Your head is for decisions, not for remembering steps.

Every process kept in memory (which step is next, what is due, what happened last time) takes attention the work itself needs. The method hands the holding to the machine and leaves the doing and the deciding with you.

In the app

• The Routine holds the order. The Run holds where you are and what is due. Triggers hold when.
• A timer tells you. It does not decide you have stopped.
• Deleted things wait 30 days in Recently Deleted, and finished runs are kept.

What the research says

• People hand memory to notes and tools, and do more of it as demands rise (Risko & Gilbert 2016). In experiments with 1,196 people, reminders improved performance (Gilbert 2015).
• Offloading frees the mind only when the store is trusted (Storm & Stone 2015).

A routine is the way something is done. A run is one time it was done. A routine can have one run in its life (a move) or hundreds (every morning), and several runs can be going at once, each on its own step.

In the app

• A Run starts from a Routine and keeps its own record: what was done, skipped, noted and changed.
• Several runs of one routine can go at once, each with its own name.
• Edit the routine and the runs going see the edit. A finished run keeps the routine as it read it.

What the research says

• Research on organisational routines draws the same line: the pattern people refer to, and each performance of it, which in turn reshapes the pattern (Feldman & Pentland 2003).
• Process software separates a template from its cases, and lets a case depart from the template (van der Aalst et al. 2005).

Start with what you can do now. A routine is not planned up front and then obeyed; it begins from the steps you already do, and it is grown: each run does what it can, whatever the run changed gets a short answer, and what should happen again goes into the next run. Small improvements, made by the person doing the work, every time through: kaizen, for one person's routine.

One step first; split it when a part needs a life of its own. Something new starts as a single step, and what it needs goes into that step as a list: "Pack for the trip", with the things to take written underneath. A part becomes a step of its own only once it turns out to need what only a step has: its own time or timer, its own place in the order, being put off on its own, being done by the machine, or a record of its own. Until then one step with a list is lighter than eight steps, and the runs show which lines have earned more. It is the other half of principle 18: a step is split when a part earns it, and folded back when it no longer needs spelling out.

In the app

• Anything can be changed for this run: a step's words, its duration, a new step, a note.
• Each change is marked orange and counted in the band at the top. Keep carries it up to the routine. Only here says it belonged to today, and the mark goes grey.
• One screen lists every unanswered change, so the count can always reach nought.
• A step holds a description, where such a list lives, and notes, values and links beside it. A part becomes its own step as a change in a run, answered Keep like any other.

What the research says

• A four-year field study found routines change through people's reflections on earlier rounds (Feldman 2000).
• A meta-analysis of 46 samples found debriefs and after-action reviews improved performance by about 25% on average (Tannenbaum & Cerasoli 2013), and more when a record helps people remember (Keiser & Arthur 2021).
• Kaizen, Toyota's practice of small improvements made by the people who do the work, treats each written standard as a hypothesis every doing tests (Imai 1986; Spear & Bowen 1999).

A routine gives a default order, not a script. Most steps do not depend on each other's order, so a run may take them in the order today allows. Where order truly matters, the routine says so, and there it holds.

In the app

• Later and At the end move a step for this run, and steps can be reordered or added for this run only.
• Every new run starts from the routine's order.
• Where order matters, the author can switch off the buttons that move a step.

What the research says

• In invoice processing at four organisations, one routine produced hundreds of different patterns over five months (Pentland et al. 2011).
• Case-handling software focuses on what can be done rather than what must be done (van der Aalst et al. 2005).

When one run finds something better and it is kept, every run of that routine still going picks it up, for the steps it has not reached yet. What a run has already done, or said for itself, is never rewritten.

In the app

• A run reads its steps from the routine as it is now and stores only how it differs, so a kept change reaches every run still going.
• A run's own changes win for that run.
• A finished run keeps the routine as it read it.

What the research says

• Knowledge built into tools and routines travels more easily than knowledge held by people (Argote & Ingram 2000).
• In 36 pizza stores, lessons spread to stores with the same owner, and learning by doing faded fast (Darr et al. 1995).
• Moving a running case onto a changed process carelessly can skip or repeat work (van der Aalst 2001).

