What shipped, and what it bought

The column you are in is marked, on every chartA week that is two days old was drawn exactly like the fifteen finished weeks beside it, so four chest sets on a Tuesday read as a muscle that had dropped out of its range. The day or week still in progress now carries a soft gold glow — the same light that marks the set you are on — so a part is never read as a whole.

And on most charts it fills as you trainYour calendar and your volume panel already moved as you logged. Weekly sets per muscle, the form cards and a lift's long-run curve did not — they showed whatever had loaded when the app opened. Today's work goes in as it happens now. Records and average duration still wait for the session to end, because a half-finished one would make both of those wrong.

The section shortcuts on Trends went nowhereThe row of chips above the Trends tab lit up whichever one you pressed and left the page exactly where it was. Tapping a section takes you to it.

A workout clock at the top of every screen, and it is a buttonWhile a session is running, a small capsule sits at the top right with your elapsed time in it. Tap it and you are back on the exercise you are on, with the list landing on the set itself rather than at the top of it.

The rest ending walks you back whatever the timer looks likeIt already did that from the full dial. Close the dial and the same rest ran out into a small pill, stood itself down, and left you wherever you were — which four sets in is a screen of history with your next set below the fold. Both now do the same thing.

Add a set to the plan, not just to todayEvery plan edit used to last until you closed the app. Now the exercise menu offers Add a set · Today and Add a set to the plan · Every time. They are separate on purpose — one more today is a good session, a permanent set is a change to your dose.

Is any of that volume loading the stretch?Training a muscle at long lengths grows it more than the same work at short ones. The projected-volume screen now ends with the muscles that look fine by set count and have nothing in them that does it — leg extensions and presses with no squat pattern, say. It only says so where the exercises are classified, and tells you how many are not.

MRV is no longer described as a wallPast the top of the volume range the app used to say more was not buying anything you could recover from. That is a claim about you it cannot support: the landmark figures are a coaching framework rather than a measurement, and recent work has returns still climbing past them. It now says the evidence thins out and the recovery cost is real, and the guide says where the numbers come from.

Coming back takes sets off before it takes weight offStrength holds up over a couple of weeks — what you lose is work capacity, so the full set count is what makes the week after your first session back hurt. Ten days off now opens a movement one set down at the same load; longer gaps take weight off as well.

It knows being away from skipping somethingIf you have been training and just have not benched in a fortnight, only the bench is stale and the sets stay. If you were away altogether, everything comes back a set lighter. Either way it applies to the first session back and clears itself once you train it.

A rough first session back is no longer read as regressionThe coach compared your last two sessions of a lift without looking at the dates, so the first session after a break counted as going backwards — and one ordinary session later it cut 7.5% off a lift that was fine. A session separated from the one before it by a layoff is no longer evidence of anything.

An unrated set can no longer earn a load increaseA set logged without an effort rating was being read as zero reps in reserve — the strongest evidence there is — which cleared the one hard rule the engine has before adding weight. It holds the load now and says why.

The trend line follows the block, not the sessionHow much the line is smoothed now scales with the span you are looking at. A year of training no longer draws noisier than a month of it.

Planned Workouts states what you programmedThe overview led with a projected week built on your training-frequency setting; it now states the rotation itself — how many workouts, how many sets, how long. Each card carries its place in the rotation, so six workouts read as a sequence.

Your training days, next to what you have loggedThe setting sits beside your actual average over the last 28 days, with a line saying what any gap costs the weekly figures below it. A setting given once and never revisited had nothing anywhere able to contradict it.

Training days are counted in your own timezoneThey were bucketed in UTC, so an evening session and the next morning’s could count as a single day. That only ever undercounts how often you train.

A new exercise opens on a real weightA movement with nothing in your log read 0 kg on every row, under a verdict saying HOLD. It now opens on the lightest the equipment makes and says FIRST TIME, because nothing had been decided yet.

Barbell jumps are a plate you ownThe step was small enough to prescribe 67.5 kg, which needs a 1.25 a side. Barbells and smith machines move 5 kg at a time now, and the plate view stops at plates most gyms actually have.

The warm-up stopped being longer than the workEvery warm-up rung took the full working rest, so a bench ramp put ten and a half minutes of countdown in front of the first real set. Rest between warm-ups is a minute, and the rungs land on whole plates instead of asking you to hang and strip three a side.

Your log shows the workout you just finishedThe log, the calendar, the trends and the career totals only refreshed when the app was launched from cold — so on a phone that keeps the app in memory you could finish a workout, watch it upload, and still not see it for days. Nothing was lost and nothing showed an error; the app simply never asked again. It now asks when everything it was holding has been accepted, and when you return to an app that has been in the background a while.

A workout logged with no signal is no longer thrown awayFailing to reach the server was being treated exactly like the server rejecting the workout, and after a few minutes of failed retries the session was discarded. That is the one situation the offline queue exists for. Not reaching the server is now only ever a reason to wait.

