Living document · last updated 10 May 2026
Nasal helps a child notice when their mouth opens during a passive task — usually watching a video on a tablet — and gently nudges them back toward nose breathing. It does this without supervision, without effort, and without removing the child from the activity they enjoy.
The principle is simple. Mouth breathing during screen time is often a default the child has fallen into without awareness. Awareness, repeated and consistently, is what lets the nervous system update the default. Nasal provides that awareness — quietly, in the moment — so the child can correct on their own.
The device's front-facing camera watches the child's face during the session. It is used as a sensor, not as a recorder. No video is ever saved or transmitted. Frames are analyzed on the device in real time and discarded immediately. The only thing that leaves the device is anonymized event data — when the mouth opened, how long it stayed open, whether the eye region suggested a yawn — sent to the clinician's dashboard.
The camera is on while the child watches. The app sees their face. Nothing is recorded. When the mouth opens, the app responds — sometimes silently, sometimes with a gentle reminder, sometimes by softening the screen until the mouth closes. Over weeks, the child needs the reminders less and less.
During a session, the child watches a video of their choice or plays a calibrated game. If they breathe through the nose, nothing changes — the experience is uninterrupted. If their mouth opens, the app may do nothing at all, show a gentle text reminder, dim the screen briefly, or pause the video, depending on which therapeutic phase the patient is in. The moment the mouth closes, the screen returns to normal.
There is never a punishment, never a score the child has to beat in real time, and never a withholding of access. The feedback is contingent (it happens precisely when the mouth opens) and immediately reversible (it stops the moment the mouth closes). This is what allows the nervous system to learn.
Children see a small visual reward — a seed that grows into a sprout, sapling, young tree, tree, mighty tree, and finally a forest — as they accumulate sessions. Both video sessions and breathing-game sessions count toward the tree. This is intentional: the tree is motivation, not measurement. Its job is to keep the child returning to the app over the months of therapy.
The tree is independent of clinical progression (described in the next section). A child can be a "Sprout" for tree purposes while still being in the clinical baseline phase, or vice versa. They serve different purposes.
Skins. Each profile in the picker has a small palette icon (🎨) next to the name. Tapping it opens a chooser with eighteen colour themes — twelve dark, six light — so the app can be tuned to what feels good for that profile. The choice saves per profile, so siblings sharing a device can each have their own look. It's purely cosmetic; the breathing detection, the prompts, and the session data work the same way regardless of which skin is selected.
This section describes the intervention engine in operational detail. It is intended for the supervising clinician and assumes a working understanding of single-case experimental designs and operant feedback principles.
A mouth-open event triggers exactly one intervention level at any given moment. The level is decided by the patient's current clinical phase floor (where the event starts), how long the mouth has stayed open (within-event escalation), and, where applicable, the active session template. Detection itself is invariant across levels — every event is logged with timestamp, duration, peak jaw-open value, and yawn-likelihood, regardless of whether the child sees any feedback.
The full within-event behaviour: an event opens at the phase's floor level. If the mouth stays open past the initial escalation delay (default 2 seconds), the level escalates by one. Every step duration thereafter (default 1.5 seconds) it escalates by another, up to the phase's ceiling level. This model is documented in detail in the Phases and progression section below.
| Level | Name | What the child experiences |
|---|---|---|
| A | Silent | No change to the screen. The event is logged for later review but no feedback is delivered. Used for baseline measurement and washout blocks. |
| B | Notice | A gentle text overlay appears on the video without dimming or pausing. A brief haptic pulse accompanies it on devices that support vibration. The child can ignore it and the video continues. |
| C | Light dim | The screen blurs gently (≈6 px) while the video keeps playing. The reminder text becomes more salient. Designed to be noticeable without being aversive. |
| D | Heavy dim & pause | The screen blurs heavily (≈14 px) and the video pauses. The reminder text is firmly worded. The session resumes the moment the mouth closes. This is the therapeutic intervention level. |
| E | Block | Heaviest blur (≈24 px), pause, and a fully blocking overlay. Reserved for clinician-enabled cases where a stronger contingency is indicated. Off by default. |
The five-level granularity exists so the clinician can match the contingency strength to the child's age, attention profile, and tolerance. A 4-year-old in early therapy may benefit from B (notice only); the same child four weeks later may be ready for C; a 9-year-old stuck on a plateau may need D; E is reserved for cases where the child has demonstrably learned to ignore lower levels. The patient app handles all five identically — only the visual and temporal parameters change.
Each patient has a clinical phase that determines how the within-event intervention behaves. Only passive-viewing (video) sessions advance the phase counter — game sessions do not, because the game itself uses a different feedback paradigm (continuous mouth-seal as forward propulsion) and is not equivalent to a measurement session.
The model is errorless learning followed by stimulus fading. Within every mouth-open event, intervention starts at the phase's floor level and escalates by one step every step-interval of continued persistence, up to the phase's ceiling level. This ensures the patient gets a chance to self-correct at the lowest contingency before stronger ones arrive, and avoids the aversive failure mode of dropping straight to D on every event. Each phase has its own floor and ceiling, configured per patient. The advancement criteria between phases are session-count thresholds for baseline → practice → discipline, then a clinician-discretionary move into mastery.
Every mouth-open event fires level A — silent logging, no feedback. The purpose is to establish each child's natural rate of mouth-breathing during screen time before any intervention is introduced. This is the no-treatment reference against which subsequent phases are compared. The child is unaware that anything is "happening" — they see only the video. Within-event escalation is effectively disabled here because floor and ceiling are both A; no level above silent can ever fire.
Each event starts at A (silent) and escalates one level every step-interval of persistence, up to D. The dynamics: a patient who self-corrects quickly never experiences anything above A; a patient who persists past the initial grace window gets B, then C, then D. This is errorless learning in operant terms — the patient is given the opportunity to perform the desired behaviour at the lowest possible contingency cost, and only experiences the firmer levels when persistence indicates they are not self-correcting. The phase is designed to teach the contingency without aversive shock — kids who first encounter D on their very first mouth-open event will refuse the camera; kids who experience the gentle ramp learn the contingency without resistance.
Each event starts at C (light dim) and escalates to D (heavy dim + pause) on persistence. The floor is raised here because the patient has demonstrated learning in practice; the system can now deliver the meaningful contingency immediately rather than ramping up to it. Therapy continues in this phase for the bulk of treatment, typically several weeks to months, with progress measured by the falling rate and duration of mouth-open events across sessions. Discipline does not auto-advance to mastery — the transition is a clinical decision. The dashboard surfaces a "Ready for mastery?" readiness signal when the patient has had at least 8 sessions in discipline and their recent mean score is 75% or higher; this prompts the clinician to evaluate readiness rather than committing the system to advance automatically.
The floor drops to B for normal mastery sessions — a gentler contingency than discipline, since the patient is now consolidating rather than acquiring. The clinical purpose of mastery is to confirm that the seal-behaviour has become intrinsic rather than externally-enforced. A patient who has been trained on a strong contingency may show high sealed% during sessions only because the contingency is present; when it is removed, the behaviour may erode. Mastery is the phase where this is tested deliberately.
A configurable fraction of mastery sessions (default 50%) are marked at session start as test-fade sessions. In a test-fade session, the floor drops to A — silent — for the entire session. No contingency fires regardless of how long the patient's mouth stays open; the system only logs. The test-fade flag is recorded with the session in the database and surfaced in the per-session appendix (column suffix "·tf") and in a dedicated Mastery durability page in the clinical PDF.
The clinical reading: if test-fade sessions show sealed% at parity with normal-mastery sessions, the seal-behaviour has generalised — the patient is maintaining nasal breathing without the contingency, and fading is working. If test-fade sessions show a substantial drop in sealed%, the skill remains contingency-dependent; the clinician should consider returning the patient to discipline phase for further consolidation before resuming the mastery probe.
Two timing parameters control the escalation rhythm within a single mouth-open event. Both are configurable per patient via the protocol panel, and both are expressed in milliseconds (1000 ms = 1 second).
The combination produces a within-event timeline whose shape depends on the patient's phase. Two worked examples:
Practice phase (floor A, ceiling D) — full graduated ramp
| Time open | Level | What the child experiences |
|---|---|---|
| 0 s | A | Mouth opens. Silent — nothing visible. Just logged. |
| 0 – 2.0 s | A | Grace window. If the child closes here, they never see any feedback. |
| 2.0 s | B | Notice — gentle text overlay + brief haptic. Video keeps playing. |
| 3.5 s | C | Light dim — screen blurs gently. Video still plays. |
| 5.0 s | D | Heavy dim + pause. Video stops. Ceiling reached. |
| 5.0 s+ | D | Stays at D until mouth closes. |
Discipline phase (floor C, ceiling D) — fast ramp, one step
| Time open | Level | What the child experiences |
|---|---|---|
| 0 s | C | Mouth opens. Light dim fires immediately. Video keeps playing. |
| 0 – 2.0 s | C | Grace window at C. Child can self-correct at the light level. |
| 2.0 s | D | Heavy dim + pause. Video stops. Ceiling reached on the first step. |
| 2.0 s+ | D | Stays at D until mouth closes. |
Visually, the two timelines look like this:
Two seconds is roughly the time it takes for an attentive child to notice that their mouth has opened, decide what to do, and start closing it. It is also long enough that a brief accidental open — a cough, a swallow, a sneeze — usually resolves itself before any feedback fires. The grace window is calibrated to that human-perceptual rhythm: short enough that persistent mouth-breathing is caught quickly, long enough that the child isn't punished for the body's normal movements.
Once escalation starts, 1.5-second steps mean the system moves toward the ceiling in a few seconds — not 10 or 15 — but each step is still slow enough that the child can perceive the change and respond to it. The pattern the child experiences is: "the screen is dimming a bit; the screen is dimming more; the screen has paused" — felt as a graded sequence, not a sudden cliff. This is the difference between a contingency the child can learn from and one that simply punishes.
