Sensor inputs
camera / depth / audio / force / telemetry
HaintEnginePatent-pending observability engine
Human-tool interaction is not fully observable.
Haint Engine is a tool-conditioned kinematic inference system for estimating hidden body state when a rigid-body tool obscures the movement that matters.
It is designed for review contexts where the visible signal is incomplete: instrument performance, simulation training, implement mechanics, and equipment-centered ergonomic analysis.
Observe
Collect what is visible.
Infer
Estimate what is hidden.
Understand
Reveal what is at risk.
REF_IDHE-OBS-01
CLASSOBSERVABILITY ENGINE
TYPEBIOMECHANICAL INFERENCE
TECHNICAL PLATE 01
FIG. 1
Generalized observability architecture.
Sensor streams, pose records, tool state, occlusion mapping, inference, and output modules are treated as one embeddable system rather than separate coaching features.
InputStateInferenceOutput
Pose records
visible landmark state + confidence
Tool-state model
identity / geometry / pose / contact zone
Occlusion map
visible / hidden / unreliable landmarks
Inference engine
anatomy + tool + prior trajectory
Output layer
score / cue / replay annotation
TECHNICAL PLATE 02
FIG. 2
Rigid-body tool occlusion region.
The tool creates a volumetric shadow where landmarks become hidden, unstable, or low-confidence. Haint Engine treats those states as inferable, not absent.
Anchored landmarks
Observed points remain tied to the skeletal chain and become constraints for hidden-state estimation.
Confidence decay
Low-confidence points are not discarded; they are marked as unstable evidence inside the frame buffer.
Inferable absence
The tool shadow assigns likely cause, contact region, and candidate landmark positions.
Rigid-body reference
Tool identity, axis, pose, and contact zones narrow the hidden-state search space.
TECHNICAL PLATE 03
FIG. 3
Anatomical constraint logic.
Candidate hidden-joint positions are reduced by body-segment lengths, joint limits, prior trajectory, and tool-contact plausibility.
Anatomical
segment length / joint range / skeletal continuity
Tool
axis / contact zone / leverage / rigid-body pose
Temporal
prior trajectory / velocity envelope / frame history
Event
impact / audio transient / force spike / task marker
Candidate-state reducerHidden landmark estimate
| State | Score | Status | Constraint result |
|---|---|---|---|
| A | 0.81 | accepted | consistent with tool axis and prior frame |
| B | 0.46 | review | joint range plausible; contact zone weak |
| C | 0.18 | reject | violates segment continuity |
The accepted estimate is not a guess from a missing landmark. It is the remaining candidate after anatomical, tool, temporal, and event constraints remove less plausible states.
TECHNICAL PLATE 04
FIG. 4
Cross-modal look-back window.
A detected event at t=0 can trigger review of preceding biometric frames to identify a likely pre-event movement condition.
baseline
visible chain stable
drift
wrist angle begins closing
risk
tool axis diverges from prior path
event
transient marker detected
impact / audio transient / force spike
pre-event biometric sequence
visible + inferred state replay
likely pre-event condition
APPLICATION INDEX
FIG. 5
Embodiments from the same inference core.
The architecture is not tied to one profession, instrument, or training environment. It is designed to support domains where a body acts through a tool: observe visible state, constrain hidden state, and return a reviewable event record.
Musical embodiment
01Instrument Performance
Instrument-relative movement economy, skeletal tension, finger travel, wrist alignment, and pre-event misalignment before the audible result.
Simulation review
02Medical Training
Tool-path review linked to inferred wrist, elbow, shoulder, and torso state during simulation, training, or post-session analysis.
Implement mechanics
03Athletics
Hidden joint state during bat, club, racket, stick, paddle, barbell, or other implement interaction.
Equipment review
04Industrial Ergonomics Review
Tool-relative posture, force direction, and ergonomic review signals while operating equipment or training around a fixed tool frame.
- Tool frame
- guitar neck / key plane / bow path / stick path
- Signal source
- movement trace synchronized with audio event
- Review output
- pre-event movement condition and inferred hidden state
FIG. 6 / Musical Embodiment
Haint Blue is the music-facing application of Haint Engine.
The musical embodiment is where the general architecture takes on a public-facing form. Haint Blue applies the same tool-conditioned inference model to guitar necks, keyboard planes, bow paths, drumsticks, and other instrument geometries.
Audio events can be synchronized with inferred body landmarks to review likely finger travel, wrist position, shoulder tension, pre-event misalignment, or movement economy before an audible result. Haint Blue carries the musician-facing language; Haint Engine remains the patent-pending technical architecture underneath it.
DOCUMENT CONTROL
Release Layer
Formal briefs stay controlled until the review context is known.
Public pages introduce the architecture. White sheets, technical notes, and embodiment-specific materials should be shared through a request path so claim scope, audience, and patent posture remain controlled.
Technical Overview
System framing, observability problem, and core inference model.
Available by requestArchitecture Notes
Sensor state, tool-state frame, occlusion map, and output-layer structure.
Available by requestEmbodiment Brief
How a domain-specific application inherits the generalized Engine architecture.
Available by requestNeed a technical packet?
Use the inquiry path for technical review, licensing, research, partnership, or embodiment discussions.
Open Inquiry Docket