Opening the app sends what it is holdingLogging a set, finishing a session and switching back to the app all triggered a send — but opening the app fresh did not. The launch most likely to be carrying an unsent workout was the one that sat on it.

Sync follows the signal, and does not wait for wifiWhen a connection comes back the queue goes out within seconds rather than waiting out a retry delay. On cellular, a hotspot, or with Low Data Mode on, sets are batched into a handful of uploads instead of one per set — the same workout, a fraction of the requests. A session logged away from home syncs where it was logged.

Tell the app you changed gymA new line at the foot of the pre-workout check-in. Mark it once and both strength charts break at that session: the line splits, a dashed rule marks the boundary, and the gap is carried by a dotted line rather than left as a hole — the series continues, the step in it is just not a reading. Each side is averaged separately too, so the old gym’s numbers no longer bleed into the new gym’s.

It changes nothing your coach doesYour next prescribed load still comes off the weight you actually lifted, because the new machine’s 60 kg is the number to beat next week. This only changes what the charts draw.

A way out of the check-in, in the cornerThe check-in gains a close button that does what Start now already did — begins the workout and keeps whatever you have answered. Closing a check-in is declining to answer the rest of it, not throwing away the answers you gave. The back button is now the same size as every other round button in the app, and the question icons lost the box behind them.

All time, on your trendsThe range bar gains All beside 7D, 28D, 3M, 6M and 1Y, and it means what it says: the window opens at your first ever session rather than at the slice of history loaded on your phone. The strength index, sets per muscle and the effort spread are all bucketed across that whole span.

A single lift, over a year or over everythingThe trend on one exercise used to stop a few months back — that was as far as the startup data reached, not a decision about training. It now offers 8 weeks, 16 weeks, 6 months, 1 year and all time, so you can ask the obvious question about a lift: is this better than it was a year ago. The two shorter windows stay instant and work with no signal.

Timeframes are named the same way everywherePicking 6M used to open a menu offering “26 weeks” — the same stretch of time, in a unit nobody thinks in. Long windows now read as months and years, a training block still reads in weeks because that is how you plan one, and anything up to a month reads in days. One deliberate exception: the volume panel still stops at 8 weeks, because the set targets it draws are defined per week and a lifetime average has no target to sit inside.

A note on any exerciseA short line of your own, sitting under the sets where you will actually read it before the first rep. Seat 4, pin 9, wide handle. The stack here is in lb. Left shoulder clicks past 45 degrees, stop there. It belongs to the exercise rather than to the workout, so it follows the movement wherever it turns up in your rotation — the same way the unit and the load step already do. Add or edit it from the ••• menu. Write nothing and there is nothing there: no empty card, no prompt, no row inviting you to fill it in.

The movement section reads at a glanceThe muscles a movement trains are pills now rather than a run-on sentence, so you can count them before you read them, and the one drawn in gold on the figure is the one written on gold underneath. The technique cues are separated by rules instead of bullets. Nothing was rewritten — it is the same words in a shape you can scan between sets.

The demonstration card shows the clipIt used to be a button that told you nothing about what you were about to watch. It now starts closed, and opening it shows the video’s own still with a play control on it. Closed is deliberate: this is the only card in the app that reaches anyone else, so while it stays shut your device contacts nobody at all.

Every exercise shows the muscles it trainsA front and a rear figure at the foot of the screen, with the muscle the movement is for lit up and the ones that only assist lit one step down — the same drawing as the muscle map on your workout, asked about one exercise instead of a whole session. Underneath, two lines on why the mechanics put the load there: the joint angle, the line of resistance, the length the muscle is actually loaded at.

And how to do it — all 317 of themThe setup in one line, three cues, and the one thing that usually goes wrong. Written per movement rather than generated, so a lat pulldown and a reverse-grip pulldown do not get the same paragraph, and pitched at someone who already knows what a working set is. This existed for seven exercises before; it now covers the library.

The exercise screen reads in three partsThe work, the numbers, the movement. The notes used to sit at the bottom of a run of charts, which is the wrong place for the one thing on that screen that does not change between sessions.

Read your volume over four weeks, or eightA rotation of six workouts does not fit inside seven days, so one week is a noisy sample of what you actually do: on the four-year training log this app was built against, one muscle came out at 28, 19, 22, 22 and 21 hard sets over five weeks with nothing about the training having changed — against a target band three sets wide. The four- and eight-week windows are the same measurement with that noise averaged out. Seven days is still the default.

Count the work where a muscle only assistsA bench press does something for your triceps and a row does something for your biceps. There is a window now that counts that at half a set, so you can see what a muscle gets in total rather than only what it gets head-on. It reads higher than the default — the landmark figures were drawn up on that basis, so counting direct sets alone was always the conservative view.