A clinician who wants the system to be gentler — more grace, slower escalation — increases both numbers. A clinician who wants it firmer — less grace, faster escalation — decreases them. Reasonable adjustment ranges are 1000–4000 ms for the initial delay and 1000–3000 ms for the step duration. Outside those ranges the protocol becomes either too lenient (child never experiences the contingency) or too aggressive (child experiences escalation before they can perceive the prior level).
Both are tunable per patient via the protocol panel. Defaults are conservative pilot values; expect to refine these from cohort data as the clinical signal accumulates.
A patient whose events rarely escalate past the floor is responding to gentle cues — they are doing the self-correction work themselves, and the system is just providing the reminder. This is the clinical win. A patient whose events routinely escalate to ceiling needs the stronger contingency to produce closure — useful information that may indicate need for protocol adjustment (raising the floor, extending the phase, or addressing engagement). The per-event maximum level reached is logged with each prompt event in prompt_events.kind, allowing the clinician to read the distribution of within-event escalation as a derived metric.
The phase logic above is the default. A clinician may override it on a per-session basis by attaching a template — a fixed sequence of timed blocks, each with its own intervention level. Templates lock intensity to time-segments rather than letting within-event escalation drive it. Templates are useful for within-session experimental designs (reversals, fading probes, dose-response) and for research protocols. The within-event escalation logic is bypassed when a template is active.
Templates remain available but are not the default model. Most patient-care sessions run on the within-event escalation logic; templates are reserved for research designs or specific clinical investigations where the time-locked structure is needed.
The Monitor modality runs differently from interactive sessions and is worth explaining in design terms, since it sometimes prompts the question "why doesn't Monitor have a calibration step like the regular sessions?"
Monitor is the off-screen observation modality. The device is placed on a stand, the camera points at the child, and the child does something else for 15 minutes — drawing, Lego, watching TV from the couch, puzzles. There is no UI for the child during this time; the app simply observes. The output is a clean read of how the child breathes when they are not in front of a feedback-delivering screen.
This modality uses fixed jaw-open thresholds rather than per-patient calibration, which is a deliberate design choice driven by three considerations:
Per-patient adjustment is achieved instead via the Monitor sensitivity setting on the protocol panel (low / medium / high), which selects between three pre-set jaw-open thresholds. The clinician picks the level that produces clinically meaningful event counts for the individual patient — high sensitivity for a child with subtle partial-open patterns, low sensitivity for a child whose mouth-open events are pronounced enough that even a coarse threshold will catch them. Default is medium; adjustment usually only matters at the extremes.
The within-event escalation + test-fade model maps onto established operant terminology: errorless learning in the within-event ramp (patient encounters the lightest contingency that produces compliance), stimulus fading in the test-fade probe (deliberate reduction of contingency strength to test for behaviour generalisation), and fading-to-mastery in the cross-treatment arc (graduated contingency removal across phases rather than abrupt termination). The pediatric mouth-breathing and orofacial-myofunctional therapy literature is consistent with this approach: patients who experience graduated fading hold the behaviour longer than patients who only ever experienced the full contingency.
The phase-to-criterion advancement defaults — 3 baseline sessions, 5 practice sessions, clinician-decided exit from discipline, 50% test-fade ratio in mastery — are pilot values calibrated to produce clinically interpretable data within a reasonable treatment window. Cohort data will refine them.
When a template is active, it fully overrides phase-based intervention selection: the level is determined by the current block, not the phase. Per-block metrics are stored alongside the session record so the clinician can read out, for each block, the number of events, total open duration, mean jaw-open peak, and yawn count. This is the data substrate for ABACA-style analysis.
If a patient is in discipline phase and a clinician attaches an ABACA template for a single session, that session runs ABACA — not discipline. After the session, the patient remains in discipline. Templates are session-level overrides; they do not change the patient's phase. This separation lets a clinician interrupt routine therapy with a measurement session without disturbing the treatment trajectory.
The clinical phase ladder (baseline → practice → discipline → optional mastery) and the child-facing tree (Seed → Sprout → … → Forest) are deliberately decoupled. The phase ladder counts only video sessions and reflects therapeutic progress. The tree counts video and game sessions together and reflects engagement. A child who logs many game sessions but few video sessions will see their tree grow while their phase remains in baseline — which correctly represents the clinical situation: they are using the app, but they have not yet accumulated the measurement basis for therapy to begin.
The default tree pacing — Forest at 60 sessions — corresponds approximately to two months of daily use. Therapy is typically expected to run three to four months; for slower-paced patients (every-other-day use, or interruptions), additional tiers beyond Forest will be added so that visual progression continues to be available. Engagement matters: a child who stops using the app does not benefit from the device, regardless of the clinical phase they are in.
Resistance to nasal airflow is determined principally by two anatomical narrowings in series: the external nasal valve and the internal nasal valve. The internal valve, formed at the junction of the caudal margin of the upper lateral cartilage with the nasal septum, is the narrowest cross-section of the entire upper airway and the dominant site of resistance in most patients. The external valve, defined by the alar rim, columella, and nasal sill, becomes rate-limiting when the alar cartilages are weak, narrow, or collapse-prone, and in any patient at sufficiently high inspiratory flow.
Patency at both valves is actively maintained against the inward-collapsing force generated by the Bernoulli effect during inspiration. Two muscles do the principal work.
The alar part of nasalis inserts onto the alar cartilage and pulls it laterally and slightly inferiorly during contraction. Its functional role is stabilisation and dilation of the internal nasal valve against inspiratory negative pressure. It is reflexively recruited via the trigeminal–facial loop in response to increased inspiratory resistance, and its tonic activity scales with ventilatory demand. The transverse part of nasalis (compressor naris) is its functional antagonist and constricts the airway during the disgust expression and during forced nasal expiration.
LLSAN originates on the frontal process of the maxilla and inserts on both the alar cartilage and the upper lip. Its contraction vector is superior–lateral, lifting the alar wing upward and outward. Functionally this opens the external nasal valve and elevates the alar rim, reducing external valve collapse and altering the inflow geometry. The redirected airstream enters at an angle that biases flow superiorly toward the olfactory cleft, which has implications for both olfactory sampling and conchal flow conditioning.
LLSAN and the alar nasalis address different valves. They are complementary, not redundant. Co-contraction during inspiration produces maximum patency along the inflow path; deficit at either valve will limit performance regardless of how well the other is recruited. Identification of the patient's rate-limiting valve guides which muscle deserves training emphasis.
Voluntary isolation of dilator naris is difficult for most patients, who can flare only via bundled co-contraction with LLSAN, often with parasitic co-activation of compressor naris. Direct verbal cues to "flare your nostrils" or "open the deep nostril" rarely produce the desired pattern in naïve patients and frequently train the wrong synergies.
The sniff is a hardwired motor program that recruits the correct synergy automatically: LLSAN and dilator naris co-contract, compressor naris is suppressed, and the activation is phase-locked to inspiration onset. It is the single most efficient entry point for training.
Although disgust shares LLSAN activation with the sniff, it bundles compressor naris co-contraction with a brief nasal expiration or breath-hold, producing net airway closure. It is the wrong motor program for patency training and should not be used as a teaching cue. Use the sniff exclusively.
Training proceeds through four stages, each building on the one before:
Patient produces a sharp investigative sniff in front of a mirror; clinician confirms visible alar elevation, symmetric flare, and absence of forced lip retraction or nasal wrinkling (which indicates compressor co-activation).
Patient extends the sniff into a 4–6 second inspiration, maintaining alar opening throughout the entire inspiratory phase. This decouples LLSAN/dilator activation from the brief reflexive sniff and trains tonic recruitment.
Patient produces the same opening pattern with reduced flow rate and minimal audible turbulence. Removes reliance on high-flow proprioceptive feedback and approaches normal tidal-breathing kinematics.
Low-grade dilator/LLSAN tone present throughout quiet nasal inspiration in unattended breathing. This is the clinical endpoint and typically requires several weeks of daily practice plus environmental cueing.
Recommended dose: 5 minutes, two to three times daily, with at least one session in front of a mirror for visual feedback. Inspiratory resistance training devices (e.g., calibrated threshold trainers) accelerate dilator naris recruitment via the trigeminal–facial reflex and may be added once the unloaded pattern is reliable.
Nasal morphology varies systematically with ancestral climate — a well-established finding in human craniofacial biology dating to Thomson and Buxton and refined by Franciscus, Noback, Holton, Butaric and others. Populations with prolonged ancestry in cold-dry climates (Northern European, Northeast Asian, Arctic) tend toward leptorrhine morphology: narrower, taller, longer nasal cavity with elevated baseline resistance. Populations from hot-humid environments tend toward platyrrhine morphology with lower baseline resistance. Hot-dry and high-altitude populations are intermediate.
Pigmentation and nasal morphology track different selection pressures (UV exposure vs. inhaled-air conditioning) and do not co-vary reliably; Horn of Africa populations are the classic decoupled case, combining dark pigmentation with leptorrhine-tending nasal form because the local climate is hot, arid, and at significant altitude. Clinical assessment should be based on direct evaluation of nasal geometry — nasal index, alar width, external valve angle, internal valve angle on rhinoscopy — not on inferred ancestry from skin colour.
Operationally: patients with leptorrhine-tending anatomy have higher baseline resistance and a lower threshold for switching to oral breathing under ventilatory load. Active dilator/LLSAN training is correspondingly more important for them. Patients with platyrrhine-tending anatomy may maintain nasal breathing more passively and require less intensive motor training, though the same protocol applies.
The argument for trained nasal breathing during exertion is not aesthetic or traditional — it rests on three measurable physiological mechanisms. Patients reasonably ask why they should fight the urge to mouth-breathe during exercise; the clinician should be able to articulate the answer.
