How Base Works: The Daily Loop and the Math

The daily loop

The interaction budget is under sixty seconds. That is not a marketing figure; it is the constraint every screen was built against, because a logging habit that costs five minutes a day survives about three weeks and one that costs one minute survives years.

Bodyweight

The pad opens already showing your trend weight with steppers either side. Two taps and you are out. The number that gets stored is the raw reading; the number that decisions are made against is the trend — an exponentially weighted moving average with an alpha of 0.25, so each morning's weigh-in moves the trend by a quarter of its distance.

That distinction matters more than it sounds. Raw scale weight can swing four pounds across two days on identical body composition — glycogen, sodium, food volume, hydration. Steering by raw weight means reacting to noise, and reacting to noise is how people abandon a plan that was working.

Food

Base does not put a search box first, because a serious athlete does not eat differently every day. You save the day you usually eat as a day template — a training day and a rest day is the common pattern — and log the whole thing in one tap.

The hourly timeline underneath shows what landed when. Tapping the + on any hour opens the add sheet, ordered by likelihood rather than alphabetically: your day templates, then your own meals sorted by how often you eat them at that hour, then recent items, then search and barcode as a fallback, then a bare macro entry for the meal you cannot be bothered to name.

The meal you always eat at 1pm is the first thing you see at 1pm. That is the whole trick, and it is the difference between logging for six months and logging for three weeks. Details are in the Food Log screen.

Sets

The set logger opens with last session's numbers already in the fields as ghost values. If you did the same thing, you tap. If you did more, you bump it and tap.

Every working set records reps in reserve alongside weight and reps, and that single extra field is what makes the progression rule work. Without it, a rep-only log cannot distinguish twelve reps at 0 RIR — a genuine ceiling that earns a load increase — from twelve reps at 4 RIR, where the load was simply too light and should jump immediately. A log without RIR will slow-walk you through weights you could already handle for months.

The rule the app applies, per exercise:

every set at top of rep range AND RIR <= 1  ->  add load, reset to bottom of range
inside the rep range                        ->  hold load, add one rep
below bottom of range, or RIR > 4           ->  hold; twice in a row, drop 10%

Endurance

You do not log an endurance session. You tap "heading out" before you leave, and the session is matched back to its prescription afterwards from your watch's own record.

That timestamp exists for one reason: if you run twice in one day and there are two prescriptions, matching them is a coin toss. Base refuses to guess rather than attaching the wrong grade to the wrong session — which would corrupt the exact history the feature exists to build. The Endurance screen covers prescription and grading in full.

What the app computes while you are not looking

Five numbers derive themselves from what you logged. Each one has a plain-English explanation behind an ⓘ on the card that shows it, with the same wording as the comment next to the code that computes it.

Adaptive maintenance. Over a trailing fourteen-day window ending yesterday, your trend weight change is converted to a daily energy balance and subtracted from your average logged intake. What is left is what you actually burned — endurance sessions included, whether or not you remembered them, whether or not your watch estimated them well. It needs ten logged days and eight weigh-ins in the window; below that it falls back to a formula estimate and says so rather than presenting a guess as a measurement. See adaptive TDEE for hybrid athletes.

Today's calorie target. Built on what this day costs rather than on a figure committed last Monday: projected resting energy plus active energy as it is measured plus any manual entries, plus your surplus. The target rises the moment you move, which for a hybrid athlete is the point — a long run on Saturday should not be paid for out of Friday's guess.

Recovery score. Your own thirty-day baselines for HRV, resting heart rate and sleep, each as a z-score, weighted 50/30/20 and renormalised if a signal is missing. It requires fourteen days of baseline and says "building baseline" below that. It is scored against you, never against a population — an HRV of 48 ms is good or bad only relative to your own history.

Threshold. One number per discipline, estimated from your best sustained effort: pace per mile for running, functional threshold power for cycling. Every prescribed band is a fraction of it, which is why it carries its own provenance and its own confidence rather than appearing as a bare number.

Ramp rate. This week's endurance minutes against the average of the previous four. It is the one number in the endurance section worth acting on immediately, because a spike week is both the classic setup for an overuse injury and a direct draw on the recovery budget your lifting is spending.

The weekly loop

Sunday morning, the check-in generates itself. It shows what the week did — trend weight and its change against the target rate, average calories and protein, days logged, sessions completed, total volume against last week, new personal records, maintenance and how it moved, the surplus you set against the surplus you actually averaged, where that surplus can physically have gone, and sleep and recovery averages.

Then it attaches its own recommendation from a rule table, so "you are 22% over your target gain rate" arrives without you having to ask anyone.

It also copies to the clipboard as plain text. That format exists because the most useful thing to do with a week of training data is sometimes to hand it to someone — a coach, a training partner, or a language model — and ask what they make of it. The check-in screen shows the full packet.

What Base does not do

Stating this plainly is more useful than another feature list.

• It does not run your session. Apple Watch and Garmin both handle live pacing, zone alerts and structured intervals natively and well. Base decides what the session should be and grades it afterwards, which is the half they do not do.
• It does not pool strength and endurance into one score. They are different capacities in different units. A blended number could not answer either question and would hide the trade-off you are actually making.
• It does not show a number it had to guess. No HealthKit data for last night means "no sleep data for last night", not a zero and not a dash.
• It does not collect analytics, telemetry, crash reports or an advertising identifier, and ships no third-party tracking SDK. What goes where is set out on the privacy page, and no marketing claim here is allowed to exceed it.

Where your data lives

Every read and write goes to a SQLite database on the device, so the app is instant and works with the network off forever. Everything except progress photos syncs in the background to a private Postgres database under your own account, so losing the phone does not lose the training history — row-level security means only a request carrying your signed-in session can reach your rows.

Progress photos never leave the device. A copy of each one is written into a "Base Progress" album in your own photo library, because the operating system erases everything an app stores when the app is deleted, and that copy is what survives a reinstall. It is your album, not the app's.

What to do next

The fastest way to understand the workflow is the screen-by-screen walkthrough — every screen, what it shows, and the arithmetic behind each number on it.