Pick the window on an exercise trendEight weeks, sixteen or twenty-one. A block, a phase, or everything there is — the two charts above it are counted in sessions, so this is the one that answers whether a lift is moving over months.

The rest timer does not always say the same thingTen words now land when the clock runs out. The commit button has worked this way for months, for the same reason: a word you meet twenty times in a session stops registering.

Your weekly volume no longer changes while you sit stillThe window ran from “168 hours ago” rather than from the start of a day, so a session logged seven days and an hour back fell out mid-morning and the count dropped without you lifting anything. It is now a calendar week, so the number only moves when you do.

The same seven days could report two different numbersThe landmark rows counted a set you never rated; the Trends section immediately above them threw it away. Both were on screen at once, disagreeing about the same week. They count it the same way now.

This week is compared against a four-week average, not last weekA rotation of six workouts does not divide into seven days, so a movement lands inside the window twice some weeks and not at all in others. On the four-year training log this was built against, one muscle swung between 19 and 28 sets across weeks that were trained identically — enough to move it two states with nothing having changed. Comparing a noisy week to another noisy week doubles that; comparing it to a settled baseline tells you whether you are actually drifting.

A deload no longer reads as falling behindCutting volume on purpose used to land you a note telling you to add sets to clear your minimum, and then made the week after look like a spike. Deload sessions are out of the measured count. They stay in the 16-week history, where a deload is something you want to be able to see.

Which answer is holding your loads backThe recovery factor is not a score — it is 1.00 plus four terms, one per question. The card draws the four terms, so soreness and sleep sit on the same scale with the same sign and you can see which one is doing the damage. Four averages in four different units, which is what was there before, is a row you look at rather than read.

Sixteen weeks of readiness, not fiveA trend line through your check-ins, with every session under it as a dot. The shading is measured from 1.00 — the point where readiness stops changing your prescription — so the size of the shaded area is the size of the effect and the side it falls on is the direction.

How often you actually answer itEvery other card on the page is only as good as this one, and a skipped check-in is read as exactly 1.00 — it can never quietly cost you a load increase. Pre-workout and creatine get a card of their own that counts adherence and claims nothing else, because neither of them feeds the number.

One timeframe language across the whole pageTrends ran on seven different windows — 7 days, 28 days, 8 weeks, 12 weeks, 16 weeks, and two counted in sessions — and the slot naming them also carried statistics and rules in the same type, so the column could not be read as “how far back”. They group into three by the question each section answers: the dose is a week (the volume landmarks are defined per week, so it cannot be anything else), the block is sixteen weeks, and per-lift sections count sessions, because a lift trained twice a week and one trained fortnightly need the same number of chances rather than the same calendar.

And you can pick it where one appliesThe segmented bar that sat in the middle of the page — governing only what was below it, with a caption explaining that — is gone. The three sections it actually reached carry the control in their own heading now, on the thing it changes.

Two headings were describing the wrong timeframe“Recent form” said twelve weeks while reading your last 30 sessions, which on a normal week is about six — and it was quietly dropping anything older than its cap. “Effort spread” had no timeframe at all when offline: it counted sets from your current rotation, so picking a longer range changed what was counted rather than how far back it looked.

Supplements read by the week, not the sessionSixteen cells, one a week, shaded by how much of that week you took it — on the same week boundaries as the calendar. Per session it was a sliver two points wide once the window grew, and it was making a per-session claim the app deliberately avoids: creatine works over weeks, which is exactly why it is not scored per session.

The coverage squares are the training calendarA square a day over the same sixteen weeks as the calendar higher up the page, at the same size and spacing, with the same wider gap where a month turns. Gold where you answered the check-in, pale where you trained without answering, empty where you rested — so the gaps are the reading, which is what a calendar is for.

The chips at the top of Trends stopped chasing themselvesTapping one used to send the highlight racing through every section the page scrolled past, dragging the strip sideways with it while the chip you pressed slid out from under your finger. A tap now holds the chip you pressed until you scroll. The strip also only moves when it has to, the chips are the size of the rail rather than a third of it, and one of them no longer points at a section that is not there.

Every bar carries its exercise’s codeBBP, IDP, CCF — built from the whole name rather than the front of it, so they tell each other apart. The last word is always in there, because on an exercise name the last word is the movement: Single Arm Cable Row is SAR, never SAC. Past seven exercises the codes come off rather than shrink below the size everything else in the app is held to.

Tap a bar for what you are actually doingThe rep range, the load, the rest, and how many warm-ups come first. The height of a column is its working sets and the code is its name; the prescription was written nowhere on Today until now. The strip stays the same height whether you have tapped it or not, so the card never moves under your thumb.

And the tonnage, under seven exercisesThe one figure on that card that is not another way of saying the set count: twenty-four sets of curls and twenty-four of squats are the same dose in hard sets and nothing like the same afternoon. It is a projection off what is programmed, so it wears a ~.

