PAD Field Guide

Predictive Adaptive Detection — Fluency Analytics

A clinician-facing view of what FluentPlay measures. Per session: Pressure, Anxiety, Ground, and the resulting Disfluency risk — tracked across treatment, broken out by phoneme, and rolled up into a six-axis clinical outcome profile.

The math is one equation: D = P × A × (1 − G·λ). The first five slides walk through what each term means and how to read the dashboard. Slide 6 is the dashboard itself.

Patent pending — 64/016,001 Single-subject Demonstration data
Will Carbone  |  FluentPlay Technologies LLC  |  April 2026  |  willcarbone@fluentplaytech.com

One equation: D = P × A × (1 − G·λ)

FluentPlay reduces every fluency moment to four numbers. Pressure (P) is the demand the speech act puts on the system. Anxiety (A) is the affective cost that demand carries. Ground (G) is the speaker's stability — the trained capacity to absorb pressure and anxiety without breaking. λ is a coupling constant that sets how much Ground actually buffers — currently 0.6 in the production model.

D = P × A × (1 − G · λ)

Why a product, not a sum

If either Pressure or Anxiety is zero, disfluency goes to zero. Both have to be non-trivial for a disfluency to occur. A speaker reading silently has no Pressure. A speaker reciting an over-rehearsed phrase has no Anxiety. Disfluency lives in the multiplication, not in either term alone.

Why Ground is subtractive

Ground does not eliminate Pressure or Anxiety. It absorbs their product. As Ground rises through practice, the same P and A combination produces a smaller D. The treatment goal is not to lower P or A — it is to raise G.

What the dashboard tracks

Per session, the four values P, A, G, D are computed from in-game performance and physiological signal. The trajectory across sessions is the clinical outcome — falling P and A, rising G, falling D.

Drag the four sliders. Watch D update live.

Live PAD calculation

Pressure (P) 0.72
Anxiety (A) 0.68
Ground (G) 0.35
Coupling (λ) 0.60
D = 0.72 × 0.68 × (1 − 0.35·0.60)
0.387
Elevated disfluency risk — naive baseline state

The P, A, G, D components

Every PAD value sits in [0, 1]. The dashboard shows them as decimals to three places. Each component captures a different layer of what's happening when a speaker tries to produce a phrase.

Why split them

A single disfluency-rate number tells you that something went wrong. Splitting it into P, A, G tells you why — and which lever to pull. A speaker with high P and low G needs Ground work. A speaker with high A and modest P needs anxiety reduction. A speaker with low G needs both.

Trajectories

Across treatment, the expected pattern is: P falls (the speech act becomes less demanding), A falls (less affective load), G rises (capacity grows), and D falls as a consequence. The dashboard's PAD Components Over Time chart shows all four traces on one plot.

Sign convention

P, A, D: lower is better. G: higher is better. The radar chart on the Clinical Profile tab inverts P and A (showing "Pressure Reduction" and "Anxiety Reduction") so all six axes point the same direction.

P
Pressure
The demand the speech act places on the speaker's motor planning system. Driven by phoneme difficulty, utterance length, time pressure, social context.
trajectory: falls with practice
A
Anxiety
The affective cost the speaker is carrying into the moment. Anticipatory tension, prior failure salience, performance stakes. Independent of physical demand.
trajectory: falls with desensitization
G
Ground
The speaker's stable capacity to absorb P and A without breaking. Trained through repetition under controlled load. The thing FluentPlay's games are built to grow.
trajectory: rises with treatment
D
Disfluency Risk
The computed output — the model's estimate of how likely this moment is to produce a disfluency. Function of P, A, G via the PAD equation. Not a direct measurement.
trajectory: falls as G rises

Phoneme-level tracking

The aggregate PAD score hides which sounds the speaker is actually struggling with. The Phoneme Map tab breaks the signal down by individual phoneme — /s/, /r/, /k/, and so on — showing baseline performance against current performance for each.

Baseline vs Current

For each phoneme, two numbers: baseline, captured during the speaker's first sessions, and current, the running average from the most recent N sessions. The dashboard shows both as percentages on side-by-side bars, with the Δ in points.

