The record · the method

How much does one person get done?

We wanted an honest number to stand under “as productive as two people.” Here is what the research says people plan, what they actually finish, where our reference day comes from, and where it is weakest. Every source linked, and the numbers we refused named.

A convention, publishedEvery figure sourcedThe refused numbers named
The (i) in the app

How this is counted

Kept counts what you meant to do and did: habits kept, tasks done, focus finished. One person's day, for this number, is five kept things · a convention we publish, not a law. Your week divided by that is your number. Anything a connection or an agent marks done counts only after you accept it. A count, not a score.

Reference day v1

1 · The numbers everyone quotes

Where the usual figures come from.

You have probably seen “41% of to-do items are never completed.” It comes from one company’s blog post about its own users, with no sample size or method, and the original page no longer exists. You may have seen “people complete about 20% of what they plan.” We looked hard and could not find a source; the nearest thing is the Pareto rule, which is about effort, not completion. And “10x engineers” descend from a 1968 experiment on twelve programmers designed to compare two ways of working, not to rank people. We are not printing any of those as facts. We are printing the ones below, with what each one measures.

2 · What the research shows

A rate of about half. Not a daily count.

Across many studies, intentions turn into action about half the time. That is the conclusion of a 2016 review of the intention-behavior gap. Read it carefully and the half is a proportion of people, measured on health behaviours like exercise and screening, weeks to months after the intention was formed. It is not a rate for a day’s to-do list. The closest thing to that is a company survey of two thousand professionals, self-reported, in which people finished about 53.5% of the tasks they planned each week. When people set their own deadlines, they miss them more often than not: honors students expected to finish a thesis in 33.9 days and took 55.5; 29.7% finished by their own best estimate. On everyday one-to-two-week projects, 37% to 44% finished on time.

Time tells the same story from the other side. Full-time workers in the United States put in 8.1 hours on the days they work. Vendor telemetry suggests under three of those hours are uninterrupted. One focused hour is a lot.

Habits are slow. In a real-world study, a new daily behaviour took about 66 days to feel automatic, and the spread ran from 18 days to a forecast of 254. A record measured in weeks is too short. Ours is a year.

And a consecutive-day count changes behaviour on its own. When a log highlights an intact run, people keep going; when it highlights a broken one, they stop, and the effect depends on how the log shows it, not on what they actually did. That is why the record has no consecutive-day counter. A quiet day is a day.

3 · How we set the reference day

Five kept things. A convention, not a finding.

There is no research-grade count of how many things an average person finishes in a day. So we declare one, and we show where it comes from. Todoist sets its default daily goal at 5 completed tasks. That is a convention, not a finding, and it lives inside a consecutive-day and points system we chose not to build. We borrow the number and leave the rest. One reference day is 5 kept things; one reference week is 35. This is Reference Day v1. If we had used a plan count instead, the research above says it would be roughly twice as large; we count kept things, so we use the kept count. If the reference were four, a forty-two-thing week would read 1.5 instead of 1.2; at six, 1.0. If it changes, this page says when and why. It never moves silently.

4 · How your number is built

Kept intentions, counted, against one week.

Kept counts what you meant to do and did: a habit logged, a task done, a focus session finished, and, when you have connections or agents, a task they finished that you accepted. Meals, notes, and workouts are activity; they show on the day, not in the number. We count up to 10 habits, 12 tasks, and 6 focus sessions a day toward the number, and up to 12 things a day that others did for you, so a task split into ten is not ten and a hundred one-tap accepts are not a hundred. The grid and the day always show every one.

Your number is the last seven days of kept things divided by 35. At one or above it reads “as productive as 1.2 people.” Under one, we just show the count. Anything a connection or an agent marks done waits for you; it counts only after you accept it, and the split between you and them is always shown. A person alone tops out at 5.6; with agents, 8.0.