The muscle map ran off the edge of its cardThe panel was missing the inset every other card in the app has, so the muscle count, the last-trained date and the end of the exercise list were all cut off by the card’s own corner. The legend wrapped for the same reason. Everything now fits on the narrowest phone the app supports.

Smoother lines on the two body figuresThe outlines were thickened with a filter that grows a shape sideways rather than outward, which left a visible staircase on every diagonal — shoulders, outer thighs. They are drawn evenly now. The muscle icons also got back an outline they were missing: a lit muscle used to run straight into the background where the unlit one beside it kept its edge.

How much you moved, at the end of the lineThe percentage across the whole window, and the change in kilos or pounds under it, set as the biggest thing on the card and sitting where the line finishes. It also names what is being tracked — your estimated max, or reps on a bodyweight exercise — because the chart draws two lines and a percentage with nothing against it left you to work out which one it belonged to.

The line is smoother, and the sessions under it are easier to seeThe trend averages more sessions than it did, so the direction of a block comes through instead of week-to-week noise, and it curves between points rather than turning a corner at every session. The dots are bigger — the same size as the ones on the set shape chart below it, which mean the same thing. What smoothing does is only ever to the drawing: the line still passes through every number it was computed from, and it cannot draw a peak higher than anything you actually did.

Two figures on one card that disagreed with each otherFolded up, the panel showed the change between your first and last session. Opened, it showed the change between the two ends of the trend line. Those are different numbers — one is your best day against your worst — so the same card could tell you +20 kg and 61 → 74 kg about the same four months. Everything now comes off the line.

A short history drew a flat lineWith only three sessions logged, the average covered all three at every point, so the answer was the same number three times and the chart drew a perfectly level line over data that plainly was not level. It now uses a narrower average when there is less to work with.

A 16-week trend, under the chart you already hadIt draws an average across your sessions rather than the sessions themselves, because over four months the raw numbers are unreadable — a deload week, a heavy single, a day the rack was busy. Your real sessions stay under the line as dots, so you can always see how much the smoothing did. Folded away by default: this is a question you ask once a block, standing still, not between sets.

Tilt the phone and the picture movesOnly the picture. The numbers, the cards and the buttons stay exactly where they are, and the whole travel is about three millimetres. It is the same parallax your Home Screen moves its wallpaper with, so the photograph leans the same way as everything else on your phone rather than inventing its own idea of level. With Reduce Motion switched on it does not move at all.

More of the picture comes through the cardsThe cards are glass, and glass has nothing to show you unless something is behind it. They now sit lighter on the photograph at the top of each screen, and are unchanged everywhere below it, where there is nothing behind them but the dark. The text on a card was measured against the brightest thing a photograph can put underneath one before the change was made rather than after.

The two round buttons match the screen they sit onThe workout switcher and the ••• at the top of Today were flat discs, left from when there was nothing behind them worth showing. They sit in the one band where the photograph is at full strength, which made them the last flat thing on a screen made of glass. The back button in the readiness check-in went with them.

Say the set and it logsIn whatever order it comes out: log 8 reps at 100 kilos, log 100 kilos for 8 reps, log 8 at 100, log 100 by 8. Say just the reps — log 8 — and it uses the weight already on the row, which after your first set is the weight you actually lifted. Say log it and it takes the set exactly as prescribed, which is most sets. You can name a different exercise on the end too: log 8 reps at 100 on incline bench.

It reads the numbers back to youYou get the exercise, the weight and reps, and how long you are resting — and if the set beat everything you have done on that lift, it says so. That is not politeness. A misheard number looks exactly like a heard one, and there is nothing on a locked screen to contradict it. A weight or a rep count that could not be real is refused out loud rather than quietly rounded into something plausible.

Or one press, with nothing said out loudGyms are loud, and half the time you will not want to talk. The same action sits in Control Center, on either Lock Screen button, and on the Action Button — one press, screen off, phone still in your pocket. It knows which exercise and which set you are on, so there is nothing to choose.

Undo, because a spoken number can come out wrongSay undo that, or press the undo button, which tells you what it is about to take back before you press it. It puts the whole exercise back the way it was rather than just the one row, because logging a set also updates the weight on the sets below it. It works on a set you typed in, too.

The Lock Screen shows the set you are onWhile you are resting, the timer card already covers the Lock Screen. The moment it finishes there was nothing there. A widget now shows the exercise, the weight and which set you are on, and logs it on a tap.

The muscle marks show the supporting muscles as wellA bench press is not only chest, and the mark said it was. The muscle an exercise is for is now in gold and the ones assisting it are in amber, so a glance down your workout tells you what a session is actually loading rather than just what each row is filed under. They are different colours rather than two strengths of one, because helping is a different kind of work and not a smaller amount of the same work.