People hold far fewer things in mind than they think, and every open thing costs attention even while nobody works on it. So only what is going stays in view, and you choose what that is.

In the app

• A going run sits on the Lock Screen and in the Dynamic Island as a Live Activity. The system shows only a few at a time.
• The bar at the foot of the app answers one question: what is going.
• An Inbox waiting for more leaves the bar until something arrives.

What the research says

• Adults hold about four chunks in the focus of attention, not seven or nine (Cowan 2001).
• Switching tasks costs time and errors, and having several tasks in play slows each one even without a switch (Monsell 2003).
• An unfinished task leaves attention behind that lowers performance on the next (Leroy 2009).

Inside a run you see the step you are on, not the list. A whole list is load that does nothing for the step in hand, and it tempts you to pick what is quickest to close instead of what comes next.

In the app

• A run's page shows the step it is on. The list of steps is one tap away.
• Moving on is deliberate: Next, or Later and At the end to put the step elsewhere on purpose.
• Put the phone away mid-step, and the run opens on the same step.

What the research says

• Asked to carry one of two buckets, people picked the nearer one and carried it further, to finish sooner (Rosenbaum et al. 2014).
• People chose urgent tasks that paid less over important ones that paid more (Zhu, Yang & Hsee 2018).
• On flight decks, checklist items get skipped when the crew is interrupted (Degani & Wiener 1993).

The unit of work is a step, not a box to tick. A box needs a measure of done, much of what matters has none, and a list of boxes rewards ticking. So a step is done by moving on, it fills with context, and a follow-up is another step with a when.

Done elsewhere is done. The job is to hand the remembering to the machine, not to tick a box and feel good about it. If you finish something that is also a step somewhere in a routine, you do not go looking for where it lives, and you do not try to make it the current step. You carry on with the step in hand. When the finished step comes up, you walk past it. The record may then say less than a report would want, and that is the trade: the method is after flow, not after managing a process for its own sake. So the same step in several routines or runs is fine. Walking past a duplicate is cheaper than keeping the system tidy.

A score, if there is one, comes from outside. A score is not forbidden. A score built on your own decisions is: ten tasks made and ten ticked rewards empty tasks, rushed work and easy tasks picked over the ones that matter. A score can only count what effort aimed at the number cannot produce. The test is one question: can this number go up without the real thing getting better? If it can, it is not a score. So the grader is outside you: a release that passed review, a customer who replied or paid, a milestone whose end somebody or something else confirms. What counts as done is decided once, when the milestone is set, so nobody has to judge success each time. And a result counts when it is above your own recent level, so repeating the same one stops paying.

In the app

• No score, no streak, no percentage.
• A step already done elsewhere is walked past with Next when it comes up. Nothing asks where it was done.
• A step collects notes and values while in hand, and the next steps read them.
• The one number a run has counts questions to answer, never achievements.
• If you want a count, a value on a step can keep it, by hand or by an automation. It never becomes a streak.

What the research says

• A streak is a result, not a goal: in a well-built process it happens by itself. People with high self-control reach good outcomes mostly through habits, not effortful resistance (Galla & Duckworth 2015).
• In two years of emergency-department data, doctors under heavy load picked easier cases, which raised throughput then and lowered productivity over time (KC et al. 2020).
• A number that becomes the goal stops describing the work (Campbell 1979).
• A specific plan for an unfinished goal quieted it much as finishing would (Masicampo & Baumeister 2011).
• Reward systems often pay for one behaviour while hoping for another, and people do what is paid (Kerr 1975).
• Across 128 experiments, tangible rewards for doing a task lowered interest in it, while positive, informative feedback raised it (Deci, Koestner & Ryan 1999).

When something happens (a time, a place, a Focus, a message from another system), a run starts or a step moves, and a step may do its own part. You decide the "if" once, and stop carrying it.

