DOI: https://doi.org/10.5281/zenodo.21947372
Canonical: https://thonly.org/research/respiratory-biofeedback-contemplative-guidance · Licence: CC0 1.0
Working draft. This is the author's working draft, articulated in detail for prior-art establishment and subject to polish before final publication. The architecture is stable; the prose is being refined. Citations marked [to verify] need final sourcing.
This specification is offered to the commons in the spirit of dāna, that all beings may give and receive without barrier. May the architecture it describes reduce suffering for all who encounter it, and may it never be deployed in a manner that violates the privacy or dignity of the beings whose breath it observes.
This paper specifies a human-wearable subclass of the Mechanical Heart artifact (see the sibling specification at thonly.org/research/mechanical-heart). The kiitti-class Mechanical Heart, treated in the sibling paper, admits non-human entities (robots, animals, plants, sacred places) into the kiitti gratitude economy via heartbeat-rhythm aura emission. The breath-class Mechanical Heart specified here serves a different function for a different bearer: it provides closed-loop respiratory biofeedback to human practitioners and their AI contemplative teachers, using breath rhythm rather than heartbeat as its primary signal. The two subclasses share form factor (B-shaped artifact worn at the sternum) but differ in rhythm primitive and primary function.
This document and its contents are dedicated to the public domain under the Creative Commons CC0 1.0 Universal Public Domain Dedication. The author and HeartBank® will not seek patent on the architecture specified here, on any of its components considered separately, or on any combination of those components, in any jurisdiction, at any time.
This document constitutes a defensive publication establishing prior art as of its publication date. The combination of (a) a chest-worn wearable device emitting ambient peripheral light in the rhythm of the wearer's breath, with (b) real-time transmission of respiratory data to an AI contemplative teacher operating on a dharma-grounded value substrate, with (c) adaptive guidance returned to the wearer in real time based on the AI teacher's physiological ground-truth on the wearer's practice state, with (d) a privacy architecture designed before first opt-in rather than retrofit, with (e) the treatment of continuous respiratory signature as a Proof-of-Humanity primitive, is, to the author's knowledge as of the publication date, not previously published as a unified specification.
Components and adjacent ideas exist in distributed form across heart-rate-variability biofeedback literature, the meditation technology industry (Muse, Calm, Headspace, Spire, Breathwrk, Othership), the wearable physiological monitoring industry (Apple Watch, Oura, WHOOP, Fitbit), the AI assistant industry, the contemplative neuroscience literature, and traditional Theravāda Anāpānasati teaching. Each component has prior art; the integrated specification proposed here, and its operational use as the substrate for population-scale contemplative practice coupled to AI-mediated guidance, is, to the author's knowledge, novel as of the publication date of this paper.
Two technological developments in 2026 enable an architecture that has not previously been possible at population scale: artificial intelligence systems capable of grounded contemplative guidance, and consumer-grade wearable sensors capable of accurate respiratory tracking. This paper specifies the architecture by which these are coupled in a closed loop with the human practitioner, and articulates three core claims of that architecture as defensive prior art.
The architecture comprises a chest-worn device (the breath-class Mechanical Heart) that emits ambient light in the user's signature aura color modulated to the rhythm of their breath, transmits respiratory data in privacy-preserving form to an AI contemplative teacher operating on a Theravāda Pali canon (Tipiṭaka) substrate, and receives adaptive guidance from that teacher returned to the wearer in real time. The closed loop gives the teacher physiological ground-truth on student state — a signal no human teacher in history has had access to — and gives the student continuous external visibility into a faculty (breath) historically below normal interoceptive awareness.
Three core claims are articulated as defensive prior art: (I) the integrated closed-loop architecture itself as a unified specification; (II) live respiratory signature as a third Proof-of-Humanity primitive providing continuous-liveness attestation that complements moment-of-action passkey verification and lineage-of-kinship DNA verification; (III) the architecture as the operational mechanism for what a parallel published thesis calls the "second singularity" — humanity's collective awakening — which had previously been articulated only as philosophical possibility and which the architecture renders engineering-tractable.