The figures are bigger, and they are all muscleEach one gave the top of its space to a head, then sat inside a margin, inside a tile that added another. Three lots of padding around a drawing about an inch tall. The head is gone and the figure now fills its tile edge to edge, which is most of a third more mark at the same size on screen.

The exercise you are on glowsThe muscle on the mark for the exercise you are currently in picks up a soft halo — the same one the set you are about to do already wears — so finding your place after putting the phone down is a glance rather than a read.

Plate math shows the weight you are loading now, not the one two rounds awayIt solved for your working set, so three warm-ups deep it was telling you what goes on the bar later while you stood there loading something else. It answers for the set you are about to do now, warm-ups included. The figure is tappable too, so you can step through every set of the exercise and see what each one looks like in plates — and logging a set moves it on by itself.

The coach says why, and says what to aim forTwo paragraphs sat either side of the numbers with nothing saying which was which, so a sentence about what adds weight could turn up directly under a verdict telling you to hold. They are labelled now — Why this and Aim for today — and the second one names the weight your session is playing for instead of describing the rule in the abstract. It also explains the rest target, which was a number on the screen with nothing anywhere saying what it was for.

A pre-workout check-in tells you what it didA check-in that comes back low can hold a load back, and nothing on the screen said so. There is a line for it now, and it says what actually happened rather than implying more than it should: the check-in moves the score, so it can keep a weight where it is, and it never changes your reps or takes a load down.

Loads and reps you can read from the groundThe set list and the logger both went up a size. Exercise names stopped getting cut off with them — BARBELL INCLINE… was as much of the name as a row could hold, and a long one wraps to a third line now instead of ending in dots.

The wait for your plan is about a third shorterEverything the app needs at launch arrives in one request — your rotation, your history, your trends, all of it. The database was doing one piece of permission arithmetic per row it read, which on a four-year log is tens of thousands of times for a single launch. It does it once now. Same data, same checks, and the request came down from about 1.4 seconds to under a second.

The demo stops re-downloading itselfThe fonts, the muscle marks and the icons on /demo were fetched fresh on every visit. They are held now until they actually change, so a second look at a shared link opens on what your browser already has.

Every muscle on the same axisThe minimum, the range worth training in and the ceiling now land in the same place on every row, whatever that muscle’s own numbers are — so the target band is drawn once, behind the list, instead of ten times. The block reads down a column at a glance and takes about half the screen it used to.

Last week, behind this weekA tick where the muscle sat seven days ago and a dot where it sits now, with the move drawn between them. Direction is the reading here: a muscle climbing toward its range and one sliding out of it are the two things a weekly number is worth checking for.

Your muscle’s week, on the exercise screenOpen a lift and you can see where the muscle it trains stands for the week, under the calendar. It also appears when you finish an exercise and on the workout report. The sets you have logged today are shown separately and are never folded into the weekly figure, because that figure was counted before you started.

How volume works, written downA guide beside the coach and rest ones, in settings and on Today’s menu. Why a warm-up counts for nothing, why a drop set counts for nothing, why a single-arm curl counts once and not twice, and why the week is a rolling seven days rather than a calendar one — every place the figure on the chart is not the one you would have counted yourself.

Twenty weeks, a square a dayAt the top of Trends for your whole account, and at the foot of every exercise screen for that one lift. Three cell states on the per-exercise grid — a day you did this movement, a day you trained something else, a rest day — because without the middle one a training block and a layoff draw the same picture.

Two shadings, because they answer different questionsPer exercise it ramps gold by that day’s top working load. Per account it ramps by working sets, because across exercises the heaviest weight of the day is a leg press on Monday and a lateral raise on Thursday — there is no shared scale to ramp. Both ends of the ramp carry their figures, and it rescales to the window rather than to your career.

Where every set lands, for the whole rotationOne row per lift on Trends: how many of its last twelve outings’ working sets fell under the rep range, inside it, or on the top edge. Sorted worst first and capped at ten, and the card says both — the lift that never reaches the top of its range is the one you want to read, and a silent cap reads as “these are all of them”.

Set shape, made readableThe strip that draws each session as a line across its own sets now prints its rep range on the band, dates both ends of its axis, marks every session that swept the top, and flags where the load changed — so double progression working stops rendering as a collapse in reps with nothing saying why.

The gate, with your own numbers in itEvery set at 12 reps · 2 of 3 and Last set at 2 RIR or under · not logged, filling in as you log. A set already in under the target reads as missed rather than pending, because nothing later can undo it — better to know at set two than at the end. The forecast stopped being a verdict handed down afterwards and became something you can still go and hit.

The next weight, with the move beside itThe one panel whose entire subject is next session’s load was the only panel with no load on it. It states it now, with the change against today’s top set, and how far round the rotation this slot comes again — in sessions, never in days, because how many days a rotation takes is a figure the app can only guess at.

