How Lighter Journeys scores effort

The effort unit

1.0 effort unit is typing a short entry on a computer: about 10 keystrokes, about 4 seconds of focused work. In Keystroke-Level Model terms: homing 0.4 s + mental preparation 1.35 s + 10 keystrokes at 0.233 s = 4.08 s.

Each action a session contains adds its cost. The effort of doing a task is the sum, so a report says, for example, "36 effort units, about 2 min 24 s of focused work". Reports use 4 seconds per unit to show that reading. Effort of doing scores 50 out of 100 at 5 effort units and approaches 100 as effort grows, so the score keeps rising as effort grows instead of stopping at a fixed ceiling.

The score is per task, not per minute: a variant that gets the same task done faster with the same actions does not score worse. Pace (effort units per minute) is shown beside it and is not scored.

Costs are absolute. Lowering every cost lowers every score. Typing and scrolls also cost per key and per screen, because a long entry or a long scroll is more work than a short one, but not proportionally more.

How the 0 to 100 score is built

The score is a weighted mean of four categories. Categories with no data in a session (for example, no audio) drop out rather than count as zero. A session in which none of the weighted categories was measured has no score: it is shown as Not scored and left out of medians and rankings, never counted as 0.

• Effort of doing (scrolling, switching, thinking): weight 3 in Balanced.
• Negative narration: weight 4 in Balanced.
• Amount of narration: weight 1 in Balanced.
• Time on task (the length of the recording): weight 2 in Balanced.

Negative narration, amount of narration and time on task keep their own 0 to 100 scales. Changing an action cost never moves them, and changing their weights never moves an effort unit.

What the numbers mean

Each category and the Effort Score run from 0 to 100, and lower is easier. Each scale is fixed: it turns what was measured in one session into a number and does not compare the session with other sessions or people. There are no benchmarks yet, so reports show no good or bad bands: compare sessions of the same task. Where a category has its own units (the recording length, switches per minute, words per minute), the report header shows them, and each info button gives the 0 to 100 value.

Cost presets

A preset is a named, versioned set of costs for one way of working. A shipped preset is never edited: new values ship as a new version, which your workspace adopts only through the preview. Presets never change your category weights.

How the devices differ

What each simple slider prices, in effort units and seconds, under every preset. One comparison always uses one set of costs, even when one variant is a phone flow and another a desktop flow: pricing each session with its own device would let a guess about the device decide which is easier. Compare views say instead whether their ranking holds under all five presets.

• Phone: Two-thumb phone use. Typing: Phones average 36.2 words per minute, so each key costs 1.43x desktop; starting an entry also brings up the keyboard. Switching: The full-screen app switcher, reloads on return, and attention shifts: about 1.3x desktop.
• Phone one-handed: One-handed phone use and phone data entry. Typing: One-finger typing averages 29.2 words per minute: each key costs 1.77x desktop (the typing 1.8 on mobile). Switching: As two-thumb phone use.
• Keyboard-heavy data entry: Trained operators, keyboard-heavy desktop work. Typing: Skilled operators type fast, and Tab replaces homing, so an entry starts with the decision alone. Switching: Switching between systems dominates this swivel-chair work.
• Equal: Transparency baseline: every detected event costs 1. Typing: Every detected typing burst costs 1, whatever its length. Switching: Every detected switch costs 1.

What matters most: category presets

Which actions count, and how accurate each is

Every measured action counts by default except clicks and typing: scrolls, context switches. Each is labeled with how well its detector did in our tests; the table says when those tests included recordings it was tuned on. An admin can stop counting any of them. An action with no detector yet cannot be counted. Cognitive effort counts too: people label stretches of a recording as low, medium or high in its report, and each labeled minute adds its cost.

Clicks and typing are detected and shown in every report, labeled Beta. Click detection is still being improved and does not count toward the Effort Score. Typing detection is still being improved and does not count toward the Effort Score. They come back into the score in a later scoring version, previewed like any other change, once their accuracy is up.

Apart from the transcript check on computer voices and 6 scripted sessions in the typing figures, every accuracy figure here comes from recordings one person on our team made on Windows, on a 4K screen and a 1440 x 1080 screen. Other people, computers and screen sizes (including 1080p and 2560 x 1440) have not been checked yet.

A label comes off only when a detector passes three checks on sessions it was not tuned on: it finds the right events, it measures their size (keys per entry, screens per scroll) within tolerance, and it does not split one action into several. Removing a label never changes a score.

Known limits

• Detectors make mistakes, as the table above shows. A false scroll adds its cost; a missed one removes it. Anyone in the workspace can correct them in a report (remove a false click, add a missed scroll, move a switch), and reports and scores then use the corrected data.
• A detector that splits one scroll into three charges the per-scroll cost three times (0.30 extra at Desktop costs). Once typing counts again, a typing burst split in two charges the per-entry cost again (0.43 extra).
• Typing undercounts keys: on a recording it was not tuned on, it counted 222 of 501 keys typed. Screens per scroll are estimated from the video and not checked yet.
• Once clicks count again, the click that opens a new page is a click; the page change itself will be priced only for reorienting on the new page. Page changes are not measured yet, and same-tab page changes are sometimes counted as context switches today.
• Effort that is not visible in a screen recording, such as screen reader use, is not captured.
• Time on task and effort of doing both grow with task length. That is intended: a longer task is more effort.
• A session is left out of medians and comparisons when one of the counted actions was not fully measured, so a partial measurement never passes for an easy session.

Scoring update

Sources

• Card, S., Moran, T. and Newell, A. (1983). The Psychology of Human-Computer Interaction (Keystroke-Level Model: K, P, B, H, M).
• Dhakal, V., Feit, A., Kristensson, P. O. and Oulasvirta, A. (2018). Observations on typing from 136 million keystrokes. CHI 2018.
• Palin, K., Feit, A., Kim, S., Kristensson, P. O. and Oulasvirta, A. (2019). How do people type on mobile devices? Observations from a study with 37,000 volunteers. MobileHCI 2019.
• Sears, A. and Shneiderman, B. (1991). High precision touchscreens: design strategies and comparisons with a mouse. IJMMS 34(4).
• Andersen, T. H. (2005). A simple movement time model for scrolling. CHI 2005 extended abstracts.
• Zhao, J., Soukoreff, R. W., Ren, X. and Balakrishnan, R. (2014). A model of scrolling on touch-sensitive displays. IJHCS 72(12).
• Holleis, P., Otto, F., Hussmann, H. and Schmidt, A. (2007). Keystroke-level model for advanced mobile phone interaction. CHI 2007.
• Bergstrom-Lehtovirta, J. and Oulasvirta, A. (2014). Modeling the functional area of the thumb on mobile touchscreen surfaces. CHI 2014.
• Mark, G., Gudith, D. and Klocke, U. (2008). The cost of interrupted work: more speed and stress. CHI 2008.
• Murty, R. N., Dadlani, S. and Das, R. B. (2022). How much time and energy do we waste toggling between applications? Harvard Business Review.
• Dixon, M., Toman, N. and DeLisi, R. (2013). The Effortless Experience. CEB.
• Hart, S. G. (2006). NASA-Task Load Index (NASA-TLX); 20 years later. HFES 2006.