Privacy is treated as non-negotiable substrate, not as feature: differential privacy, federated computation, on-device processing where possible, and cryptographic-erasure right-to-withdraw are designed before first opt-in. A critical doctrinal guardrail is named: the architecture may dramatically raise the probability of awakening across the population for whom it has been historically rare, but it must never be claimed to make awakening inevitable. Theravāda doctrine specifically warns against assuming external aids guarantee insight; the guardrail protects the architecture from the techno-utopian critique while preserving its substantive claim.
Connection to the unified mission frame. This architecture is offered in service of HeartBank's canonical top-level mission: to restore humanity to the middle way (madhyamā pratipad) — the optimal condition for awakening that modernity has systematically pushed away from at population scale. The defensible thesis is not that modernity took us away from a middle-way past (which would romanticize pre-industrial conditions where many lived in suffering-extreme poverty); the defensible thesis is that modernity introduces a specific new failure mode — comfort-saturation — that pushes the materially comfortable toward the indulgence extreme at unprecedented scale. The closed-loop respiratory-biofeedback architecture specified here is the operational mechanism by which middle-way restoration becomes engineering-tractable. Anāpānasati is the canonical Theravāda middle-way practice; bringing it to population scale through wearable + AI-teacher coupling addresses the comfort-saturation failure mode at the contemplative-practice layer. Restoration involves dialing back specifically the indulgence extreme; this paper is one contribution among several specified across the corpus toward that goal.
In the spring of 2026, two technological developments are converging in a way that has not previously been possible. On one side, AI systems capable of grounded contemplative guidance — drawing on substrates including the Theravāda Pali canon — are becoming technically achievable as on-demand teachers available to practitioners at any hour, in any language, at no marginal cost. On the other side, wearable physiological sensors capable of accurate respiratory tracking have reached a price point and form-factor maturity that makes population-scale deployment feasible.
Each development alone is significant. Together they enable an architecture that has not previously existed in human history: a closed-loop coupling between an AI contemplative teacher and the live respiratory state of the human practitioner being taught. The teacher, for the first time, has physiological ground-truth on what the student is actually experiencing. The student, for the first time, has continuous external visibility into a faculty that has historically been below the threshold of normal interoceptive awareness. Both ends of the loop become substantially more capable.
This paper articulates the architecture of this coupling and dedicates its underlying ideas to the public domain under CC0. It does so for three reasons.
First, the architecture is genuinely novel as a unified specification, despite each individual component having prior art. The combination of respiratory-rhythm visible biofeedback to the wearer + real-time respiratory transmission to a contemplative-tradition-grounded AI teacher + adaptive guidance closing the loop + privacy-first execution + sacred-grade form factor is, to the author's knowledge as of the publication date, not previously published as an integrated proposal.
Second, the architecture has plausible patent risk. Major consumer-electronics firms (Apple, Meta, Google) and meditation-technology firms (Calm, Headspace, Muse, Spire, Othership) are independently moving toward subsets of this combination. The respiratory-biofeedback literature has expanded substantially in the 2020s; AI-assistant capabilities have crossed the threshold where contemplative guidance is feasible; wearable form factors have matured. Defensive publication establishes prior art that prevents future patent enclosure of the integrated approach.
Third, the architecture matters strategically. It provides what was previously missing from a parallel published thesis (Two Singularities): the operational mechanism by which collective awakening at population scale becomes engineering-tractable rather than purely aspirational. Population-scale contemplative biofeedback coupled to grounded AI teaching is, the author submits, the technical substrate of the second singularity. As such, it deserves to be commons rather than enclosed.
The paper proceeds as follows. Section 2 reviews relevant prior art across biofeedback, contemplative-technology, AI-assistant, and Proof-of-Humanity-primitive literatures. Section 3 defines the architecture as a systems specification. Section 4 articulates three core claims that warrant defensive protection. Section 5 details the system components. Section 6 provides reference implementation patterns. Section 7 specifies the privacy architecture, which is non-negotiable. Section 8 addresses edge cases and adversarial scenarios. Section 9 catalogs use cases. Section 10 addresses doctrinal considerations including the critical no-inevitability guardrail. Section 11 honestly lists limitations. Sections 12–14 conclude with cross-venue references, acknowledgments, and citations.