A forecast says when it is finalIt appears on your first working set and can move as you go, which read as a fault rather than as a call firming up. 2 OF 3 SETS sits opposite the heading while it is still provisional.

Target, rest and effort, side by sideThe rest target used to live on a button inside the logger and on a row of a menu, and nowhere at all on the screen you read between sets.

Which session the call is forTODAY · PULL A above the verdict, or PLANNED · PUSH B if you arrived from the plan rather than from today. A prescription belongs to a slot in the rotation, and the same screen was showing today’s call and a call for a session you are not in with nothing to tell them apart.

Why these numbersA sheet, for that one exercise: what today asks for, where the load came from, your last session, the score with both of its terms spelled out, where the rep range came from and whether you set it or the app did, and what the increment is. The general guides are correct and general, which is exactly why neither could say why this lift is asking for 132 kg today.

A session preview instead of a stat barA workout used to open on 6 exercises · 18 working sets · 4 muscles, and a count of muscles is the same number whether the session gives your chest nine sets and your triceps one or the other way round. It now shows hard sets per muscle — the unit the whole volume model is defined in — and an estimated length that says it is an estimate.

Every page title at one sizeThe same workout name was set at 38 points on one tab and 76 on the next, and the smaller of the two was the screen you stand in. They all moved up to match, and a display title that wraps now takes a set leading rather than most of a line of dead air between two rows of capitals.

A silhouette behind the exercise nameCrushed to a shape and faded into the background before the first card, with the face deliberately past legibility — a readable face makes the picture about somebody rather than about a movement. Where a photograph of that exact movement exists you get it, framed on the muscle or the lift rather than on the person; every other exercise draws one of a set of gym pictures instead, so nothing opens on plain ground and the same one does not come round often.

Technique notes, where there are anyWhat the movement works and what only assists, the setup in one line, and two or three cues. Never more: a list long enough to need scanning is one nobody reads between sets. An exercise with nothing written for it shows no block at all rather than a placeholder — invented cues and real ones look identical to the person reading them.

Plate math, above the coachA whole loaded barbell for barbell and Smith lifts, solved in that exercise’s own unit, with every plate carrying its own number. The bar’s own weight is a tag on the shaft and tapping it cycles, because which bar is on the rack is a fact about your gym. What the plate set cannot make exactly is reported, never quietly rounded away.

Numbers you can read from the floorThe load and reps on a set row went from 27 to 33 points, and the exercise name is set at signage scale — both were smaller than the heading above them, on the one figure you have to read standing over a bar.

Which row you are standing onThe set you are about to do is marked three ways at once — full opacity while the rows around it fade, the load and reps in gold, and a play mark in the slot a logged row puts its checkmark in, so that column reads as one state column top to bottom. It is a mark, not a control: tapping the row already opens the logger.

Panels fold, and stay how you left themTap a panel’s label pill and it collapses to that pill plus the figure beside it. Across exercises, across launches. If you already know what 132 kg looks like on a bar, you should not have to shut that panel twice.

Pick up exactly where you wereTicked sets, the exercise you were on, your check-in answers — and, more important than any of it, the same session, so the rest of your training is not filed as a second one. It restores for up to three hours, which is the same point at which the server decides a session was abandoned, so the two can never disagree about whether you are still training.

A session you forgot to finish closes at your last setNever at the moment somebody noticed. Stamping the discovery would invent a duration out of how long the phone stayed in a pocket. It reads slightly short of a session you ended by hand, and short and true beats long and invented. A session with no sets in it is deleted rather than counted, because a zero-length workout would quietly rescale every weekly volume figure you have.

A nudge if you walk offOne notification after half an hour with nothing logged. The threshold is measured rather than picked: across the sessions carrying real per-set timing, the longest gap between two logged sets is six minutes. It cannot end a session and does not try.

A skipped exercise is a factThe log used to record only what you did, so a four-exercise session and a six-exercise session you walked out of two thirds through were the same row. Every planned exercise is sent now, with what was asked of it frozen alongside — and the report you read back weeks later can show what you did not do.

A cut lands on the stackA backed-off machine load used to come back at 188.5 lb, which a 10 lb stack does not make. Cuts come down a whole number of pin positions from what you lifted and land on a real weight. A load too light for the rule is deliberately left alone — a 20 lb machine has nothing between 20 and 10, and forcing a 5% cut onto that stack would turn it into 50%.

An increase adds one increment to the weight you liftedIt used to round onto a grid counted up from zero, which is a grid no gym is on: a pulldown at 145 lb was sent to 160 rather than 155, and a lateral raise was offered a jump of 3.66 against a step of 5. The load you just lifted is loadable by definition, which is better evidence about your gym than any table.