The three turbinates (inferior, middle, superior) project into the nasal airstream as scroll-shaped surfaces. Their geometry, combined with the curved three-dimensional path of the nasal cavity, generates organised secondary flow with vortical and helical components. Computational fluid dynamics studies (Keyhani, Doorly, Kim and others) consistently demonstrate complex three-dimensional flow patterns with rotational structure, particularly through the middle meatus. Air arrives at the trachea with embedded rotational kinetic energy and turbulent mixing characteristics fundamentally different from the bulk parabolic flow produced by oral inspiration.
The functional consequence is improved distal mixing and more uniform alveolar ventilation per unit volume of inspired air. Nasally conditioned air contacts more alveolar surface area and reduces preferential streamlining through dominant pathways, with corresponding gains in gas-exchange efficiency. Oral breathing delivers an equivalent tidal volume but with less effective distal distribution. This mechanism complements, and is partially independent of, the well-known conditioning functions of the nasal airway: warming inspired air to core temperature, humidification, particulate filtration, and pathogen capture by the mucociliary escalator.
The paranasal sinuses, particularly the maxillary sinuses, constitutively produce nitric oxide (NO) via NOS isoforms in the sinus epithelium, maintaining sinus cavity NO concentrations in the hundreds of parts per billion to low parts per million range. During nasal inspiration, NO is entrained from the sinus ostia into the inspired airstream and delivered to the lower airways and pulmonary vasculature. Lundberg and colleagues established this pathway and demonstrated its physiological significance.
Inhaled NO produces pulmonary vasodilation with preferential perfusion of well-ventilated alveoli, improving ventilation–perfusion matching; it is bronchodilatory; and it has documented antimicrobial activity. Oral breathing bypasses sinus NO entrainment entirely. The patient inhales ambient air without endogenous bronchodilator/vasodilator delivery. Over high cumulative ventilation — e.g., during endurance exercise or chronic mouth breathing — the foregone benefit is substantial.
Trained nasal-only breathing during submaximal endurance exercise produces equal or improved performance at matched workload, with measurable reductions in minute ventilation, increased CO₂ tolerance, and improved tissue oxygen unloading via Bohr-effect facilitation. At true VO₂max and during sprint-duration efforts the upper-airway resistance becomes rate-limiting and oral breathing recruits as a necessary compensation. For the great majority of training volume (sustained submaximal work) and for activities of daily living, nasal breathing is functionally superior. The "extended sniff" trained as a sustained inspiratory pattern is the technique that allows nasal-only breathing to scale into moderately high ventilatory loads.
Two valves, two muscles. LLSAN opens the external nasal valve; dilator naris stabilises the internal valve. Both must be recruited for maximum patency. Identify the rate-limiting valve in each patient.
Train via the sniff, not the disgust expression. The sniff bundles the correct synergy (LLSAN + dilator naris + inspiration) and suppresses compressor naris co-activation. Disgust bundles the opposite respiratory phase and produces airway closure.
Four-stage progression. Acquisition → sustained activation → quiet inspiration → tonic carryover. Five minutes, two to three times daily; mirror feedback in early stages; resistance loading once the unloaded pattern is reliable.
Anatomy guides intensity. Leptorrhine-tending patients (more common in cold-climate ancestry but assessed individually) benefit most from intensive training. Avoid pigmentation as a proxy for nasal anatomy; assess geometry directly.
Three mechanisms justify nasal-only breathing under load. Conchal vortical flow improves distal mixing; sinus NO entrainment provides pulmonary vasodilation, bronchodilation, and antimicrobial activity; air conditioning protects the lower airway. Oral breathing bypasses all three.
This section is the technical companion to the plain-language "Self-learning" material that follows. It addresses why this kind of intervention works — and where the limits of the working theory are. The goal is to give clinicians a defensible mental model for what Nasal is doing, what it isn't doing, and how it fits alongside other behavioural interventions.
Nasal is, mechanically, an operant-conditioning system: the behaviour (mouth open) produces an immediate, reliable, contingent consequence (screen dim and audio attenuation), and the absence of that behaviour produces the absence of that consequence (uninterrupted content). Over enough repetitions, the brain learns the contingency, anticipates it, and modifies the behaviour to avoid the consequence. This is reinforcement schedule design at the most concrete level.
But the operant frame, as classically formulated, doesn't fully describe what Nasal is doing — and it's worth being honest about that. Three ways the frame is incomplete:
The clinical takeaway: when explaining Nasal to a referring partner, "operant biofeedback" is a fine first sentence. The full picture is closer to "feedback-mediated procedural learning of a partly-autonomic motor pattern, with the contingency designed to land in implicit memory." That's the technical phrase, but the clinical reality is simple: the child watches their show, their mouth learns to stay closed, and after enough sessions the pattern carries over.
The single most important design decision in Nasal is the latency between the behaviour and the feedback. The screen dim is delivered as quickly as the camera-based detection pipeline allows — measured in fractions of a second, not seconds. This is not a configurable parameter; it is the defining feature of the intervention. The question of why such low latency matters has three substantive answers:
This is also why parental nagging fails. Decades of clinical experience document that telling a child to keep their mouth closed produces behaviour change only in the moment of the instruction, with no consolidation and no carryover. The intervention has to land in the right memory system at the right moment, and verbal instruction reliably misses on both counts. It's not a parenting failure; it's a memory-system failure.
The endpoint of biofeedback training is automaticity — the new pattern running without conscious attention. This is the same endpoint motor learning aims for: a tennis serve, a shoelace tie, touch typing. The literature on procedural learning gives us a fairly stable rough timeline, with the strong caveat that individual variation is the rule rather than the exception.
The phases described below — acquisition, consolidation, automaticity, carryover — are theoretical phases of procedural learning, drawn from the broader motor-learning literature. They are not the same as the clinical phases described in the Clinical mechanism section (baseline / practice / discipline / mastery), which are operational phases of the Nasal protocol. The two sets of phases run roughly in parallel — a child in clinical-discipline phase is typically in the acquisition or consolidation phase of procedural learning — but they are not synchronised, and the dashboard does not classify or treat patients based on the procedural-learning phases. The clinical phases are what the system tracks; the procedural-learning phases are the conceptual framework for understanding what's happening underneath.
These phases are not crisp boundaries; they overlap and individuals vary. The rough shape is consistent across the procedural-learning literature, and the clinical implications are practical: a flat plateau in the early months is normal and expected; a flat plateau extending well past the consolidation window without improvement on the Sustainment or Recovery indices warrants protocol review.
HRT is the closest cousin in the behavioural-therapy literature. Originally developed for tics and trichotillomania, it has been extended to nail biting, thumb sucking, bruxism, and other habit-pattern disorders. Its three core components — awareness training, competing response training, social support — map roughly onto Nasal as follows: the dim provides the awareness training (the child becomes aware of mouth-opening at the moment it occurs); the lip closure itself functions as the competing response (it physically prevents the unwanted behaviour); the parent and clinician roles provide the social support layer. The differences are also informative: HRT is conducted entirely through verbal instruction and self-report, with the patient consciously practising the awareness-and-response sequence. Nasal automates the awareness training and removes the conscious-practice requirement, which is what makes it accessible to younger children whose verbal-cognitive scaffolding for HRT is incomplete.
Myofunctional therapy directly trains the muscular and postural components of nasal breathing — orbicularis oris strengthening, tongue posture, palatal contact, oral resting-position correction. It is the natural complement to Nasal: where myofunctional therapy builds the capacity for nasal breathing, Nasal builds the habit of using that capacity. The two address different parts of the problem and pair well clinically. A child with weak lip-seal musculature will struggle with Nasal until myofunctional work has restored baseline tone; a child with adequate muscular capacity but a long-standing mouth-breathing habit will struggle with myofunctional work alone because the home practice is sporadic and effortful. Both interventions together address both axes.
EMG biofeedback has documented efficacy for tension headache, TMJ dysfunction, urinary incontinence, and dysphagia. Heart-rate-variability biofeedback is established for anxiety, hypertension, and stress-related cardiovascular dysregulation. Respiratory biofeedback is used in asthma management and panic-disorder treatment. These traditions are mechanistically similar to Nasal — sensor → signal extraction → contingent feedback → physiological learning — and the clinical outcomes are broadly positive across them. The biofeedback literature is in a curious position: it has decades of supportive efficacy data but rarely reaches randomised-controlled-trial gold standard because the interventions are difficult to blind and adherence-dependent. The honest summary is that biofeedback works in the cases where the underlying physiology is amenable to feedback-mediated regulation, which is most of them.
EEG neurofeedback addresses central rather than peripheral signals, and it is by now a substantial clinical field with evidence for ADHD, epilepsy, anxiety, PTSD, and several other indications. Its theoretical evolution — from operant conditioning of specific frequency bands toward more recent infra-low-frequency endogenous-neuromodulation models — has informed how peripheral biofeedback thinks about its own mechanism. The clinical question for Nasal-as-a-system is whether the peripheral feedback shapes only the peripheral motor pattern or whether it also shifts central regulation of resting respiratory tone. The honest answer is we don't yet know. Pilot data should help differentiate.
Three layers of evidence are relevant to a clinician deciding whether to recommend Nasal:
The honest framing for parents and referrers is something like: "Nasal applies a well-established class of intervention — feedback-mediated procedural learning — to a problem (habituated mouth-breathing) where that class of intervention has not previously been deployed in a portable, home-deliverable form. The mechanism is well-grounded; the product itself is new and is being studied as it deploys."
Several questions are important and not yet answered. Listing them honestly:
The point of listing these openly is that the absence of answers does not preclude responsible clinical use. Most behavioural interventions are deployed in advance of complete answers; the ethical bar is informed clinical judgment, transparent communication with families, and willingness to adjust as data accumulates. That bar Nasal aims to meet.