Grouping

Phonemes are grouped by manner of articulation: Stops (/k/, /g/, /b/, /d/), Fricatives (/s/, /θ/, /ʃ/, /f/, /v/), Liquids (/r/, /l/), Affricates (/tʃ/). Patterns within a group often reveal articulation strategies; patterns across groups reveal more general control issues.

Why this matters clinically

Two speakers with the same overall fluency score can have very different phoneme profiles. Treatment plans should target the specific phonemes showing the smallest Δ, not the speaker's overall score.

Pick a phoneme. Drag the sliders to see the Δ change.
/s/ Fricatives · 18 sessions
Baseline45%
Baseline 45
Current82%
Current 82
+37 pts 18 sessions

Six-axis outcome profile

The Clinical Profile tab rolls everything up into a single radar plot — six axes, each in [0, 1], all oriented so that higher is better. The shape of the polygon is the speaker's clinical signature at a glance.

The six axes

  • · Pressure Reduction — drop in P from baseline
  • · Anxiety Reduction — drop in A from baseline
  • · Ground Strength — current G level
  • · Fluency Rate — % of utterances without disfluency
  • · Session Consistency — adherence and frequency
  • · Phoneme Range — breadth of phonemes showing improvement

How to read the shape

A round, evenly-distributed shape near the outer ring = strong overall response. A spiky shape = uneven progress, with treatment gaps the dashboard surfaces by name. A small shape = early in treatment or limited engagement.

Drag any axis slider. The polygon redraws live.

Adjust the six axes

Loading PAD Clinical Dashboard…

Methods & limitations

PAD Clinical Dashboard is a single-subject visualization built on the FluentPlay PAD framework. The data shown in this build is simulated for demonstration. Read this slide before drawing any conclusion from a session profile.

Methods

The PAD framework

PAD = Predictive Adaptive Detection. Per-syllable scoring system modeling pre-articulatory planning instability. Provisional patent pending: Application 64/016,001, filed March 24, 2026.

The equation

D = P × A × (1 − G·λ)

P, A, G, D ∈ [0, 1]. λ = 0.6 in production. Computed per session from in-game performance signals; integration with EEG-derived physiological anchors (Pleasure dimension via PAD Ready State) is in development.

Phoneme scoring

Per-phoneme baseline captured during the speaker's first 3 sessions. Current value is a rolling average of the last 5 sessions in which that phoneme was produced. Δ = current − baseline, expressed in percentage points.

Composite outcome

Six-axis radar combines normalized changes across Pressure, Anxiety, Ground, fluency rate, session adherence, and phoneme breadth. Each axis is independently scaled to [0, 1] before plotting.

Data sources

  • In-game: Speech Console, Articulation Trainer, Sound Bridge, Rainbow Syllables, Summit, Rhythm Pad
  • Self-report: pre/post session anxiety check-ins
  • Physiological (planned): EEG via PAD Ready State station, fNIRS via NIRx collaboration

Limitations

Demonstration data

The numbers shown in this build are generated for demonstration — they do not reflect real treatment outcomes for any individual. A clinical deployment connects the dashboard to actual session output from FluentPlay games.

Not a diagnostic device

This dashboard does not diagnose, classify, treat, or evaluate any condition. It visualizes a single speaker's performance against their own prior baseline. It is not normed against any clinical population.

Single-subject design

Conclusions are within-subject only. Trajectories on this dashboard cannot be compared between individuals — baselines are person-specific by construction.

Model status

The PAD equation is a working model under active development. λ may be revised. Component definitions may be refined as the EEG and fNIRS integrations mature. Treat metrics as provisional until the framework is published in a peer-reviewed venue.

Self-report limits

The Anxiety component partly relies on speaker self-report. Self-report introduces social desirability and recall bias. The EEG anchor under development is intended to reduce that dependence.

Phoneme coverage

The phoneme map only shows phonemes the speaker has produced in enough sessions to have a current estimate. Absent phonemes are not zero — they are unmeasured. Empty slots are not deficits.

Patent. US Provisional Application 64/016,001 — Predictive Adaptive Detection per-syllable scoring system. Filed March 24, 2026 (micro entity). Non-provisional deadline March 24, 2027.

FluentPlay Technologies LLC. Will Carbone, Founder & CEO. Somerville, MA. willcarbone@fluentplaytech.com. April 2026.