A comma is a decimal point, not noiseA number pad draws its separator from the phone’s own locale and there is no way to ask it for a point, so half the world can only type 32,5. That was reaching the log as 325. Grouping and decimals are now read the way a human types them, in any locale, including an Arabic-Indic keypad.

A held load stopped carrying conversion crumbsImported history is pound-shaped — 121 lb is 54.885 kg — so a lift the coach was holding prescribed 54.9 kg every session, on a rack whose finest plate is 1.25. Every figure the engine hands forward is tidied in the unit you are reading it in. Nothing stored is touched: a set reads back exactly as you typed it.

Kilograms are the stored truthPounds are computed at render, which is what makes the two units one log rather than two roundings of it. Switching a lift to pounds changes what you read and never what was recorded, and four years of history follows the switch, halves included.

One answer per exercise, not one per slotThe coach used to compare a session only against past sessions from the same workout. That protects something real — a movement run heavy on one day and for volume on another should not read as a collapse — but the workout was standing in for the rep range, and on most plans both slots chase the same range. Comparison is by rep range now. A 4–5 day and a 6–9 day stay apart exactly as before.

It was skipping the session that mattered mostOne bench session was scored against another three weeks earlier, because that was the last one in the same workout, instead of against the one three days earlier at the same weight for eight fewer reps. It concluded you had gone backwards and cut 7.5% off a lift that had gone up.

The exercise screen tells one storyThe weight at the top, the Prev line, the chart and the verdict were on two different footings — the prescription counted one set of past sessions and everything explaining it counted another. That is how 220 lb could sit directly above Prev 209.4 lb × 10 · 3 days ago with neither number wrong.

Every loaded lift charts an estimated maxMachines and isolation work used to chart the weight alone, and the weight is the one variable double progression deliberately holds still — so the bars sat flat for weeks while your reps climbed, then stepped on the session the load reset. One machine row drew its lowest bar in ten weeks on what was its best session in the window and its fourth improvement in a row.

The exercise row, on a fixed gridThe number column used to size itself to whichever was wider, the load or the row of set markers — and there is one marker per set, so a three-set exercise and a six-set one handed the name completely different amounts of room. That is why names broke in a different place on every row. Every load on the screen sits on one line now.

Rows that grabbed and would not openA thumb resting on a row before a flick armed a reorder, and arming switched the scroller off — so the screen simply stopped moving. The same press outranked the row’s own tap, so nothing opened either. Every fix aimed at the timing broke the next one; the answer was to stop hand-rolling it and take the platform’s.

What a swipe offers depends on where the set isThe last row carries Add set. The first working set carries Warm-up, and the last warm-up carries Working. Everything between is delete alone, because warm-ups are contiguous at the top of a list and flipping a row in the middle would strand one under three working sets.

Where you are is where you last put a set downStep over an exercise because the rack is busy and everything that asks where you are used to point at the one you skipped — the button, the session rail, the breathing markers. It follows your newest logged set now, and picks anything stepped over back up at the end rather than dropping it off.

Reorder reaches every accessibility routeA drag is not an interface for VoiceOver, Voice Control or Switch Control, so both lists carry Move up and Move down as real actions, with an announcement of where the row landed.

An alarm a silenced phone still makesA notification is a notification: flick the silent switch on the way in, or run a gym Focus, and the one thing the app has to say is the one thing it cannot. It uses iOS 26’s alarm API, which sounds through both. Exactly one alarm is ever armed for a rest, and the notification stays underneath as the fallback — switch the alarm off and everything behaves as it did.

The last five seconds, out loud and in your handFive ticks climbing a scale and a stab at zero, built from detuned oscillators with a filter closing on every hit and a small room under it — because five identical beeps is the noise every kitchen timer has made since 1985, and on a gym floor you stop hearing it. Five haptic taps under the five you hear, getting stronger and sharper, scheduled as one piece so the two cannot drift.

The taps mean different things nowBanking a set and a rest ending used to be literally the same system buzz, so from a pocket you could not tell which had happened. Banking a set is one blunt press. Rest ending swells into two sharp hits, because it is the one you might not be looking at.

Your heart rate, inside the dialFrom an Apple Watch or AirPods Pro 3, beating at whatever rate it is reading. Live is red; a reading that has gone quiet turns grey, stops beating and shows its age. It is read, drawn and thrown away — nothing is stored, nothing is sent, and the coach cannot see it. A heavy set is not an aerobic event.

The card hands you back to the exerciseREADY holds for a beat, the timer folds away, and you land on the set list you were going to reach for anyway. No tap.

A Lock Screen card and a Dynamic IslandTicked by the system rather than by the app, so it is correct with the app suspended, backgrounded or force-quit. Both controls — fifteen seconds either way, and Skip — work without unlocking.