The thing Nasal helps with is essentially a habit. The mouth opens, then closes, then opens, hundreds of times a day, and over years the open default becomes the resting state — even though the body would prefer to breathe through the nose. The work of changing this is the work of replacing one habit with another, and that follows rules. This section describes those rules in plain language.
Most habits run below conscious attention — that's part of what makes them habits. The mouth opens, no one notices, life goes on. The first thing biofeedback does is interrupt that invisibility. The screen dims at the moment the mouth opens; the child sees the consequence; awareness arrives in the moment of the action. The dimming itself isn't a punishment, and the child doesn't need to think about why it happened. They just learn, after enough sessions, that the dim is connected to their mouth.
This is the operant-conditioning frame: behaviour that's followed by a contingent outcome shifts. The outcome doesn't need to be unpleasant — it only needs to be reliable, immediate, and connected to the behaviour. When the connection becomes obvious, the brain starts to anticipate. The child closes their mouth before the dim arrives. That moment of anticipation is where the new habit lives.
A habit consolidates through repeated, distributed practice. Five sessions in one day matter less than one session a day for five days. The brain is doing a sleep-mediated consolidation between sessions — the connections strengthen overnight. Compressed practice doesn't allow this; it overwhelms the working-memory channel without reaching the long-term storage where habits actually live. So the protocol asks for short daily sessions rather than longer occasional ones.
For the same reason, gaps don't help. A two-week pause doesn't lose all progress, but it does mean the next session feels harder than the last one before the gap. The new habit needs the daily contact to become the new default.
Improvement is not linear. The first few weeks usually show the cleanest gains — the mouth-open rate drops noticeably, sessions feel easier, the child seems calmer. Then a plateau is common. Sessions feel "the same" for a while; the score doesn't move much. This is the consolidation phase, where the new pattern is moving from explicit awareness into automatic behaviour. The plateau is uncomfortable to watch but it isn't failure — it's the work happening below the surface.
After the plateau, the carryover starts. The child begins to keep their mouth closed during activities outside the app — at school, in the car, while reading. This is the goal. The dashboard's Sustainment and Recovery indices, and the Monitor sessions, are designed to detect carryover starting to happen. When it does, the protocol moves into the mastery phase, which deliberately reduces the contingency to test whether the new habit holds without prompting.
Three to four months of regular use is the timescale on which we expect the protocol to mature. Inside that window, fluctuation is normal. Sessions get worse for a week, then better. Children get sick, miss days, return. None of this is a problem. The clinician's dashboard will surface real concerns automatically — the recommendation cards flag declining trends or stalled progress, and a real signal stays visible across many sessions, not just one or two.
Reasons to contact the clinician outside the routine: you suspect an underlying airway problem (chronic congestion, snoring, audible mouth-breathing during sleep) that isn't improving with practice; you notice the child has emotional pushback against the app that goes beyond ordinary boredom; or you see a sustained downward trend across several weeks rather than the normal week-to-week fluctuation. The clinician can review the data, adjust the protocol, or refer for ENT evaluation if anatomy is the issue rather than habit.
Biofeedback is one tool in a broader set, and clinicians frequently combine it with other approaches when the situation calls for it.
Myofunctional therapy. Direct strengthening of the lip seal, tongue posture, and oral-resting position. Myofunctional exercises and biofeedback work well together — myofunctional builds the muscular capacity for nasal breathing; biofeedback builds the moment-to-moment habit of using it. Many clinicians who refer to Nasal also prescribe myofunctional homework alongside it.
ENT/airway evaluation. Some children mouth-breathe because the nose is mechanically obstructed — enlarged adenoids, deviated septum, allergies that have been ongoing for years. No amount of behavioural work will fix this on its own. If the clinician suspects an anatomical or chronic-inflammatory cause, an ENT or airway-focused dental review may come first or in parallel.
Pediatric medical hypnosis. A small but growing pediatric subspecialty uses self-hypnosis and guided imagery to help children take direct control of habit-pattern conditions — bedwetting, tics, habit cough, IBS, and conditions with strong behavioural-mediation components. Practitioners like Jeff Lazarus (Menlo Park, CA — pediatrician with 25+ years of medical hypnosis practice) and Ran Anbar (pediatric pulmonologist, SUNY Upstate) report clinically meaningful improvement in 80% range across habit-pattern presentations after 2–3 visits. The approach pairs naturally with biofeedback: hypnosis works at the volitional layer (the child chooses the new pattern); biofeedback works at the contingency layer (the new pattern gets reinforced in real time). For families who already have a relationship with a hypnosis-trained pediatrician, layering the two is reasonable. The American Society of Clinical Hypnosis maintains a directory of practitioners; the National Pediatric Hypnosis Training Institute is the relevant North American training organisation.
Nasal stands on a long tradition of work in brain self-regulation through feedback. The clinical neurofeedback field — the broader practice of using real-time signals to teach the nervous system to find better states — has been carried for nearly four decades by Sue and Siegfried Othmer at the EEG Institute in Woodland Hills, California. Their work began in 1985 with the brain training of their son Brian and grew into the Othmer Method, which has trained thousands of clinicians worldwide and continues to evolve through Siegfried's research on infra-low frequency neurofeedback. While Nasal's method is peripheral biofeedback rather than EEG, the underlying conviction — that the body and brain can learn from contingent feedback in ways that don't require effortful conscious control — is something this field owes to the Othmers and the generation of clinicians and researchers around them.
Within pediatric airway and habit-pattern work specifically, the lineage runs through Karen Olness's foundational text on pediatric hypnotherapy, the myofunctional-therapy tradition that traces from Beatriz Padovan and Irene Marchesan, and the broader respiratory-feedback tradition documented in work by Ran Anbar and others at pediatric pulmonary centres. Nasal is not a synthesis of any single one of these — it is one specific implementation of a real-time biofeedback contingency for nasal breathing — but it is built on the assumption that all of these traditions point at the same underlying truth: behaviour shaped by reliable, immediate, well-timed feedback can re-pattern itself in ways that more verbal or cognitive instruction often cannot.
None of these adjuncts are required to use Nasal — the protocol stands on its own. They are options for the cases where biofeedback alone is producing slower-than-expected progress, or for clinicians who already integrate them in their practice.
The dashboard is the clinician-facing companion to the patient app. It receives anonymized event data from each session and renders it as charts, tables, and rule-based suggestions. This section walks through what each component shows and how to read it. None of the panels require interpretation in isolation — they're designed to triangulate.
The dashboard is designed for a desktop or tablet in landscape. It works on a phone too — but only in landscape orientation. Portrait is too narrow for the layout, and we deliberately don't try to make it fit; the dense data layout depends on horizontal space. If you open the dashboard on a phone and something looks broken or cramped, rotate to landscape first.
The dashboard's structural layout. Topbar across the top, patient list down the left, the main area on the right. The main area starts on Overview and switches to Patient detail when you select a patient — that's where the clinical reading happens, across the panels stacked above.
The dashboard has two layers. The overview grid lists every patient you've enrolled with at-a-glance metrics: recent sealed %, trend arrow, current phase, alert badges. Click any patient to drop into patient detail, which is where the clinical reading actually happens. The detail view stacks the panels you'll consult: header and actions, indices summary, by-modality table, session matrix, modality trends, suggested actions, session history, and the protocol panel. Most of your time will be in indices, suggested actions, and session history.
The indices summary panel is where the clinical reading begins. Nasal computes five composite indices organised into three tiers — tier 1 (clinical headline): Performance, Stability, Compliance; tier 2 (session quality): Engagement; tier 3 (downstream outcomes): Regulation. Each is on a 0–100 scale, higher is always better, and each answers a distinct clinical question. The full framework — what each index measures, the formulas, how to read them in combination, data-sufficiency thresholds, and the ceiling-effect caveat — is documented in its own section later in this manual (Indices — the three-tier framework). For now, the practical orientation: Performance and Stability are the two numbers a clinician reads first; the other three add depth once enough data has accumulated to make them meaningful.
Every index formula is versioned (e.g., performance_v1, stability_v1) and every clinical report is stamped with a bundle version, so case-study reproducibility is preserved across formula changes. The bands are deliberately wide for now; as cohort data accumulates the formulas will tighten and version. The dedicated indices section covers this in full.
One row per session type the patient has used: Video (passive viewing — the primary clinical reference), Reader (eyes-on-text, low-stim), Monitor (off-screen passive observation, rear camera, no UI), and Game (interactive sessions, including the fish-game). Each row shows total sessions, mean Sealed %, a 4-session trend arrow (↑ improving, → flat, ↓ declining), time since the most recent session in this modality, and a quality flag for the last three sessions.
The clinical use is identifying modality-specific weakness. A patient who scores 85% in Video but 50% in Reader has not yet generalised the skill — the contingency learning hasn't transferred from the Video context where it was trained to the Reader context where there's no comparable feedback. Extend sessions in the lagging modality, or escalate intensity for that activity type.
A heat map of the last 20 sessions in each modality. Rows are modalities; columns are chronological order. Green = sealed ≥ 80%, amber-green = 60–79%, orange-red = < 60%, empty = no session at this position.
Patterns to look for: a stable green row means a modality is solid; an oscillating row often means content variance or a phase transition; degrading rightward means recent sessions are worse than older — investigate; one bright row and others dim means cross-modality inconsistency, the same skill-transfer signal as the by-modality table but visualised over time.
Sealed % over time, one line per modality, with each session as a dot. Vertical dashed lines mark phase transitions in the patient's protocol with the new phase labelled. The clinical reading is whether sealed % moves systematically after a phase change — a sustained jump after the transition line is a positive signal worth recording in your session notes.