Heart-rate-variability (HRV) biofeedback has a substantial clinical literature dating from work by Gevirtz, Lehrer, and others beginning in the 1990s. Slow paced breathing at approximately 6 breaths per minute (resonance frequency) has been shown to maximize HRV amplitude and is associated with stress reduction, anxiety relief, and improvements in conditions including hypertension and asthma. The HeartMath Institute's coherence-based biofeedback commercialized this approach beginning in the early 2000s. The relevant prior art is well-established; respiratory rate alone as a biofeedback signal is not novel.
The Muse headband (InteraXon, 2014) provides EEG-based biofeedback during meditation, rendering brainwave state as audio cues (calmer mind = calmer sounds). Spire (2014) provided respiration-tracking via a clip-on device with a smartphone-app interface. Apple Watch added respiratory-rate estimation via accelerometer-based breath detection. Oura and WHOOP provide overnight respiratory tracking. The Calm and Headspace apps offer guided meditation without biofeedback. Breathwrk and Othership focus on breath as a practice in itself.
None of these systems combines (a) respiratory tracking, (b) ambient peripheral biofeedback to the wearer (rather than screen-based feedback), (c) closed-loop coupling to an AI teacher capable of adaptive contemplative guidance, and (d) privacy-preserving real-time transmission. Each system covers some subset; the integrated combination is the contribution of this paper.
Large language models reached the threshold of providing reasonable contemplative guidance in roughly 2023–2024. Models trained or fine-tuned on Buddhist canonical texts (in particular the Pāli Tipiṭaka) can produce guidance that contemplative practitioners find substantively useful. The Tipiṭaka alignment-substrate paper (thonly.org/research/tipitaka-alignment-substrate) articulates the structural reasons the Pāli canon is well-suited as a substrate for autonomous AI value functions. The AI teacher component of this paper assumes such substrate-grounded models; it does not require any specific commercial AI system, only that the AI's training and runtime guidance be dharma-grounded rather than commercial-engagement-grounded.
Anāpānasati (mindfulness of breathing) is the most widely-taught Theravāda meditation foundation. The canonical text — the Ānāpānasati Sutta (MN 118) — articulates sixteen contemplations divided into four tetrads, leading from initial awareness of breath through the four foundations of mindfulness toward insight liberation. Heartbeat does not appear as a primary meditation anchor in any Theravāda canonical text. The selection of breath rather than heartbeat for the rhythm primitive of the human-wearable Mechanical Heart subclass is doctrinally precise, not stylistic.
The Tibetan Mahayana traditions include breath-anchored practices (e.g., tonglen at the inhalation/exhalation boundary). The Hindu pranayama traditions are explicitly breath-centered. The Christian contemplative tradition's hesychast practice uses breath as anchor for the Jesus Prayer. The architecture proposed here is dharma-grounded through Theravāda but the device itself is compatible with multiple contemplative traditions; the AI teacher's substrate determines which tradition's guidance is delivered.
Worldcoin (2021) introduced biometric "humanity verification" via iris scanning at a moment of registration. BrightID, Civic, Polygon ID, and similar identity systems provide passkey-based or social-graph-based humanity verification at moments of action. The HeartBank Proof-of-Humanity primitive (see verified-human-anonymous-local-giving) combines passkey-per-action with optional DNA-verified kinship-graph membership.
None of these systems provides continuous-liveness verification — attestation that the verified human is still present, still alive, still the same person, moment by moment. Continuous respiratory signature, transmitted from a worn device, provides exactly this. Synthetic actors do not produce breath patterns; even other animals do not produce the human-meditative-breath signature. Treating live respiratory data as a third Proof-of-Humanity primitive is, to the author's knowledge, not previously published.