In the app

• A Trigger starts a run or moves a step: a time, a place, a Focus, a workout, another app, an event from outside.
• An Action is what a step does by itself: run a shortcut, set a Home scene, write a value.
• A trigger can carry conditions, such as weekdays only.

What the research says

• A plan tying a situation to a response is carried out far more often than a goal alone. A meta-analysis of 94 tests found d = 0.65 (Gollwitzer & Sheeran 2006), and one of 642 tests found larger effects for plans written as if-then (Sheeran et al. 2025).
• Formed habits are set off by context, not by a goal (Wood & Neal 2007).

Some things have no moment to wait for: the appraisal, the parcel, the reply. Nothing tells you it has happened, so you keep checking in your head. A check-in asks a short question every so often, and the answer ends the asking and moves the run.

In the app

• A check-in is a trigger with a question, on a run or a step, asking every so many minutes, hours or days, with its answers beside Not yet.
• Each new card replaces the last. An answer takes its cards away.
• While a step's check-in waits, the step goes to the end, so the run is not stuck on it.
• A run's own check-in holds the run while it asks, and going again answers it.

What the research says

• Holding an intention to act on a future event slowed people's other work before the event came (Smith 2003).
• In randomised trials at three banks, reminders raised savings by about 6% (Karlan et al. 2016).

Many deadlines are not dates: "ten days after the offer is accepted", "the evening before the flight". The event is not known when the routine is written, and it moves. So a step's moment can be counted from a date the run carries, and moves when that date moves.

In the app

• A step can be due by a time, or start at a time, counted from a date the run or routine carries, plus or minus an offset.
• By default, anything with a date is mentioned a day before, and when late it asks rather than being skipped.
• These deadlines can go into your calendar or Reminders, and are written again when the date moves.
• Alerts on your own calendar event can drive a run: start a routine, bring the run to the step for that stretch, or start another routine. They are counted from the event, so moving the event moves the plan. This needs an iPhone on iOS 27 with the Routine Run shortcut set up.

What the research says

• The critical path method made every activity hang on what it depends on, so a change upstream moves everything after it (Kelley & Walker 1959).
• Students predicted their thesis would take about 34 days. It took about 56, longer than their worst case (Buehler et al. 1994).

A step that cannot be done now is not skipped and not left hanging. It is put somewhere it will come back from, and there are as many ways to say where as there are real reasons for not now.

In the app

• From Not now… on the step, Later moves it one place on and At the end after everything else. Pause holds the whole run.
• A step can be put off to a time or a place, and comes back then or there.
• Decide later leaves it in its place with a quiet question on it, and the run comes back to it at the end.
• Nothing is deleted. The record says what happened.

What the research says

• Intentions tied to an event are remembered more reliably than ones tied to a clock time (McDaniel & Einstein 2000).
• A short plan for how and when to come back to an interrupted task removed the attention it left behind (Leroy & Glomb 2018).

On the go, the only job is to catch what comes up and keep moving: one tap, no decision. In time set aside, what was caught gets sorted. Mixing the two means deciding badly on the pavement, or not catching things at all.

In the app

• A routine can stay open: out of steps, it waits for more. Inbox, To read and To watch come this way, and whatever is sent to them from the share sheet or Shortcuts lands as a step.
• What arrives is answered Only here as it arrives, and the list can say once a day how much is waiting.

What the research says

• Getting Things Done separates capturing from clarifying, with a weekly review (Allen 2001).
• Batching phone notifications three times a day lowered stress (Fitz et al. 2019), and so did checking email three times a day (Kushlev & Dunn 2015).

The cheapest interruption is the one that never arrives. Ignoring a notification still costs attention, so the method narrows what can reach you to what this part of the day is for, and gathers the rest.

In the app

• A Focus filter tells the iPhone which routines' cards may come through while that Focus is on.
• Each Focus can say which routines may start, and whether they need a yes.
• Repeats fold into one interruption.

What the research says

• Receiving a notification, without answering it, disrupted performance on an attention task (Stothart et al. 2015).
• In a study of 221 people, a week with alerts on brought more self-reported inattention (Kushlev, Proulx & Dunn 2016).
• Batching notifications lowered stress; turning them all off raised anxiety (Fitz et al. 2019).