Watch for the moment a slower modality (often Reader or Monitor) catches up to Video. That's the visible signature of skill transfer — the breathing pattern is generalising beyond the originally-trained context. It is the goal of the protocol.
A panel of rule-based bullets that fire when specific data patterns are detected in the patient's recent sessions. Each bullet has an icon: ↑ (advance — the patient may be ready for the next phase), ! (concern — quality issue, plateau, inactivity), or i (info — pending parent reports, scheduled actions, neutral status).
Many bullets carry an action button. Advance phase opens the protocol panel pre-set to the recommended next phase; View sessions scrolls the page to session history; Review drafts opens the most recent pending parent report. Each rule has a clinical threshold under the hood — for example, "ready for discipline" fires when within-session retention has averaged ≥ 80% over the last 5 practice-phase sessions. The bullets are advisory, not mandatory; you decide whether to act.
All sessions for the patient, grouped by week. Most recent week is expanded by default. Filter by date range, modality, "issues only" (sessions with quality flags), or hide short (< 2 min) sessions.
Click any row to expand the session detail: full per-session metrics, the jaw-signal timeline, intervention-block summary if a template was running, and prompt-event records. The detail view is where you go when an index moved unexpectedly and you want to see the underlying single-session data.
The PDF link on each session generates a single-session report suitable for the patient's records or printing.
The configuration surface for an individual patient: clinical phase (baseline / practice / discipline / mastery), maximum intensity (capped at D by default; E enabled only if "Allow block level (5)" is on), session template (or "Use phase default"), auto-advance toggle, sessions-required thresholds, activity types, and Monitor sensitivity.
Changes apply to the patient's next session — the patient app loads the protocol at session start and uses that snapshot. A session in flight is not affected by mid-session edits.
Two distinct actions, intentionally separated. Deactivate is the soft-stop: the enrollment code stops working, the patient stops showing in alert and recommendation logic, but all their session data, reports, and protocol history are preserved. Use this when therapy ends, when a patient transfers to another clinician, or when a code is no longer needed but the record might be referenced later. Deactivated patients remain visible in the dashboard with reduced opacity, and a "Reactivate" button is available to undo.
The 30-day reactivation cooldown. Once a patient is deactivated, the system blocks reactivation for 30 days. This is a billing-integrity rule — it prevents cycling a patient on and off across billing periods to dodge the active-seat charge. The dashboard surfaces the exact reactivation-eligible date if you try to reactivate within the window. If you might want a patient active again within the month, simply leave them active rather than deactivating; their cost to you is the same either way. After 30 days, reactivation works normally.
Delete permanently is the hard-stop: the patient row and all dependent data (sessions, events, protocols, reports) are removed from the database. Used for legitimate test-data cleanup, data-entry errors, or in response to a parent's right-to-erasure request under GDPR. The action requires the patient to first be deactivated, and the confirmation modal asks you to type the patient's alias to proceed. There is no undo.
Where the buttons appear. An active patient's detail panel shows a single Deactivate button at the right of the actions row. After deactivation, that button is replaced by two: Reactivate (subject to the 30-day cooldown) and Delete permanently. The header also shows a "Deactivated" badge so you can see the state at a glance. The patient remains in the sidebar list with reduced opacity.
The default for any uncertainty is Deactivate. Once a patient has session data, that data has clinical-record value — preserving it is almost always the right call. Reach for Delete only when you have a specific reason to remove the data entirely.
The bullet rules will surface "ready for next phase" automatically. But the dashboard signals worth your manual attention are quieter:
For the full metric reference — formulas, units, what every base metric means in detail — see the Practitioner report metric reference, which is auto-generated from the report-metrics module and stays in sync with the engine.
Three PDF report types generate from the dashboard. Each serves a distinct audience and conversation. Briefly: the session report is a per-session archival document, useful for case files and granular review; the progress report is the parent-facing summary, designed to support the practitioner-parent conversation; the clinical summary is the longitudinal, research-grade document for your own reading and for case-report or retrospective work.
All three reports are generated server-side from the same data store. They contain only pseudonymous identifiers (alias + enrollment code). The clinician decides the cadence for parent reports — typically end-of-phase, monthly, or whatever schedule fits the case. Session reports generate on demand from the session history view. Clinical summaries generate on demand from the patient detail page.
A single-page (occasionally two-page) document covering one session. Generated from the session-history list with a per-row PDF action. Designed to be filed alongside notes from the visit it relates to, or referenced when reviewing a specific behavioural event in detail.
Schematic only. Actual reports use the patient's real data.
Patient block. Alias and enrollment code, age band if known, current phase. No PII.
Session key-values. Date and time of the session, modality (video / reader / game / monitor), phase the session was recorded under, duration, score, total mouth-open events, longest single open-mouth episode, the highest intensity intervention that was triggered (A silent → E block), and the template that ran. The template field is meaningful when you've used a custom or non-default block sequence — it tells you what within-session structure produced this data.
Behavioural signals (this session). The session-level view of the compliance and framing data captured during this session. Useful when you want to see exactly when and how the patient was responding to prompts during a specific session — the values feed the aggregates that appear in the progress and clinical reports, but here they are at the per-session granularity. Mean latency and mean persistence are derived from this session's prompt_events array.
Jaw timeline. A sparkline of the patient's jaw-open ratio across the session, sampled at 1 Hz. Dashed threshold line indicates the personal calibration cutoff above which the jaw is considered open. Peaks above the threshold are mouth-open events. Pattern reading: clustered peaks vs. evenly distributed peaks tells you whether the patient lost stability all at once or had a steadier session; height of peaks tells you how wide the patient was opening (taller = more emphatic, possibly speech vs. passive open-mouth).
Two pages. Designed to be readable in five minutes by a parent without clinical training, while still containing meaningful numbers for the parents who want detail. This is the report that goes home — it can be emailed, printed and handed over at a visit, or both. The cadence is your call; typical practice is end-of-phase plus monthly during long phases.
Period. The date range covered by this report. Auto-set to "since last sent report" for periodic drafts; you can adjust it before sending.
Patient block. Alias, code, sessions completed in the period. Age band omitted on the parent-facing report by default.
Where you are in the journey. The four-phase bar visualises which phase the patient is currently in, plus a count of sessions in this phase relative to the typical expected count for the phase. A short phase-keyed paragraph follows ("you are in the main therapy phase…", etc) — these come from STRINGS.parentJourney* and you don't normally edit them.
Summary numbers. Sessions, total practice time, average score, trend. Trend is computed as (recent third mean) − (early third mean). The threshold for the "rising / falling / steady" label is ±10 points. Holds the parent's attention to the directional signal rather than the absolute number, which they should not over-anchor on.
Score trend. Per-session sealed % over the period, plotted chronologically. If you have fewer than four sessions in the period, the chart renders as a placeholder note.
Performance by modality. Inherits from the modality matrix you read in the dashboard. Reduced to parent-readable form — sessions, mean sealed %, simplified trend arrow (rising / steady / falling).
Behavioural signals. Two short paragraphs and concrete key-values. The compliance paragraph reports prompts received, compliance rate (closed within 2s), typical response speed, and typical hold-after-closure. The framing paragraph reports good-framing percentage and lean-in / head-turn event counts. These are the same numbers that drive the clinical-PDF version of this section; they are presented here without distribution histograms because parents don't need that level of detail.
What this shows. Plain-language inline-computed findings — 3 to 4 short >-marked observations derived from the period's data. Findings cover score trend, phase pacing, compliance pattern, and framing quality. The voice is descriptive ("Your child is doing well at responding to the reminders"), not diagnostic — interpretation is anchored to the data, action belongs in the practitioner notes and the conversation.
Practitioner notes. Your free-text block. The dashboard generates a draft that you edit before sending. The draft uses the tree-stage metaphor (Seedling / Sapling / Young tree / Tall tree) and avoids citing the Performance / Stability indices directly — parents don't get index numbers in their report by design.
Closing line. A brief encouragement line drawn from STRINGS.progressEncouragement. Editable in code; the same line currently ships across all reports.
Seven pages. Designed for your own reading and for case-report or retrospective work. Density is QIK-style — research-defensible structure with raw data, derived metrics, distributions, and rule-based interpretation. Generated on demand from the patient detail page in the dashboard. Not sent to parents by default; you may choose to share specific pages or excerpts in clinical conversation, but the report's voice and density assume a clinical reader.
Page 1 — results summary. Patient header strip with alias, code, sessions, age band, enrollment date, last-session date. Two index panels — Performance and Stability — each with their decomposition (Performance: sealed % at 60%, latency at 20%, retention at 20%; Stability: cross-modality consistency at 60%, phase progression rate at 40%). Per-session derived metrics table covering events per minute, longest sealed and open streaks plus the closed-to-open ratio, and latency to first event. Every row includes its formula inline so the reader can verify what the number means.
Page 2 — detail by modality and phase. Modality matrix with sessions, mean sealed %, PI scored per modality, and trend slope per modality (pp/session). Treatment phase visualisation with the four-phase bar and the patient's sessions-in-phase relative to the typical expected count. Session statistics — mean and median score, improvement (early-to-recent delta), and the linear regression trend slope across the full history. The ceiling caveat appears here as well — when the headline metric is saturated, the trend numbers are explicitly noted as non-informative and the reader is redirected to Stability and Monitor data.
Page 3 — trajectory. Three views of the data over time. Score trend over time is the per-session line chart. Event count development is the bars chart of mouth-open events per session, ordered chronologically — useful as a coarse complement to the composite score trend (a falling score with a falling event count is real improvement; a falling score with stable event count is worth investigating). Score distribution is the binned histogram of session scores — particularly informative for ceiling cases where the trend is flat but the distribution may reveal occasional dips or bimodal patterns.