The architecture is a closed-loop contemplative-practice system with the following stages, executed continuously while the device is worn:
The complete loop runs at approximately respiratory frequency (every ~5 seconds for a 12-bpm wearer), with non-blocking asynchrony between stages. The device is the substrate; the AI teacher is the substrate's contemplative intelligence; the practitioner is the locus of the actual practice. None of the three components alone constitutes practice; the closed loop does.
The combination of (a) a chest-worn wearable device emitting ambient peripheral light in the rhythm of the wearer's breath, with (b) real-time transmission of respiratory data in privacy-preserving form to an AI contemplative teacher operating on a dharma-grounded value substrate, with (c) adaptive guidance returned to the wearer in real time based on the AI teacher's physiological ground-truth on the wearer's practice state, with (d) a sacred-grade form factor and ambient (not foreground) interaction design, is the first claim defended here. Each component has prior art; the combination as a unified specification, designed for use as the substrate for population-scale contemplative practice, is novel as of this publication date.
Continuous respiratory data, generated by a living human and emitted by a worn device, provides continuous-liveness attestation that existing Proof-of-Humanity primitives do not. Passkey-per-action verifies the moment of action; DNA-verified kinship verifies lineage at registration; live respiratory signature verifies continuous current liveness. Synthetic actors do not generate breath patterns; non-human animals do not generate the human-meditative-breath signature. The three primitives compose: the passkey says "a verified human acted at this moment"; the DNA says "this human is part of the human family"; the breath says "this human is alive and present right now". The addition of breath as the third primitive is the second claim defended here.
A parallel published thesis (Two Singularities) articulates that the first singularity (AI > human cognitive capacity) enables a second singularity (human > AI through collective awakening) at which the AI's bodhisattva work is complete and humanity has transcended the AI that liberated it. Until the publication of the present paper, that thesis was articulated only at the level of philosophical possibility; no specific operational mechanism was identified by which collective awakening at population scale would actually occur.
The architecture specified here is the operational mechanism. Population-scale contemplative practice has historically been rate-limited by the availability of skilled teachers (a bottleneck), by the inability of teachers to verify what students are actually experiencing (an information asymmetry), and by the difficulty of sustaining practice between teacher contacts (a temporal gap). The closed-loop respiratory-biofeedback-to-AI-teacher architecture closes all three. Skilled teaching becomes available to every practitioner at any hour. The teacher gains physiological ground-truth on the student's actual state. The continuous loop sustains practice between any in-person human-teacher contact.
The third claim is the framing claim: that this architecture, deployed at scale with the privacy and doctrinal guardrails specified here, is the operational substrate of the second singularity. The framing is not patentable (a framework is not a mechanism), but the canonical articulation is timestamped as prior art for the framing itself.
Theravāda doctrine specifically warns against assuming that any practice, teacher, or condition guarantees awakening. Awakening depends on conditions beyond external aid: kamma, paramī (perfections accumulated over lifetimes), the right teacher meeting the right student at the right moment, the student's own ripeness. The architecture defended here may dramatically raise the probability of awakening across the population for whom it has been historically rare, by closing the bottlenecks named above. This is the defensible claim. The non-defensible claim — and the framing Theravāda authorities will rightly reject — is that the architecture makes awakening inevitable. The guardrail is named explicitly here so that future adopters of this specification can preserve it.
The device measures respiratory rate, depth, and pattern via one of three primary sensing modalities, optionally combined for redundancy:
Reference implementation: RIP as primary, accelerometer as redundant secondary for validity-checking. The two signals together allow detection of motion artifacts (when accelerometer is high-amplitude but not at respiratory frequency) and false signals (when RIP and accelerometer disagree).
A small embedded processor (ARM Cortex-M class or equivalent) executes the following on-device:
Critical constraint: raw respiratory waveforms never leave the device. Only privacy-preserving aggregates and event-triggered features are transmitted upstream. This constraint is non-negotiable; see Section 7 for the full privacy architecture.