A checklist is a score, played as written, and that is its virtue. A routine is a tune, played a little differently each time, and what worked can join the tune. The method is built around the conditions for flow: one clear next thing, feedback straight away, room to play. The method teaches flow and flexibility, not obedience to a habit. "Be water", as Bruce Lee put it: the habit does not run the mind; you choose to do what you are used to.

In the app

• The routine is the tune. Each run is how it is played today: steps moved with Later and At the end, changed or added for this run.
• What was played well can join the tune with Keep. What was only today's gets Only here.

What the research says

• Organisation researchers treat improvisation as a matter of degree, never as the absence of structure (Weick 1998; Barrett 1998).
• Flow comes with challenges that stretch skills without overwhelming them, clear near goals and immediate feedback (Nakamura & Csikszentmihalyi 2002, Handbook of Positive Psychology).

Whether a step is done by a person, a script or a software agent, the machine does what can be undone, and the person decides what cannot.

In the app

• What can be undone is not asked about: Stop and Delete put up an undo bar.
• An event from another system can be made to wait for a yes, and a journal records who pressed what.
• An assistant writing a routine cannot add webhooks or Home scenes. In the app, everything is shown before it is saved; asked through Siri or Shortcuts, it saves at once and keeps the earlier version, which one Undo puts back.

What the research says

• The level of automation should be chosen per function, weighing reliability against the cost of a wrong action (Parasuraman et al. 2000).
• Across 18 experiments, more automation meant easier routine work and worse performance when it failed (Onnasch et al. 2014).

There are two ways to be serious about time: fill it well, or win it back. Routine Run is for the second, and there are two ways in. One is organising what already fills your days, so the evening becomes free. The other is a new habit, built the Kaizen way: plan one action, run it, and add to it in the runs until it has become a routine.

In the app

• A new habit starts as a routine of one step. A run can add the next one, answered Keep, until the routine is the whole habit (3).
• Not a planner that lays out your day. It adds to the calendar and Reminders you already use, and does what they cannot: the walk from one thing to the next, which none of them holds (between).
• Not a tracker of streaks. A skipped step is information, not a broken chain.
• Not an autopilot. It asks before what cannot be undone.

What the research says

• Who it is for is a position, not a finding.
• Research on ADHD describes it as trouble doing what one knows rather than knowing what to do, and the help it describes is information and time made visible where the work happens (Barkley).
• Highlighting a streak that broke lowered later engagement, across seven studies (Silverman & Barasch 2023).

The guide steps back. Every step starts guided: read it, then do it. Run by run, while the routine is being run, each step can move one of two ways: to habit, where it no longer needs spelling out, or to the machine, where the app does it. Habit matters more: a step the machine does is one you stop practising. What cannot be undone stays with you on either road. Habit is there to free attention, not to take the decision away; and the next habit is added once the last one runs by itself, since adding too early can break the old one without forming the new.

In the app

• To habit: start a routine with one or two steps you can do, and add the next once the first is done without thinking. Cut a step down to its name, move its details into a note, or fold several steps into one, by hand or by describing the change to an assistant. Keep writes that into the routine. In a run's list, done steps fold into one row.
• To the machine: a step's action is set off Only when I press it, or When the run arrives. A step's timer can Move on by itself when it runs out.
• Every run records what was done and what was skipped.

What the research says

• Good teaching support is meant to be withdrawn as the learner takes over (van de Pol et al. 2010).
• Guidance a beginner needs can slow down someone experienced (Kalyuga et al. 2003).
• Reminder-based apps supported repetition but got in the way of the habit forming (Stawarz et al. 2015).
• Atomic Habits asks for a process small enough to do today, made a habit, then the next, and for hanging each new habit on one you already have (Clear 2018). A popular book; it is the everyday form of the if-then research on principle 9.
• The level of automation is best chosen per function, weighing reliability against the cost of a wrong act (Parasuraman et al. 2000), and reliance should follow what the automation has shown it can do (Lee & See 2004).