Pilot preview, September 2026
Cadence and MARS
Cadence is an iPad app for recording and reviewing a person’s own story, with MARS-style narrative measures alongside the transcript. Recording and review happen on the iPad; cloud transcription and model analysis require consent. The comparison below describes the current app and the questions for clinical testing.
The walkthrough
- 0:00Physical iPad
- 0:06Spontaneous storytelling
- 0:20Saved demonstration story
- 0:37Narrative words
- 0:45Word class
- 0:53Utterance check
- 1:07Phrases
- 1:17Clauses
- 1:25Dark mode
- 1:29Does the verb have everything it needs?
- 1:44Who does “we” mean?
- 2:00Story grammar
- 2:13Clinical metrics
- 2:24Chart-note draft
- 2:36Export
- 2:49September 16th closing
Start with the person’s own story
A client tells a spontaneous story about something that happened to them, in their own words. They do not need to read a supplied passage. The therapist can then work from words and sentences to the story as a whole.
Does the verb have everything it needs to make the sentence work? If someone says “we”, is it clear who they mean? Does the story establish who was there, where and when it happened, how events unfolded, and how it ended?
Cadence shows counts and proposed structures for the therapist to check against the speaker’s words and intended meaning. Its basic utterance check cannot establish whether the verb has everything it needs.
What each app shows
This table maps the terms used in MARS to the corresponding views in Cadence.
| MARS tab | MARS | Cadence | Comparison |
|---|---|---|---|
| Adjusted transcript | Edit after upload; save explicitly. Losing edits on navigation was the top pain point in discovery. | Edit in place; saved on every keystroke; each analysis run records which transcript version it used. | Saves edits automatically |
| Narrative words | Count and per-minute rate. | Same two measures, highlighted in the transcript. | Adds transcript highlighting |
| Word class | Part-of-speech counts; MARS's most accurate tab. | Same eight counts plus connectives, with the whole transcript coloured by class. | Adds connectives and highlighting |
| Utterances and completion | Two tabs. | One numbered list and MLU in words. “Passes check” means at least two words and a verb; full sentence completeness still needs clinical review. | Combines two views; basic completeness check |
| Phrasal analysis | Bracketed phrase nesting. Slow (40 min or more); flagged as a candidate to drop. | Phrase counts and a bracketed phrase view. Not yet timed or validated on clinical material. | Timing and clinical usefulness to be tested |
| Multi clause | Multiclause utterance count. | Same, plus single-clause count and coordination vs subordination. | Adds clause breakdown |
| Argument structures | Clause-level 0 to 4 argument counts. | Predicate-level approximation. Six of the 27 metrics come from here and the app labels them as approximations in the panel and the PDF. | Uses a different method |
| Reference chains | Chain count. | Proposed links shown in the transcript. New analyses can locate unresolved occurrences; older saved results may contain only a count. Every link needs review. | Shows proposed links |
| Story grammar | Single model. The project's own blog notes the results are inconsistent. | Two models score the six Stein and Glenn elements independently; disagreements are shown for clinician review. Evidence availability depends on the saved result. | Shows agreement between two models |
| Protocol Table 1 (27 metrics) | 18 at Stage 1, 27 at Stage 3. | 27 reported on one sheet and three PDF pages; six are the argument-structure approximations above. | Reports all 27; six are approximations |
Other Cadence features
| Cadence | Why it matters |
|---|---|
| Saves edits automatically | Edits are saved as you work. Reopen a session to continue editing. |
| You approve what is sent to online services | Audio and transcripts are stored on the iPad. If you approve online transcription or analysis, the relevant material goes to the selected service. Some analysis runs on the iPad. |
| Measures from the audio, not just the text | Pause structure and speech rate come from the waveform. A transcript cannot give you those. |
| Across sessions | Compare a client’s story structure and other measures across sessions. |
| Chart note and exports | A draft note for the therapist to check and copy into the client record. Export reports and data as PDF, CSV, JSON or Praat TextGrid. |
| Recording quality check | Checks recording quality before analysis. |
What we’d like to learn from using Cadence
Cadence hasn’t yet been tested in clinical use. We’d like to find out how accurate the results are, how much checking they need, and how long a session takes.
MARS has published clinical testing results. We don’t yet have equivalent evidence for Cadence.
- How closely do the results match the therapist’s analysis, and where do they need correction?
- How much clinician time does a session take, including checking and corrections? Cadence’s target is under an hour.
- How accurately does it transcribe Scottish, Welsh and Northern Irish speech? Its dialect support still needs testing against checked transcripts.
- How long does phrasal analysis take, and does it help the therapist?
A difference in argument analysis
The two apps count argument structure differently. Cadence currently works around individual verbs; MARS reports counts by clause. Those results aren’t directly comparable yet.
Getting it on an iPad
Distribution is through Apple's TestFlight. It needs an iPad Pro from 2018 or later on iPadOS 26. The on-device language model additionally needs an M1 or later; everything else runs on older Pros.
TestFlight link will follow (ask Fiona)
- Install TestFlight from the App Store on the iPad.
- Open the invite link on the iPad. Tap Accept, then Install.
- Open Cadence. The first launch loads the same sample cases as the video.
Questions for your review
- Where do the results go after a session: the client record, an email to yourself, a printout?
- Phrasal analysis: keep, simplify, or cut?
- Which speech varieties should we test first?