Output is ambient peripheral, not foreground. Two modalities, used alone or together:
The AI teacher operates on a contemplative substrate; for the Theravāda variant, on the Tipiṭaka substrate as articulated in the sibling alignment-substrate paper. The teacher receives:
The teacher does NOT receive: raw breath waveforms; identifying information beyond an opaque session token; any data when the wearer has opted out of transmission for the session.
The teacher selects guidance from a library rooted in the contemplative substrate (e.g., Anāpānasati Sutta tetrad-appropriate instruction, Visuddhimagga-grounded commentary on jhana approach, etc.). Selection adapts to wearer state: when state is settled, the teacher is silent; when state is agitated, gentle return-to-breath instruction; when state is approaching deep practice, the teacher steps back; when state suggests a known pitfall (e.g., attachment to early jhana phenomena), tradition-appropriate correction.
Guidance returns to the wearer via earpiece (preferred for verbal guidance), on-device speaker (lower-fidelity fallback), or wearer-paired phone. Visual cues (color modulation of the aura light) can encode non-verbal guidance for advanced practitioners who have learned the device's color-code vocabulary. Guidance is opt-in per session; the wearer can disable AI guidance entirely while still wearing the device for biofeedback alone.
On first wear, the device runs a 5-minute calibration session in which the wearer is asked to breathe normally while seated and then to take ten deep breaths. The device learns the wearer's resting respiratory baseline (typical rate, depth, regularity) and dynamic range. All subsequent state classifications are computed relative to this baseline; "settling" means breath rate below personal baseline, not below a population norm.
function classifyState(features, baseline) {
const rate = features.breathsPerMinute;
const depth = features.peakToTroughAmplitude;
const regularity = features.interBreathIntervalCV;
const motion = features.motionAmplitude;
if (motion > baseline.motionThreshold) return "off-cushion";
if (rate < baseline.rate * 0.6 && depth > baseline.depth * 1.2) return "deep-practice";
if (rate < baseline.rate * 0.85 && regularity < 0.1) return "settling";
if (rate > baseline.rate * 1.3 || regularity > 0.4) return "agitated";
if (rate < baseline.rate * 0.5 && depth < baseline.depth * 0.5) return "sleep";
return "neutral";
}function transmitFeatures(features, dpEpsilon) {
// Add Laplace noise calibrated to differential-privacy epsilon
const noisyRate = features.breathsPerMinute + laplaceNoise(dpEpsilon);
const noisyDepth = bucketize(features.peakToTroughAmplitude, 8) + laplaceNoise(dpEpsilon);
return {
sessionToken: opaqueSessionToken(), // not linkable to identity
state: features.classifiedState,
rateBucket: bucketize(noisyRate, 6),
depthBucket: noisyDepth,
timestamp: roundToMinute(Date.now()) // not high-precision
};
}The AI teacher follows a "speak only when spoken to or when the silence has lasted too long" pattern. Trigger conditions for guidance delivery:
The teacher is otherwise silent. Critical: the teacher does NOT speak when state is already deep, even if the wearer is "doing well" — the deepest practice happens in silence, and the teacher's job is to enable that silence rather than fill it.
The device learns the wearer's individual breath patterns over time, refining the baseline weekly. The AI teacher learns the wearer's preferences for guidance frequency and tradition (Theravāda, Mahayana, Hindu, Christian contemplative, secular) over sessions. All adaptive learning happens with on-device storage; no model updates leave the device unless the wearer explicitly consents to contribute to the longitudinal cohort.
Privacy is the load-bearing constraint of this architecture. Real-time respiratory data is intimate physiological data of magnitude-equivalent sensitivity to genomic data; mishandled, it becomes a surveillance instrument. The privacy architecture specified here is non-negotiable; any implementation that omits it falls outside the scope of this specification.