Page 4 — per-session appendix. One row per session in chronological order. Columns: date, modality, duration, score, total events, mean recovery latency (prompt → close), longest closed streak, longest open streak, yawn count, frame quality (with check / partial / issue marker plus percentage), phase the session was recorded under, and compliance % for that session. This is the receipts table. Every summary statistic in the report can be independently verified from this page. For research / case-report use, this is the page that needs to be cited.
Page 5 — compliance prompts. Four headline stats: total prompts (with gentle/firm breakdown), compliance % (closed within 2s), mean latency, mean persistence-after-closure. Two side-by-side distribution histograms: latency-to-close and persistence-after-closure, both filtered to populated bins for sparse-data clarity. Sample sizes in each titlenote. The persistence histogram is the consolidation signal — right-shifting across sessions = the closed state is becoming the default rather than something the patient has to actively maintain.
Page 6 — ergonomic framing. Four channel stats: too-close events (postural concern, fires a real-time prompt), face-missing events (data-quality only), head-turn events (regulatory / restlessness signal, observational only), frame quality % across the period. Head-turn-events-per-session trend chart underneath — falling trends often parallel the breathing consolidation as the nervous system settles, but this is a *confirming* indicator only, not a replacement for the primary indices.
Page 7 — interpretation. Findings in five sections, each with one or more >-marked observations and an explanatory paragraph below. I. Overall covers the two indices and the score trajectory. II. By modality surfaces between-modality gaps and individual modality trends. III. By phase reports pacing relative to the expected sessions-per-phase. IV. Notable patterns covers the closure envelope (longest closed / longest open / ratio), distribution shape findings, and the ceiling-aware redirection if applicable. V. Behavioural is the inline-computed findings layer over the compliance and framing data — compliance-rate findings stratified at 60% / 85% thresholds, persistence findings at 10s / 30s thresholds, frame-quality findings, and head-turn early-vs-late trend findings.
Nasal exposes five composite indices, organised into three tiers. Each tier answers a different clinical question, and each gets its own visual placement in the reports and dashboard. The tier separation matters: collapsing all five into one "performance" line would obscure the distinction between how the patient does in sessions, how good the session conditions were, and how downstream regulation is shifting — three different phenomena that deserve different reading-states.
The three numbers a clinician reads first when opening a patient's record. They answer the core clinical question: how is this patient doing? Surfaced at the top of the patient detail panel in the dashboard and on page 1 of the clinical PDF.
What it measures. In-session compliance quality: how much of the session the patient kept their lips sealed, how fast they recovered when they didn't, and how well they sustained compliance across the session arc. Direct measure of "are they doing the work."
Formula (performance_v1). 0.6 × sealed% + 0.2 × latency-component + 0.2 × retention-component. Sealed% is pass-through (already on a 0–100 scale). Latency-component normalises the mean correction-latency-after-prompt to a 0–100 scale (5s → 0, 0ms → 100). Retention-component is the within-session retention percentage (first-half to second-half compliance comparison). Weights chosen to anchor the index in the primary closure measure (sealed%) while keeping the response-loop and within-session-sustainment signals visible. Computed server-side in the fetch_patient_indices SQL RPC.
How to read it. ≥80 is the on-track range. 60–79 is the monitoring range — watch the trajectory. <60 is below expected and warrants a direct review of protocol fit, content engagement, or technical setup. For ceiling cases (patients consistently at 95–100), the absolute value carries less signal than the Stability Index and Monitor-mode data.
What it measures. Whether the in-session compliance is generalising — across modalities (does video performance match reader performance match game performance?) and across phases (is the patient pacing through the protocol roughly on schedule?). Performance answers "are they doing the work"; Stability answers "is the skill durable and transferring."
Formula (stability_v1). 0.6 × cross-modality consistency + 0.4 × phase progression rate. Cross-modality consistency penalises gaps between modality means (a patient at 95% in game and 60% in reader has lower consistency than one at 82% in both). Phase-progression rate is the ratio of completed phases to expected phases given time-since-enrollment. Computed server-side.
How to read it. ≥80 is high consistency — the skill is generalising, and progression is on schedule. 60–79 is moderate; check the modality matrix to see which contexts are lagging. <60 suggests the skill is content-dependent or the patient is stuck in a phase, both worth a direct intervention review.
Reading Performance and Stability together. The two headline indices answer different questions — Performance is how well overall, Stability is how consistent across contexts — and the clinically useful information is often in the combination rather than either number alone:
| Stability high | Stability low | |
|---|---|---|
| Performance high | Consolidating well — the protocol is working as intended. | Doing well in some contexts but not generalising. Check the modality matrix for which context is lagging; extend sessions there. |
| Performance low | Steady-state struggling — consistent, but consistently below target. Review protocol fit. | Below expected across the board. Escalate intensity, and check non-app factors (illness, sleep, home setup). |
The high-Performance / low-Stability cell is the one most easily missed: the headline number looks fine, so nothing draws the eye, but the skill isn't transferring across modalities. Stability is the index that catches it — which is the whole reason it sits in the headline tier rather than being folded into Performance.
What it measures. Cross-session quality of the response-to-prompt loop: how quickly the patient closes their mouth when prompted, how durably the closure persists, and whether gentle prompts are sufficient or firm escalation is being needed. This is the "are they learning to respond" question, distinct from Performance's "are they doing the work" question — a patient could have high Performance (high sealed%) with low Compliance (slow to respond when they slip), or vice versa.
Formula (compliance_v1). 0.5 × latency-norm + 0.3 × persistence-norm + 0.2 × (1 − firm_ratio), computed per session and then EMA-weighted across sessions (half-life 5 sessions). Latency-norm rescales mean prompt-response latency 1500ms → 0, 200ms → 100. Persistence-norm rescales mean hold-after-closure 0ms → 0, 30000ms → 100. The firm_ratio inverse rewards sessions where gentle prompts sufficed.
The EMA weighting is intentional and distinguishes Compliance from Performance/Stability. Performance and Stability answer trajectory questions and use flat-mean across history. Compliance answers a current-state question — how is this patient responding now — so recent sessions are weighted ~2x older ones at 5-session intervals. A patient who needed many firm prompts in week 1 but is now responding to gentle prompts in week 8 is currently highly compliant; flat-mean would understate this.
How to read it. Read alongside Performance. High Performance + high Compliance = consolidating well. High Performance + low Compliance = compliant during sessions but brittle when prompted; watch for sustained closure to develop. Low Performance + high Compliance = engagement issue rather than learning issue; check content. Low both = the protocol isn't producing response-to-stimulus learning yet; conversation about engagement, content, or protocol fit.
Data sufficiency. Compliance requires prompts to have fired. A patient who has done many sessions but never triggered a prompt (rare; usually a ceiling case) has no Compliance value. See the confidence section below for the threshold rules.
The middle tier addresses a process question rather than a clinical-outcome question: were the sessions readable enough for the clinical data to be trustworthy, and was the patient psychologically present during them? A surprisingly common failure mode is misreading a clinical-decline signal that's actually a setup-degradation signal — the patient's data got worse because their attention slipped or their camera setup drifted, not because their breathing pattern regressed. Tier 2 distinguishes these.
What it measures. A composite of camera-setup quality (was the device positioned well; was the patient framed) and attention stability (was the patient looking at the screen vs. turning their head frequently). Frame-quality alone tells you whether the data is technically readable; head-turn rate alone tells you whether the patient is psychologically present. Together they distinguish four cases: engaged + good setup, present-but-distracted, engaged-but-bad-setup, and checked-out.
Formula (engagement_v1). 0.5 × frame-quality% + 0.5 × attention-score, flat-mean across sessions with framing data. Attention-score normalises head-turn rate per minute on an inverse scale: 0/min → 100, 5/min → 0. The 5/min cap comes from clinical inspection of pilot data; expected to refine post-pilot.
How to read it. ≥80 means session conditions are clean and the breathing data can be interpreted without caveats. 60–79 means moderate session quality — interpretable but worth checking whether one of the two components is dragging (frame quality drop usually points to home setup; attention drop usually points to content or fatigue). <60 means session quality is questionable enough that clinical signals from those sessions should be interpreted carefully.
The bottom tier addresses a different clinical question entirely: is the patient's autonomic state shifting as nasal breathing establishes? This is a downstream consequence of the protocol, not a measure of in-session performance. The literature on pediatric mouth-breathing and orofacial development describes a consistent pattern: as nasal breathing establishes, parasympathetic activation increases, restlessness decreases, attention regulation improves, and sleep architecture shifts. These are different outcome variables than behavioural compliance with the protocol — a patient could have high Compliance (responding well to prompts) and low Regulation (still restless, still seeking distraction) early in treatment. The distinction is clinically meaningful.
Tier 3 currently has one member because it's what we can compute from existing data. It is designed to expand: sleep-quality measures from PSG or wearables, parent-reported behavioural measures, clinician-rated facial growth measures — all eventually sit in this tier as they become available.
What it measures. Head-turn frequency as a proxy for restlessness, combined with the early-vs-late trend in head-turn frequency as a proxy for the autonomic shift. A patient who started restless and is now calm has high Regulation (autonomic shift has happened). A patient who was always calm has high Regulation (state is good even without a shift). A patient with rising head-turn frequency has falling Regulation (regression — worth investigating).
Formula (regulation_v1). 0.4 × current-calmness + 0.6 × improvement-signal. Current-calmness inverse-normalises the EMA-weighted overall head-turn rate (5/min → 0, 0/min → 100). Improvement-signal compares early-third-of-history vs. late-third-of-history mean head-turn rate; the delta is rescaled −2/min (regression) to +3/min (strong improvement) onto 0–100. Weighted toward improvement because the autonomic-shift signal is the change. Re-normalises over available components when the improvement signal can't be computed (early sessions).
