Skip to content

Eat · Sleep · Train · Work · Repeat

It remembers why.

Your watch, your scale, your nutrition app and your training platform produce data every single day. Nothing joins them up, and nothing remembers what you already tried.

ESTWR brings all four together, plans the week on what they actually say, and writes down every decision next to the reasoning behind it. In March you'll know what you changed in August, and why.

See the record

What makes it different

A decision history, not a metrics dashboard.

A dashboard tells you HRV is down. A coach with a memory remembers you already cut volume back in May for the same reason, and that it didn't work. The gap between those two things is what makes advice accumulate instead of starting from scratch every month.

04 AUGveto

Running volume frozen

Reasoning

The right foot has sat at 2 out of 3 for three days: clear discomfort that doesn't change the gait. The rule freezes running volume until it drops back to 1. The soleus protocol carries on untouched, because that's what holds up October's volume.

Data behind it

right_foot 2/3 · 3 consecutive days

28 JULveto

Burpee broad jumps cut to 30 % of the split

Reasoning

Each station's split is assigned on comparative advantage. Not this one: a plyometric landing puts four to six times body weight through the forefoot, and the athlete's own protocol flags it as a movement to avoid. The constraint wins.

Data behind it

Permanent constraint · non-negotiable

21 JULajuste

Deload pulled forward by a week

Reasoning

The seven-day rolling average for HRV has spent five days below personal baseline, with sleep steady and no schedule changes. Accumulated load was running ahead of recovery, so the deload moves up rather than forcing the week.

Data behind it

HRV 7-day average: −11 % vs baseline · sleep steady

09 JULobjetivo

Target revised to 1:09:30

Reasoning

The 2025 Roxzone came in at 6:21, against the 4:30–5:00 that efficient teams post. The 2025 breakdown puts 80 to 90 seconds there: jog the transitions and arrive at each station with the split already agreed.

Data behind it

2025 baseline: 1:12:43 · Roxzone 6:21 (9 % of total)

Every change to the plan writes a decision. No exceptions. One decision can revoke another, and the whole history stays readable — nothing is deleted.

The five domains

Five verbs, and the fifth is the one that counts.

The first four are data domains. The fifth is the loop that ties them together: plan, execute, measure, adjust, record. Without that loop, the other four are just charts.

  1. Eat
    Eat

    Calories, macros, adherence and body composition. What you ate, not what you meant to eat.

  2. Sleep
    Sleep

    Duration, stages, latency, regularity, HRV and resting heart rate.

  3. Train
    Train

    Sessions, training load, paces, progression and reported symptoms.

  4. Work
    Work

    Energy, focus and perceived load. The domain almost nobody measures, and often the cause of poor sleep rather than its victim.

  5. Repeat
    Repeat

    Plan, execute, measure, adjust, record. The loop is the product.

How it decides

The week plans. The day adjusts.

A single night's HRV is noise: a late dinner, a drink, a loose strap. A system that reacts to that lurches from one extreme to the other, and gets abandoned within three weeks. ESTWR plans on seven-day rolling averages, and only moves tomorrow's session when a deterministic rule holds across a window of two days or more.

A single day's reading

Noise. It moves nothing on its own.

Seven-day rolling average

Signal. This is what decisions rest on.

You are the exception. A symptom you report counts from the first minute, with no window to wait for.

Adjustments are forward-only: they change the days still ahead, never rewrite the ones already gone.

Averages, deltas and thresholds are deterministic SQL, not a language model's judgement. The model interprets data that has already been aggregated; it doesn't calculate, doesn't compare against baselines and doesn't evaluate thresholds.

The limits

Some sessions the system can never even propose.

A constraint with veto severity doesn't warn: it blocks. The right foot is the reference case, and it's enforced in code rather than in the wording of a prompt — because a prompt can be talked round and a check cannot.

Preference

Steers. «The gym has no ski erg», «prefers to train at 7am». The system takes it into account and proposes something else.

Veto

Blocks. A session that breaks it never gets generated. It isn't a warning you can dismiss.

And one limit that doesn't move

ESTWR observes and correlates. It does not diagnose. Wearable data are estimates, and vendor-proprietary metrics — estimated VO2 max, energy scores, in-house zones — are stored for reference and decide nothing. When an anomalous signal persists, the right output is a referral to a professional.

Where it stands

Two athletes, one race, 12 November.

Hyrox Barcelona 2026, Doubles. The baseline is 1:12:43. The target, 1:09:30. Whatever the system decides along the way will be on the record, with its reasoning.

2025 baseline1:12:43
2026 target1:09:30
Difference−3:13

Internal use. No sign-up, no waiting list, nothing to sell.