice for training intensity and duration distribution in endurance athletes? International Journal of Sports Physiology and Performance: polarized distribution.
- Schoenfeld, B. et al. (2016, 2017, 2019): volume, frequency and rep range in hypertrophy.
- Coggan, A. and Allen, H. Training and Racing with a Power Meter: power zones and FTP.
- Friel, J. The Cyclist's Training Bible: block periodization.
Missing something, or spotted a mistake? Write to soporte@entrenai.app.
Try it yourself
Download EntrenAI and tell it what you train: free workout logging costs nothing.