# Norysr working hypothesis register

**Revision:** 02
**Project state:** research and specification; no completed prototype
**Initial intended population:** pediatric cerebral palsy with swing-phase foot drop
**Evidence status:** citation verification in progress

This file records provisional Norysr design hypotheses for benchtop planning. It is not a prescription, a safe-use envelope, a regulatory determination, or a report of measured performance.

The earlier public research paper was withdrawn after release review found citation identifiers that resolved to unrelated works. Until a source-by-source audit is complete, this register does not claim that its provisional values are established by external literature.

## System concept

`distal-shank IMU + tibialis-anterior/gastrocnemius sEMG → local preprocessing → gait-phase estimate → isolated surface-FES command → session record`

- Surface electrodes only for the initial concept.
- The stimulation decision is intended to remain on the wearable.
- Norysr has no completed prototype, participant dataset, measured outcome, or clinical authorization.
- A licensed clinician, electrical-safety specialist, ethics process, and regulatory specialist are required before participant-facing work.

## Status labels

| Label | Meaning |
|---|---|
| Working hypothesis | A proposed value or architecture choice for prototype planning. |
| Derived budget | Arithmetic assembled from working allocations; not a measurement. |
| Needs specialist review | A safety, clinical, ethics, or regulatory question outside the project team's authority. |
| Evidence audit pending | External support has not completed source-by-source verification. |

## Sensing hypotheses

| Item | Provisional value | Status |
|---|---|---|
| IMU placement | Anteromedial distal shank | Working hypothesis |
| IMU acquisition | 6-axis at **100 Hz**; compare 200 Hz | Working hypothesis |
| sEMG channels | **2:** tibialis anterior and gastrocnemius medialis | Working hypothesis; clinician placement review required |
| sEMG acquisition | **1 kHz, 16-bit**; filtering to be established during bench work | Working hypothesis |
| Feature window | Evaluate **150–250 ms** windows | Working hypothesis |

## Stimulation boundary

No frequency, pulse-width, current, electrode-placement, command-duration, or gait-window value in this document is a safe-use instruction. Those parameters are withheld while the evidence and safety basis are re-audited.

Proposed architecture-level requirements:

- Default-off stimulation after startup, reset, communication loss, watchdog expiry, or sensor fault.
- Hardware current and charge limits independent of application software.
- Electrical isolation appropriate to the final architecture.
- Clinician-defined limits before any participant-facing use.
- Benchtop fault injection before any human study is proposed.

Every item above needs specialist review.

## Inference hypotheses

| Question | Provisional criterion | Status |
|---|---|---|
| Can a compact model classify gait phase? | Compare with a transparent IMU-only rule baseline | Experiment to design |
| Does sEMG improve timing or robustness? | Report participant-level false and missed triggers | Experiment to design |
| Can inference remain local? | Profile a quantized model on candidate microcontrollers | No hardware selected |
| Can short calibration help? | Compare population and per-user models | No Norysr dataset |

## Latency ledger

| Stage | Provisional allocation | Status |
|---|---|---|
| Sensor read | ≤5 ms | Working allocation |
| Window update and buffering | ≤10 ms | Working allocation |
| Inference | ≤5 ms | Working allocation |
| Command transfer | ≤3 ms | Working allocation |
| **Sensor to stimulation command** | **≤23 ms** | **Unmeasured derived subtotal** |

The subtotal is arithmetic, not an oscilloscope trace. Norysr has measured no latency on project hardware.

## Testable questions

1. Does fused IMU+sEMG improve trigger timing or robustness over an IMU-only rule baseline?
2. Can a quantized on-device model stay within a measured sensor-to-command budget on candidate hardware?
3. Does per-user calibration reduce participant-level error relative to a population model?
4. Can the complete sensing and command path fail safely under dropped data, low confidence, reset, and processor fault?

## Validation order

1. Complete the citation and claim audit.
2. Select components and verify sensor synchronization, isolation, and default-off behavior.
3. Measure latency distributions, memory, power, and fault behavior on the bench.
4. Compare model output with transparent baselines on an appropriate dataset.
5. Obtain clinical, ethics, electrical-safety, and regulatory review before participant-facing work.

## Regulatory status

No Norysr regulatory classification, product code, submission route, predicate strategy, or jurisdictional class has been confirmed. Those are questions for qualified regulatory counsel after intended use and system architecture are defined.

## Non-goals for the initial concept

- Implantable stimulation.
- Cloud-controlled stimulation.
- Multi-joint assistance.
- Populations other than cerebral palsy.
- Unsupervised participant use.
- Diagnosis, prognosis, or autonomous clinical decisions.

## Source status

The public review paper is unavailable while its citations and dependent claims are audited. A corrected evidence document will be republished only after each primary identifier, attribution, population, method, and quantitative claim is verified against the source.