How to read it. Read alongside Compliance and Performance, not in isolation. Regulation is a confirming indicator: it tells you whether the work the patient is doing in sessions is producing the downstream autonomic effect the protocol targets. High Compliance + rising Regulation = consolidation arc on schedule. High Compliance + flat Regulation = the patient is responding to prompts but the autonomic shift isn't visible yet; this is normal early; watch over time. Compliance high but Regulation falling = unusual; check for protocol intensity, sleep, life events.
Important caveat. Regulation is one signal. The literature points strongly enough that we believe it's worth surfacing, and pilot data will refine its sensitivity. But low Regulation alone is not a clinical concern; it's a flag to look at the broader picture. As more downstream measures land in Tier 3, Regulation becomes one of several converging signals rather than the single autonomic readout.
Each of the three new indices (Compliance, Engagement, Regulation) carries a confidence flag based on how much data has accumulated for that specific index. Below the threshold, the index displays as "insufficient data" rather than a number — showing a tight 87% Compliance for a patient with only two recorded prompts would be misleading.
| Confidence | Compliance threshold | Engagement threshold | Regulation threshold | Display |
|---|---|---|---|---|
| high | ≥20 prompts | ≥10 sessions w/ framing | ≥10 sessions w/ framing | Number, no caveat |
| medium | 10–19 prompts | 5–9 sessions | 6–9 sessions | Number, full formula |
| low | 5–9 prompts | 3–4 sessions | 3–5 sessions (absolute only, no trend) | Number + "limited data" note |
| insufficient | <5 prompts | <3 sessions | <3 sessions | Hidden — index not displayed |
Performance and Stability do not carry the same confidence flags currently — they have always been computed across whatever sessions exist and have always been displayed regardless. This is a minor inconsistency worth noting: a patient with one or two sessions has a Performance Index displayed but no Compliance Index. The reasoning is that Performance and Stability are primary anchors that clinicians expect to see; Compliance is a secondary tool whose value comes from having enough prompt data behind it. As the index framework matures, expect Performance and Stability to also surface confidence flags in v2.
Even though the flags aren't enforced on the display yet, Performance and Stability should be read with the same caution. The interpretive tiers:
A Performance Index of 92 from 25 sessions and a Performance Index of 92 from a single session are not the same clinical object. Until the display carries the flags, the clinician carries them mentally.
Every formula is named with a version: performance_v1, stability_v1, compliance_v1, engagement_v1, regulation_v1, plus a bundle version 1.0 stamped on every clinical report. The clinical PDF carries a small footer note: "Indices computed under formula version 1.0."
The reason: retroactive comparability for case reports. If a paper or case-report is written off these reports today, and the formulas later evolve (e.g., we discover post-pilot that 50/25/25 weights perform better than 60/20/20 for Performance), the historical reports stay computable under performance_v1 while new reports compute under performance_v2. A patient's trajectory remains reproducible across formula evolution; published case-reports remain verifiable.
The discipline is: never silently overwrite a versioned formula. Changes to weights, scaling endpoints, or component composition produce a new version alongside the old one; the old version stays computable indefinitely.
The three-tier architecture is partially surfaced as of v0.12. To be transparent about the current state:
The next implementation step is bringing the three new indices to the dashboard patient-panel surface with the same three-tier visual separation used in the PDF. Until that ships, the clinical PDF is the primary place to see Compliance, Engagement, and Regulation.
This section addresses the clinical positioning of Nasal: who it is for, who it is not for, what it can and cannot address, when other interventions should come first, and what oversight the deploying clinician retains. The framing throughout is conservative — Nasal is a clinical-grade biofeedback prototype, not a medical device, and the prescribing clinician carries the responsibility for appropriate selection and ongoing monitoring.
Nasal is designed for children and adolescents with habituated mouth-breathing where the underlying airway is functionally patent — that is, the child can breathe through the nose, but doesn't, by habit, by postural drift, or by acquired pattern. The intervention assumes the airway is mechanically capable of supporting nasal respiration at rest and during light activity. Where this assumption fails, the intervention will not produce its intended effect and may produce frustration without benefit.
The age range Nasal is built for is roughly 4 to 16 years. The lower bound is set by attentional and cooperation requirements: a child needs to sit through a calibration sequence and watch content for a few minutes at a time, which is age-appropriate from around four onward. The upper bound is set by motivational and developmental considerations: by mid-adolescence, the appropriate intervention is usually a more agentic one — myofunctional therapy with self-directed practice, conscious lip-seal training, or referral if anatomical contributors persist. Adults can use the system, and some do, but the system was not designed around their motivational profile and the protocol may need clinician-led adjustment.
Within the 4–16 range, the children most likely to benefit are those with: a clear daytime mouth-breathing pattern at rest, no current acute upper-respiratory infection, no untreated airway anatomy contributors, parental engagement sufficient to support daily sessions for several months, and at least minimal cooperation with an unsupervised tablet activity. Where any of these conditions is unstable, expect slower progress and more frequent protocol adjustment.
Nasal targets the habit pattern of mouth-breathing. It is well-suited to:
Nasal is not designed to address — and should not be the primary intervention for:
A baseline airway and ENT-orientated evaluation is recommended before enrolling any child in Nasal whose history or examination suggests possible anatomical or chronic-inflammatory contributors. The threshold for referral should be lower rather than higher; addressing anatomy first is almost always the right ordering. Specific triggers for pre-enrolment referral:
A child who passes a basic patency assessment and whose history and examination do not suggest the above can reasonably begin Nasal without further evaluation. A child with any of the above triggers should have those addressed first, after which Nasal becomes appropriate.
Hard contraindications — Nasal should not be used:
Relative contraindications — caution and clinician judgment:
Nasal is delivered through screen-based content, and that warrants honest discussion. The protocol prescribes daily sessions of typically 10–20 minutes; this falls within ranges that paediatric guidelines consider acceptable for younger children when content is purposeful, supervised, and time-limited. Several aspects of the design address common screen-time concerns:
That said, Nasal's screen-time should be considered additive to the family's overall screen-time accounting. A reasonable framing for parents: Nasal sessions count as screen time, but they are productive screen time, and most families can absorb 15 daily minutes of clinically-directed video viewing without disrupting their broader screen-time norms. Where a family has very strict no-screen policies, consider whether Reader-based sessions or game-based sessions can carry the protocol load instead.
Nasal is non-invasive and does not deliver any therapeutic agent, electrical stimulation, or physical force. The clinical risk profile is correspondingly low. Specific safety considerations:
No serious adverse event from the use of feedback-mediated screen dimming for nasal-breathing training has been documented in the broader biofeedback literature. Nasal-the-product is in pre-pilot phase; ongoing safety monitoring is part of the pilot protocol and any findings will be reflected in subsequent revisions of this document.
Nasal is a clinical biofeedback prototype, not a medical device. It is used under the prescribing clinician's supervision, with the clinician's protocols, the clinician's decisions about progression, and the clinician's direct relationship with the patient and family. This positioning has practical implications.
A reasonable cadence for active dashboard review during the early weeks of a patient's protocol is weekly; once the child is in stable consolidation, every two-to-four weeks is usually adequate; in mastery and beyond, monthly review and as-needed responsive contact is appropriate. The clinician sets the rhythm based on the patient and the practice's patterns.
Nasal is appropriate for children 4–16 with habituated daytime mouth-breathing in a functionally patent airway, where parents can support daily sessions and the child can tolerate the screen-based feedback. It is not appropriate as a primary intervention for anatomical obstruction, sleep-disordered breathing, or active inflammatory airway disease, all of which need to be addressed first. The clinician owns selection, protocol design, monitoring, and the family relationship. The system supports clinical work; it does not replace clinical judgment.
What the parent reads in the dedicated parent guide is summarized here so clinicians can speak the same language. For full depth, including all the reassurance and parenting-context material, refer parents to the parent guide directly. This section is a clinician's compressed map of what parents are looking at.
When a parent opens the Nasal app on their child's device, they see a single screen with a small set of tiles: YouTube, Library, Device videos, Reader, and (once enabled) Games and Monitor. Each tile starts a session in that modality. There is no "start session" button — the act of choosing a tile is the start.
During a session, the parent sees almost nothing. The child sees the chosen content. If the mouth opens, a gentle dim and a small text prompt appear on-screen; if the mouth stays open, a firmer prompt and a deeper dim follow. When the lips close again, everything returns to normal. The child experiences this as "the video pauses when my mouth opens"; the parent experiences it as a brief flicker if they're watching from across the room.
A small bottom panel during sessions shows elapsed time, event count, and an "End" button (gated by parent PIN if set). Most families never touch this panel — they just let the session run.
In the app's settings, parents can see (and edit, though rarely should) a panel showing the two text prompts ("Close your lips" / "Breathe through your nose"), a firm-prompt delay, a Response dropdown (default "Pause video + blur"), and an Audio cue on trigger toggle. These settings predate the ABCDE rhythm-pattern protocols and are now operationally minor — the response patterns the patient actually experiences are driven by the protocol you set in the dashboard, not by these app-side toggles. The settings remain visible for two purposes: (a) parents in non-English households can rewrite the prompt text in their language, and (b) the Audio cue on trigger is reserved for a future feature, currently off.
If a parent asks about these settings, the simplest answer is: "Those are an older surface. The actual response is managed from my dashboard. The only thing you might want to change is the prompt text if you'd like it in your language."
Two optional PINs, both four digits, both set by the clinician from the dashboard (parents cannot set or reset PINs themselves). The child PIN gates profile-switching on devices shared between siblings; the parent PIN gates session-ending and profile-removal. Most families don't need the child PIN; many find the parent PIN useful once the child becomes capable enough to skip sessions. PINs are hashed server-side (SHA-256) and stored on the device only as a local boolean (PIN-set / not-set). Forgotten PIN is reset by the clinician from the dashboard — there is no email-based recovery path. Talk to the parent at enrollment about whether either PIN suits their household, and document the choice; revisit at any phase transition.