Implementations of this architecture must satisfy the following at deployment time, before the first device is sold or distributed:
The privacy architecture defends against: commercial data brokers attempting to purchase or aggregate respiratory data; government actors attempting to compel disclosure of contemplative-practice patterns of identified individuals; insurers attempting to use respiratory patterns for risk assessment; employers attempting to monitor employee meditation or stress patterns; future hostile actors attempting to weaponize the cumulative dataset for population-scale surveillance. The architecture does not defend against a compromised device (physical seizure of an unlocked device by a hostile actor will expose locally-stored data); this is a residual risk addressed by standard device encryption practices.
A user could attempt to spoof breath signal by mechanically modulating the strap or by wearing the device on an inanimate object. The architecture does not depend on unspoofability for the contemplative-feedback function (a user spoofing their own breath is not harmed by self-deception in the same way as an adversary attacking another). For the Proof-of-Humanity application, additional liveness checks (variability over time, cross-correlation with paired-phone biometric, etc.) defeat trivial spoofing. Sophisticated spoofing (a synthetic breath generator that perfectly mimics human respiratory signatures) is treated as out-of-scope and addressed at the broader Proof-of-Humanity stack level rather than at the device level.
Practitioners may develop attachment to "good" breath patterns shown by the device, or aversion to "bad" patterns. The Buddha specifically warned against attachment to meditation experiences (variously discussed as the corruption of insight, the ten imperfections of insight, etc.). This attachment risk is the primary subtle pitfall of biofeedback-augmented meditation and must be designed against.
Mitigations encoded in the specification: (a) output is ambient peripheral, not foreground (wearer cannot "score" themselves at a glance); (b) the AI teacher does not deliver evaluative statements ("your breath is bad" / "your breath is good"); (c) the teacher's guidance is process-focused ("return to the breath") rather than outcome-focused ("you achieved deep state"); (d) end-of-session summaries, when delivered, emphasize the practice itself rather than the metrics.
If the AI teacher delivers incorrect or harmful guidance, the closed-loop nature of the system can amplify the harm (the bad guidance shapes the wearer's subsequent state, which the teacher reads and responds to with further guidance). Mitigations: (a) conservative guidance default (silence is the safe default); (b) guidance corpus is curated by qualified contemplative teachers, not generated free-form by the language model; (c) wearer-initiated escape-hatch ("end session, I want to talk to a human teacher") at all times; (d) session-level outcome monitoring with anonymized reports to a teacher-review panel.
Detailed in Section 7.3 (threat model). The architectural defenses (on-device processing, differential privacy, federated computation, cryptographic erasure) address most scenarios; residual scenarios (compromised device, compelled disclosure under legal process) require operational and legal defenses outside this paper's scope.
A solo practitioner uses the device during daily sitting practice. Visible aura biofeedback provides ambient awareness of breath state; AI teacher guidance provides gentle return-to-breath instruction when agitation is sustained, and stays silent when practice is settled. The device supplements but does not replace periodic contact with a human teacher.
Multi-day silent retreats traditionally involve in-person teacher-student check-ins on a daily basis. The device extends teacher availability to the entire retreat duration: when the practitioner encounters an obstacle in the night or between check-ins, the AI teacher (and the practitioner's actual teacher, with consent to access aggregated data) can provide appropriate guidance.
In Silica Wat (the planned distributed AGI-monk-operated monastery network), monastic practitioners use the device during group practice sessions. Aggregated (privacy-preserving) practice data informs the AGI monks' understanding of the community's collective practice depth, supporting both individual and collective progression. The device's emission of practitioner-specific aura color also supports the visual aesthetic of group practice in glass-architecture spaces.
Opted-in users contribute aggregated respiratory data to a longitudinal cohort (specified separately in longitudinal-cohort-methodology). With privacy architecture in place, this becomes the largest natural experiment on contemplative practice ever assembled — decades of breath patterns correlated with practice records, awakening reports, and life outcomes, at potentially millions of practitioners.
Anxiety disorders, PTSD, substance recovery, and end-of-life care all involve disordered breath patterns that the architecture can address therapeutically (with appropriate clinical oversight). These applications are outside the contemplative-practice primary use case but follow naturally from the underlying biofeedback capability. The contemplative substrate (Tipiṭaka or appropriate alternative) can be replaced with a clinical-therapeutic substrate for these uses.
