A prototype is a question with wires.
Each build exists to retire one uncertainty. Tests are defined before results, raw evidence is retained responsibly, and failed gates stay failed until a new build changes the conditions.
Program gates
| Gate | Question | Minimum evidence | Decision |
|---|---|---|---|
| G0 | Is lateral contact speech usable? | Multi-user, multi-day wired recordings; comfort; wearer/other rejection | Continue, change transducer/placement, or stop |
| G1 | Does passive modulation faithfully carry a calibrated waveform? | Shielded fixture sweeps, linearity, drift, tolerances, spurs | Select topology |
| G2 | Can it survive body loading and orientation? | Phantom/enclosure coupon matrix; repeatability; range under approved conditions | Choose antenna/resonator architecture |
| G3 | Is the wearable safe and useful on people? | Reviewed protocol, supervised cohort, stop events, audio and fit metrics | Integrate or redesign |
| G4 | Can a consumer reader replace research instruments? | Integrated RF/audio/compute prototype, emissions pre-scan, thermal and security tests | Design verification build |
| G5 | Can it be sold responsibly? | Certification, FTO, manufacturing validation, privacy ops, support readiness | Limited pilot |
Design verification plan
The canonical DVP.md defines test IDs across mechanical, acoustic, RF, DSP, security, privacy, and product workflows. The CSV template records build, operator, environment, instrument, criterion, result, raw-data hash, and deviation.
Evidence quality ladder
- Observation: useful for generating hypotheses; cannot support a claim.
- Exploratory test: controlled enough to choose ranges and metrics; thresholds may still move.
- Confirmatory bench test: frozen method and criterion on new builds/samples.
- Representative validation: production-intent hardware/software, representative users/environments, blinded analysis where possible.
- Post-market evidence: monitored reliability, complaints, incidents, drift, and supplier changes.
Sample identity
Use IDs like TAG-P1-003-R2, POD-M2-014, and RDR-P4-002. Never reuse an ID after rework; append a revision and record the change. Every audio/RF file name begins with session ID and contains no person’s name.
Acceptance discipline
- Freeze a criterion before the confirmatory set.
- Report every tested unit, not only the best trace.
- Include uncertainty and calibration status.
- Separate algorithm development data from final validation.
- Measure worst-case mute, orientation, body, and environment.
- Record deviations contemporaneously.
- Link every external marketing claim to a claim-ledger row.
Kill criteria
Pause or pivot if the useful throat signal requires uncomfortable contact; if the physical mute cannot make wearer recovery impractical across approved conditions; if the compliant frequency plan destroys useful range; if a consumer-cost reader cannot maintain dynamic range; if contact mechanics are too anatomy-specific for a manageable fit system; or if FTO blocks the differentiated architecture without a viable license.
Gate review packet
One page of conclusions, build/configuration list, frozen protocol, raw-data manifest with hashes, calibration records, analysis version, complete results including failures, safety/privacy deviations, open anomalies, and a signed continue/pivot/stop decision.