Rest separates rounds, not setsA working set with a drop hanging off it used to start a countdown at the exact moment you were meant to be stripping the bar, and the drop itself started nothing. That near-zero gap is the technique. The clock belongs after the drop, at the exercise’s own target — and it is deliberately not extended for it, because nothing puts a number on what a past-failure set adds.

A megabyte the app was carrying and never usedThe database library ships as one piece and this app uses two parts of it — signing in, and calling the handful of functions everything runs on. Realtime subscriptions, file storage and edge functions were being built into every copy for nothing. Linking only the two takes the code down 18%.

Offline is your own last payloadCached whole, written only once it decodes, keyed on your account and deleted on sign-out. A phone with no signal comes up on your own plan instantly rather than on a placeholder.

The launch decode went 266 ms to 6 msNine tenths of it was building a date formatter for every timestamp in the payload. Hand-parsing is 22 ms, it runs off the main thread now, and the payload itself dropped by roughly two thirds by not sending this client the part of your log it throws away.

Sign in with a six-digit codeIn a designed email rather than the stock one, with the code as the hero and the link under a rule. A magic link cannot come back into a native app at all, and the code never leaves the screen you are on.

The screen does not sleep while the app is openThe phone is on the floor next to the rack and the point of that screen is being readable from standing. There is a switch, because it is real battery over ninety minutes.

How rest worksIt exists because the behaviour is surprising by design. You log a set and no clock appears, and an app that silently does nothing where it usually does something reads as broken rather than as deliberate.

Settings that belong to the rack, not the accountUnit, load step and rep range, per exercise — gyms are mixed, and a kilo plate-loaded rack sits next to pound dumbbells. Every row says where its value came from, so “machine default” is never mistaken for a number somebody chose.

Trends over a real yearThe long ranges used to redraw the same 40 days of cache spread across empty month buckets, which reads as a plateau on the one screen you would use to decide whether a block worked. Long windows are computed on the server against your whole log now, and every state that is not — loading, offline, nothing in this window — says which one it is.

How the coach worksThe loop, the binding gate, the two ways weight comes off, the rep ranges and rest targets per class, the effort ladder, what the check-in can and cannot do — and a section on what the engine deliberately does not know, which is the list every app in this category leaves off its own page.

The app learns how often you trainA plan is an ordered rotation, not a Monday-to-Sunday week. Six workouts trained daily is 1.17 rotations a week, so reading the rotation total straight understates every muscle by enough to hide a group sitting past what it can recover from. It counts your last four weeks, and it says whether the number is yours or its own guess.

Plan checkWeekly hard sets per muscle against the three volume landmarks, projected off the template rather than measured off the log — so a plan can be checked the day it is written, by somebody with no history at all. Every muscle is listed, zeros included: a muscle you have not trained cannot appear in a count of what you trained.

24 of 47 lifts were prescribing a cutThe rep ratio was measured against the top of the range — and the bottom of a range is exactly where double progression puts you the session after every increase, so correct execution and failure scored the same number. Across 598 evaluations of the four-year test log, 300 fired the reduce branch and 171 of those were sessions whose reps sat inside the prescribed range. It is position within the range now.

Adding weight was being read as going backwardsReps fall when the load rises. That is the mechanism, not a failure of it. Every backward comparison now requires the weight not to have gone up — without it the engine punishes the exact session it just asked for.

Three of the four score terms were constantsOne of them had a single distinct value across every evaluation ever made, because nothing anywhere collected it. The score was one term wearing a four-term product’s clothes. It is two terms now, and the guide says which.

Volume became a leverEvery action the engine could take moved load, and hard sets per muscle per week is the stronger one. Your weekly sets are read against the three landmarks and it names the exercise that should absorb the next one — always a lift it is already progressing, never one it is backing off.

A deload needs two reasons, and at most one can be the calendarA week dropped into the middle of a block purely because it was due produced no growth benefit and measurably cost strength. A due date will suggest one; it will never announce one.

Four years of history, importedThe app was built against a real training log rather than a seed — around 13,000 sets — which is what every measurement quoted on this page was taken against, and the reason a rule that looked fine on paper could be shown to be wrong across 300 sessions.

Sets survive the basementWrites go into a local queue and return immediately — nothing waits on a request to consider a set logged. What is queued is a whole session rather than a single set, so retries are free and the queue collapses to one entry per workout.

Units follow the exerciseOn top of an account default, because a gym is mixed. The choice travels with the movement across the rotation, so a machine you read in pounds stays in pounds wherever it turns up.

Supersets and drop setsA drop set is real work and belongs in your tonnage. It is not another hard set — it extends the set it hangs off, and counting it as one would inflate the weekly number that decides next week’s volume. A superset’s whole mechanism is the rest, so the clock starts after the last member and not before.

Correct a logged set, delete a sessionThe session is re-scored, so a rep count the coach has already turned into a prescription stops being wrong. Both ride the same offline queue as everything else.