Every kept thing is placed on the day your device said it was, in your timezone at the moment you did it. Before September 2026 Kept did not store the moment a task was finished, so older task days are placed by the task’s last edit and marked approximate when you open them.

person(day)  = min(habits you kept, 10) + min(tasks you did, 12) + min(focus, 6)
agents(day)  = min(habits accepted + tasks accepted, 12)
units(day)   = person(day) + agents(day)
index(week)  = sum of units over the trailing 7 days / 35
Caps apply to the number, not to the tallies. Below 1.0 the headline is the count. The multiple never reads 0.0, and “You are” never precedes it.
Worked days under Reference Day v1. Units are capped; kept is the uncapped tally.
DayRowsUnitsKeptDay indexWhat the day sheet says
Quiet1 habit110.2Monday · 1 kept (no multiple below 1.0)
Normal3 habits + 2 tasks + 1 focus661.2Tuesday · 6 kept · the work of 1.2 people
Heavy5 habits + 14 tasks + 3 focus20224Thursday · 22 kept · the work of 4 people
Agent-heavy4 habits + 3 tasks + 1 focus + 42 accepted agent tasks + 3 waiting20504Friday · 50 kept · the work of 4 people · your agents 2.4 of it · 3 waiting for you

A 42-unit week reads 1.2. A 12-unit week reads “12 kept this week.”

How the count stays honest

  • A task split into ten is not ten. Tasks count toward the number up to 12 a day. Tallies and the day sheet always show all of them.
  • Subtasks never count. They live in their own table. The parent counts once.
  • A habit counts once a day. Keyed on the habit and the day, the database's own uniqueness, never on a row id.
  • Focus sessions count up to 6 a day. And only when the session finished with minutes above zero.
  • Workouts, meals, notes, and metrics never enter the number. If they are promises, they are habits. They can still show as activity on the day.
  • Reopen removes; redo restores once. One unit per task at its current day. Un-keeping a habit deletes the row; re-keeping writes a fresh one. Set semantics, never an append log.
  • Deleted rows leave the record. The ledger derives from rows that exist.
  • A connection's or agent's completion counts only after you accept it. Up to 12 a day. The label comes from the grant, never from model text. Acceptance is a write only you can make.
  • No writer can raise its own number. Connections and agents see counts, never the reference or the multiple.
5 · What this number is not

A count, not a score.

It is not a measure of your worth. It does not know what your work cost you. It counts what you wrote down, so a person who plans less will look smaller here than they are. The half-of-intentions figure above is measured over weeks and months, not over a single day; we cite it as the best available context and say so. It is not a comparison with Kept’s other users; we have not measured them. It is a count, not a score. Nothing here goes red, and nothing here counts down against you.

6 · Sources

Every figure, with a working URL.

16 linked sources. Hosts that block automated checks carry a dated manual check. This page says what each number measures and what it does not.

Peer-reviewed and federal (verified in the primary text)

Lally, P., van Jaarsveld, C. H. M., Potts, H. W. W. & Wardle, J. (2010)

How are habits formed. European Journal of Social Psychology, 40, 998-1009.

https://doi.org/10.1002/ejsp.674

Silverman, J. & Barasch, A. (2023)

On or off track: how (broken) streaks affect consumer decisions. Journal of Consumer Research, 49(6), 1095-1117.

https://doi.org/10.1093/jcr/ucac029 · publisher blocks automated checks; abstract checked by hand 2026-09-05

Peer-reviewed (link re-checked before publish)

Surveys and telemetry (a company's data; method partly stated)

Conventions

Numbers we refused

Refused

"20% of planned tasks completed": no source found.

Refused

"41% never completed": company blog, no method, original page gone.

Refused

"2 h 53 min productive per day": marketing survey, primary unavailable.

Refused

"Five intentions with half kept = 2.5": our own first draft. The five is a completion goal, not a plan count, and the half is not a daily rate. We caught it before publishing.

Changes · v1 · September 2026 · Reference day 5 kept things. First publication.