Parents see the four phases as Baseline → Practice → Discipline → Mastery in the dashboard summary that comes back in their reports, and as a growing tree in their child's view (Seedling → Sapling → Young tree → Tall tree). They are told that Baseline is invisible measurement, Practice is gentle introduction of prompts, Discipline is the firmest phase where the bulk of behavioural change happens, and Mastery is the dial-down phase that confirms the new pattern holds without contingent feedback. They are also told that you, the clinician, decide phase transitions — they don't.
Parents are told explicitly: do not become the breathing police. The whole architecture of Nasal is built to keep the parent-child relationship out of the correction loop. They are told to set the routine (daily session, same time of day if possible), make sure the device is charged and positioned reasonably, and otherwise step back. They are explicitly told not to coach during sessions, not to praise sealed lips, not to call out open lips, and not to comment on app numbers in their child's hearing. If the parent has been doing breathing reminders before Nasal, they are told that they can now stop — the app does that part.
They're told to contact you when: the child is resisting sessions for more than a few days; sleep quality changes noticeably (worse OR markedly better); allergy or congestion symptoms appear; school or daycare reports a change; or anything else they find unsettling. They are told not to wait for the next scheduled visit if any of these come up. They are told that they can also ask for Monitor sessions to be enabled earlier than the default — it's the clinician's call, but parents know they can request it.
The dashboard generates a written report you review and send to the family. Today, drafts are created automatically when you advance a patient to a new phase — the system inserts a draft row triggered by the phase transition, with the period framed around the time since the last sent report. You open it from the Parent reports panel, the dashboard fills in the headline, "what's working" body, and "what we're focusing on next" paragraph via an RPC, you review and edit, then send via the notify-parent-report edge function which emails the parent.
4-week periodic auto-drafts are not currently firing. The infrastructure for them is built (the ensure_periodic_parent_report_drafts RPC exists in the schema), but nothing currently invokes it — neither the dashboard on load nor a scheduled cron. So between phase transitions, no draft will appear on its own. If you want to send an interim update for a stable patient, the current path is to advance them through a phase or, if no advancement is due, contact the family directly without using the report system. Wiring up the periodic trigger is on the roadmap and will land in a future release without changing the review/send flow you already use.
The draft uses the tree-stage metaphor and plain-language descriptions of what changed in the period. Numbers (Performance Index, Stability Index, modality breakdowns) are intentionally kept out of the parent report; if a parent asks for them, that's a clinical conversation, not a routine handoff.
Working definitions of terms used throughout this manual, the parent guide, the dashboard, and the patient app. Where a term has a specific operational meaning inside Nasal that differs from common usage, the Nasal-specific meaning is given here.
The initial three sessions. All events fire level A — silent logging, no feedback delivered to the patient. The purpose is to establish each child's natural rate of mouth-breathing during screen time, against which subsequent phases are compared.
The second clinical phase, default five video sessions. Intervention escalates within a session along an A → B → C → D ramp. Trains the child to associate gentle reminders with firmer ones and to self-correct before the firmer ones arrive. (Renamed from "graduated" in the April 2026 vocabulary update.)
The terminal therapeutic phase. Every mouth-open event fires level D (heavy dim and pause). This is the operating mode for the bulk of therapy, typically several months. Auto-advance does not progress past discipline. (Renamed from "full" in the April 2026 vocabulary update.)
An optional clinician-set phase used when sustained nasal breathing is observed in discipline. Level D fires for the first half of each session, level A for the second — the second half functions as a within-session probe of whether the learned behaviour persists when the contingency is absent. Not part of the auto-advance ladder. (Renamed from "weaning" in the April 2026 vocabulary update.)
The event is logged but no feedback is delivered to the patient. Used for baseline measurement and for washout blocks inside templates.
Gentle text overlay on the video without dimming or pausing; brief haptic pulse on supporting devices. The child can ignore it and the video continues.
Light dim (≈ 6 px blur) while the video keeps playing. Reminder text becomes more salient. Designed to be noticeable without being aversive.
Heavy dim (≈ 14 px blur) and pause of the video. Firmly worded reminder. The session resumes the moment the mouth closes. The therapeutic intervention level.
Heaviest blur (≈ 24 px), pause, and a fully blocking overlay. Off by default — only fires for patients whose protocol has "Allow block level (5)" enabled. Reserved for cases where the child has demonstrably learned to ignore lower levels.
A fixed sequence of timed blocks attached to a patient's protocol, each block running a specified intervention level. Templates override phase-based intervention selection on a session-level basis. Built-in presets include Silent baseline, ABACA, ABA, BAB, and Discipline (sustained). Custom templates can be created via the dashboard's template editor.
One segment of a template, defined by a label, duration in minutes, and an intensity level (A–E). The patient app automatically transitions between blocks at their scheduled times; the kid sees no boundary marker.
A classical within-session experimental design: silent baseline (A) → notice (B) → silent return (A) → dim/intervention (D) → silent washout (A). The pattern lets the clinician read out, in a single session, how the child responds to escalating cues and whether the response carries across the silent washout. The internal letter pattern is structural, not the same as the A–E intensity vocabulary, although both happen to use letters.
A discrete moment where the patient's measured jaw-open value crosses above the calibrated threshold and stays above for long enough to be classified as a real opening (rather than speech, a yawn artefact, or a transient face-tracking glitch). Each event has a start time, end time, peak jaw-open value, and a classification (mouth-breathing event, yawn, speech, etc.).
Proportion of session time during which the mouth was closed below the calibrated threshold. Reported per session, and the largest single component (60% weight) of the Performance Index. Higher is better.
A continuous numeric output from the on-device face-tracking model, from 0 (fully closed) to 1 (maximally open). Calibrated per-session to a personal threshold during the calibration step, since natural resting mouth posture varies between individuals.
The technical term for a single facial-feature signal output by the face-tracking model. Jaw-open is one blendshape; eye-squint, mouth-pucker, and dozens of others exist. Nasal uses a small subset of these for event detection and yawn filtering.
A heuristic that classifies a wide-mouth-open event as a yawn (rather than a mouth-breathing event) based on duration, peak amplitude, and concurrent eye-region squinting. Yawns are logged separately and excluded from the prompt-firing logic, since they are reflexive rather than habitual.
A per-session indicator of whether the session telemetry was usable: face was in frame consistently, distance was in range, lighting was sufficient, calibration succeeded. Sessions with quality issues are still recorded but tagged so the clinician can weight them appropriately. The dashboard's session matrix and the by-modality view both surface quality flags.
Tier-1 clinical-headline index. 0–100 score summarising in-session compliance quality: 0.6 × sealed% + 0.2 × correction-latency-component + 0.2 × within-session-retention-component. The primary "how is the patient doing" number. Formula version performance_v1.
Tier-1 clinical-headline index. 0–100 score for whether in-session compliance is generalising: 0.6 × cross-modality consistency + 0.4 × phase-progression rate. Catches the patient who performs well in one modality but hasn't transferred the skill. Formula version stability_v1.
Tier-1 clinical-headline index. 0–100 score for the cross-session quality of the response-to-prompt loop: 0.5 × latency-norm + 0.3 × persistence-norm + 0.2 × (1 − firm_ratio), EMA-weighted across sessions (half-life 5 sessions). Answers "is the patient learning to respond," distinct from Performance's "is the patient doing the work." Formula version compliance_v1.
Tier-2 session-quality index. 0–100 score combining camera-setup quality and attention stability: 0.5 × frame-quality% + 0.5 × attention-score. Distinguishes a clinical-decline signal from a setup-degradation signal. Formula version engagement_v1.
Tier-3 downstream-outcome index. 0–100 score using head-turn frequency as a restlessness proxy and its early-vs-late trend as an autonomic-shift proxy: 0.4 × current-calmness + 0.6 × improvement-signal. A confirming indicator, read alongside the tier-1 indices rather than in isolation. Formula version regulation_v1.
A per-index data-sufficiency indicator — high, medium, low, or insufficient — based on how much data has accumulated for that specific index. Below the insufficient threshold, an index displays as "insufficient data" rather than a potentially misleading number. Currently surfaced on the display for the tier-2 and tier-3 indices; Performance and Stability carry interpretive tiers documented in the manual but not yet enforced as display flags.
Feedback that arrives precisely at the moment of the target behaviour, every time it occurs, and stops the moment the behaviour stops. Distinguished from after-the-fact instruction (e.g., a parent telling a child to close their mouth ten minutes later) and from non-contingent feedback (e.g., a generic reminder app firing on a schedule). Operant conditioning research consistently finds contingent feedback to be the active ingredient in habit-reversal therapies.
The persistence of a learned behaviour into a context where the contingency is no longer being delivered. Within-session carryover is what the silent washout block of an ABACA template measures: did the child stay closed during the A-block at the end, after the D-block did the contingency-delivery work? Across-session carryover is what the Mastery phase tests at a longer timescale.
Specifically: the second-half sealed % minus the first-half sealed %. Positive = the patient improved across the session. The clinical advancement rule "ready for discipline" looks at a 5-session rolling average of this metric.
The kind of activity the session ran in. Four are currently supported: Video (passive viewing), Reader (eyes-on-text), Monitor (off-screen, rear-camera passive observation), and Game (the fish-game and other interactive activities). Modality matters clinically because skill transfer is the goal — the patient needs to maintain nasal breathing not just in the originally-trained Video context but in the others.
The child-facing visual reward — Seed, Sprout, Sapling, Young tree, Tree, Mighty tree, Forest. Driven by total session count across all modalities. Independent of clinical phase. The tree's job is engagement (keeping the child returning to the app for months); the phase's job is therapy (measuring and shaping the breathing pattern).