Limen: what the deeper evidence changes

The few terms needed

Limen

the project codename for researching a digital connection to peripheral nerves. It is not a selected implant or available product.

ENG

electroneurogram: electrical activity recorded from a nerve. Direct natural ENG must be separated from muscle activity, electrically evoked responses and stimulation artifact.

Afferent return

information delivered through a peripheral nerve toward the central nervous system. A stimulation command, delivered pulse, neural response, perceived sensation and useful effect are separate checkpoints.

Duplex

recording and return operating together under a declared timing mode. Same participant, same nerve or sequential demonstrations do not by themselves prove simultaneous duplex.

Reversibility

evidence about what remains after removal relative to a defined baseline. Explantation alone does not prove that the nerve, function, data or support burden returned to baseline.

The question Science actually tested

Does primary evidence establish a chronic, selective, stable and reversible peripheral-nerve interface that records direct natural ENG and delivers controlled afferent stimulation? If not, which dependency must close before any architecture can be considered for selection?

The review kept route identity, timing, duration, denominators, calibration, tissue response, failures, removal and comparator evidence separate. It did not combine unrelated studies into a fictional complete device.

What was researched

Science inspected human and animal studies covering longitudinal nerve recordings, neural and muscle routes, stimulation and percept mapping, artifact handling, tissue response, mechanical failure, infection, revision, explant and post-removal evidence. It also inspected official trial and program records and publication-specific patent material only for overlap—not for novelty or freedom-to-operate conclusions.

Legal mapped the questions that would later govern intended use, restoration versus enhancement, consent, withdrawal, lifetime support, neural-data privacy, command integrity, cybersecurity, accessibility, coercion, dual use, regulatory framing and intellectual-property review. All 44 governance questions remain OPEN.

What the evidence establishes

• Separate human components exist. Short studies demonstrate direct peripheral-nerve recording and neural stimulation in the same participants, but the strongest task example used mixed neural and muscle input.
• Late natural-neural recordings have been reported. One inspected record includes recording at day 502. That does not establish continuous availability, stable identity of the same neural sources, or long-term return in that participant.
• Specific sensory effects are real but person- and window-specific. Location, quality, intensity, naturalness and usefulness are different outcomes. Stable subsets cannot erase lost contacts, changing percepts or recalibration.
• Tissue and system failures are part of the record. The inspected studies report infections, wire or lead failures, delamination, revisions, explants, cellular responses and incomplete post-removal evidence. They do not support one pooled harm rate.
• The field is crowded. Electronic, living-intermediary and combined interface ideas already appear in programs, papers and patent publications. That defeats any casual broad-novelty story but does not decide patentability or freedom to operate.

One correction materially changed the record

The official correction to the implanted neuromusculoskeletal-interface report was read directly. It limits clinical outcomes to three of four patients and reports two serious adverse events in Patient 4, including sepsis, infection-related removal, antibiotic treatment and later local infection. It also states that Patient 4 received no sensory feedback.

Those facts strengthen denominator and burden requirements. They do not create a pooled rate, an extra set of independent explants, a direct-ENG result, or evidence that removal restored baseline. The historical uncorrected clinical wording remains quarantined.

Read the official correction

What remains unresolved

No architecture earned selection

• A — electronic: has the strongest direct component precedent, but chronic source identity, duplex behavior, tissue burden and matched advantage remain open.
• B — biohybrid or living intermediary: has biological and adjacent experimental precedents, but no accepted evidence establishes the required chronic human transfer behavior or unique value.
• C — combined: could theoretically divide functions across biological and electronic layers, but useful contribution, attribution, coupled drift and failure isolation remain unestablished.

These are research folders, not finalists. More evidence in one folder does not make it a selected product.

What the governance review adds

Even a stronger scientific record would not automatically authorize development. A future decision would still need a defined population and task, a defensible restoration or enhancement boundary, complete consent and withdrawal logic, long-term service and removal obligations, neural-data rights, command-integrity and incident rules, accessible alternatives, coercion safeguards, jurisdiction-specific regulatory review and qualified patent/FTO work.

The legal package asks those questions without pretending to answer them. Its US and EU family maps are analytical prompts only. No regime, class, pathway, standard, legal applicability or patent position has been selected.

The decision that follows

Keep Limen as a bounded research direction. Do not select a device or begin engineering from this record. The deepest current unknown is whether source-separated natural neural information stays specific and useful during controlled return across longitudinal change, with the full denominator, missing periods, contact losses, remapping, failures and later tissue/removal state visible.

This is an OPEN evidence question, not evidence of impossibility. It is also not permission to fill the gap through biological, animal or human work.

What happens next

Stop or redirect rules

• Stop the direct-ENG claim if eligible evidence identifies the measurement as muscle-derived, evoked-only or artifact-only.
• Do not call the system continuous duplex if usable recording excludes stimulation intervals or the functions are only sequential.
• Do not claim stable identity from a surviving subset while losses, gaps or remapping remain outside the denominator.
• Stop the invasive justification for a bounded target if an optimized removable comparator meets or exceeds the same user-valued outcome with lower burden.
• Do not call removal reversible without later evidence relative to a defined baseline.

Accepted records behind this brief

A SHA-256 digest identifies exact bytes. It does not prove a scientific claim. Producer completion, Independent QA, Integrator acceptance, Git landing, publication and release remain separate states.