Skip to content

Research

What Lodestar records

Lodestar is a free Mac app people use all day for its own sake. While they do, it records how their hands type and point, on their Mac, for as long as they use it. It is not a diagnostic.

What we are looking for

Research groups who want to find out whether this record can help identify conditions, and follow them over time.

Parkinson's disease is where the published work is strongest, but the record is not limited to it. A first study could be small: people with a diagnosis and people without, recording for a few months alongside your clinical assessments. Lodestar is free, we would build the consent and data export with you, and the data is yours to analyse.

Write to Rocco Vaccone, research@vaccone.software

Duration

There is no task to do, so the record runs as long as the person keeps using it. Across eight remote smartphone studies of more than 100,000 people, median retention was 5.5 days (Pratap et al. 2020).

Breadth

It keeps every keystroke and every pointer movement, not summaries, so the published keyboard and mouse measures can be recomputed from it, and so can ones nobody has proposed yet.

What it records

As of Lodestar 0.54. Everything is kept on the Mac, sealed, and can be exported.

Keyboard

  • Every keystroke: when the key went down and how long it was held, in microseconds, from the keyboard event's own stamp
  • The hand, the finger and the kind of key (letter, space, backspace and so on), never which key
  • Every 90 seconds of typing: hold time quantiles, the log fluctuation index, flight and overlap, by hand and by direction

Pointer

  • Every movement as the device reports it: motion that ends in a click, motion that comes to rest, and drags
  • Every click, with its source, the device, Force Touch pressure and how long it was held
  • Every scroll
  • Each movement's speed profile: peak speed, submovements, and power in the 1 to 3, 4 to 7 and 8 to 12 Hz bands

Voice, when dictating

  • Pitch and its spread, loudness, jitter, shimmer and harmonics to noise
  • Pauses, speech rate and articulation rate
  • Never the audio

Daily rhythm

  • Interdaily stability and intradaily variability
  • The most active ten hours and least active five, onset, offset and midpoint
  • Bouts of work, and the drift within each

Context

  • The kind of app (browser, terminal, chat and so on) and its name
  • The keyboards attached and the report interval of each, the screens, power and the lid
  • When it could not see (secure input, asleep, not running), and every change of build or device

About the person

  • Year of birth, sex as assigned at birth, dominant hand and whether they touch type
  • All optional

The keystroke record never names a key, but the pattern of hand, finger and kind of key can partly reconstruct what was typed. Treat it as sensitive.

How it compares

The published studies closest to Lodestar. Several measured people once, in one typing session or a short test. Lodestar records the same signals during ordinary work, every day.

StudyConditionWhat it measuredIn Lodestar's record
Giancardo et al. 2016Early Parkinson'sKey hold times from one typing session of about fourteen minutes (MIT neuroQWERTY)Yes, every hold, every day
Lan and Yeo 2019Early Parkinson'sA simpler index of how hold times fluctuate, on the MIT dataYes, computed every 90 seconds
Arroyo-Gallego et al. 2018Early Parkinson'sKey hold times from everyday typing at homeYes
Adams 2017 (Tappy)Early Parkinson'sHold, latency and flight times by hand and direction, typing on their own keyboardsYes
Gajos et al. 2020Ataxia and parkinsonismMouse movement in a short test in a web browserYes, in everyday use rather than a fixed task
Zulueta et al. 2018 (BiAffect)Bipolar disorderKeystroke timing and corrections on a phone keyboard, over eight weeksPartly: the same kind of timing on a Mac keyboard, no phone sensors
mPower (Bot et al. 2016)Parkinson'sTapping, voice, walking and memory tasks on a phonePartly: voice when dictating, no walking

How far it has been checked

Against the MIT study
Replaying the MIT neuroQWERTY study's published data through Lodestar, on all 85 people, gives an AUC of 0.750 for Lan and Yeo's index computed their way, against their 0.752. Lodestar's own 90 second windows give 0.743. This shows the code computes the measure as published, nothing more.
Its own timing
Lodestar measures itself as it records: it hears each key about 5 ms after the keyboard event's own stamp, and sees the hold of 99% of the keys struck.
Clinical validation
Not done. So far the record is one person, with no diagnosed participants and no clinical labels. That is what working with researchers is for.

What participants need

A Mac with Apple silicon and macOS 27 or later. Lodestar would be installed on the participant's own Mac.

Who we are

Lodestar is built by Rocco Vaccone. The full design of the record is in the design specification. Write to research@vaccone.software.