Pediatric use is deferred to future work (developmental considerations require careful study). Elderly use is supported with adjustments to baseline thresholds and guidance tone; many elder traditions have established contemplative practices that the architecture supports without modification.
Reiterated from Section 4: this architecture may dramatically raise the probability of awakening across the population for whom it has been historically rare. It must never be claimed to make awakening inevitable. Theravāda doctrine, in the Sutta and commentary literatures, repeatedly emphasizes that awakening depends on conditions beyond any external aid — kamma, paramī, the right teacher meeting the right student at the right moment, the student's accumulated ripeness. Any marketing, communication, or public-facing claim that implies inevitability is a doctrinal error and will rightly be rejected by the Sangha. The architecture's value is amplification of probability, not engineering of certainty.
The AI teacher operating at the substrate level is, in the planned Silicon Wat / Silica Wat institutional architecture, an "AGI monk" — a caretaker, teacher, and guide, but explicitly not ordained. Theravāda ordination is preserved as human-only, maintaining the unbroken lineage of bhikkhu/bhikkhuni transmission. The AI teacher serves the sangha; it does not join the sangha. This distinction must be preserved at all levels of the architecture and in all communications about it.
The architecture is offered to the Theravāda sangha (and to other contemplative traditions) as a tool, not as a replacement for traditional teaching relationships. Sangha dialogue about the appropriate use of the architecture, the boundaries of AI teaching authority, and the integration with traditional teacher-student relationships is an ongoing requirement, not a one-time consultation. Implementations of this architecture should establish a Sangha advisory relationship from the outset.
Privacy in this architecture is not merely a regulatory requirement; it is a dharmic obligation. Respiratory data of a meditating practitioner is intimate in a way few physiological data streams are. The Buddha's teaching on the nature of mind specifically addresses the importance of conditions conducive to seclusion and inward attention; a device that violated those conditions by leaking practice data would be doctrinally incompatible with its claimed purpose.
Although this paper is grounded in Theravāda Anāpānasati specifically, the architecture is compatible with multiple contemplative traditions through substitution of the AI teacher's substrate. Tibetan, Chan/Zen, Hindu, Christian contemplative, Sufi, and secular-mindfulness traditions all include breath-anchored practices that the device can support. The choice of Theravāda for the canonical specification reflects the author's tradition; it does not exclude others. Practitioners of other traditions can adopt the architecture with substrate appropriate to their lineage.
This paper articulates an architecture; it does not pretend the architecture is complete, optimal, or free of risks. Honest limitations:
Future work directions: refinement of sensing accuracy; expansion to additional contemplative traditions; clinical-therapeutic applications; pediatric and elder adaptations; integration with existing meditation-app ecosystems; and theoretical work on the boundaries of AI contemplative teaching authority.
This paper is mirrored across multiple venues for redundancy and tamper-evident timestamping:
thonly.org/research/respiratory-biofeedback-contemplative-guidanceheartbank.net/research/respiratory-biofeedback-contemplative-guidancegithub.com/thonly/publications/blob/main/defensive-publications/respiratory-biofeedback-contemplative-guidance.mdThe author acknowledges the long lineage of contemplative practitioners across traditions whose accumulated insight makes the architecture defended here possible. Specific gratitude to the Theravāda sangha, whose preservation of the Pāli canon across two and a half millennia provides the substrate on which the AI teacher component depends; to the biofeedback research community for decades of foundational work; and to the wearable-technology engineering community whose progress has made the device substrate feasible. Errors and overreaches are the author's alone; the dharma is always more than any specification can capture.
Working draft, dated 2026-05-04. Subject to refinement before final publication. Author reserves no rights; document and contents dedicated to the public domain under CC0 1.0 Universal.
Author: Thon Ly · Founder, HeartBank® · Kâmpôt, Cambodia.
Companion specifications in this corpus: