Limen Step 6: Why the Decision Is Still Open
Concept study · Research only · No human testing · Not available · No development program exists
Neither route has earned selection. The inspected record contains useful component demonstrations, but no eligible study compares the complete invasive and optimized removable routes under the same population, endpoint, duration rule, and burden accounting.
The question in plain language
Would an invasive route—direct natural peripheral-nerve recording for forward control plus afferent nerve stimulation for return sensation—give a user-valued advantage over an optimized removable route using surface muscle recording plus external vibrotactile or electrocutaneous feedback?
The comparison was deliberately narrow. Both sides had to serve the same frozen population and a vision-occluded contact-and-release task, use the same adequate-duration rule, and report enough denominators, missing periods, maintenance, recalibration, adverse events, interruption, and abandonment to compare burden honestly.
What was researched
Science integrated a corrected, source-by-source audit of 21 records. Eleven invasive and nine removable near matches were extracted without pooling separate demonstrations into a fictional complete system. Full text or a readable peer-reviewed manuscript was available for 18 records; two paper bodies remained abstract or metadata only, and one dataset record exposed a manifest while 19 data or code payload bodies remained inaccessible.
The near matches included short direct-nerve recording, acute vibrotactile feedback, blanking and calibration studies, an eight-completer home surface-EMG study, mixed-population external-feedback work, and forward-only high-density surface-EMG work. Each answers a component question. None meets the complete frozen comparison.
Candidate and comparator burdens were sought symmetrically. For a removable system, the audit treated abandonment or non-wear as the analogue of explant, refit as the analogue of revision, doffing or electrode shift as the analogue of interruption, skin breakdown as an adverse-event field, and per-donning recalibration as calibration burden. Missing or unreported fields remained unknown; they were not converted to zero.
What the evidence establishes
- Direct neural recording, muscle recording, nerve stimulation, and external sensory feedback have each been demonstrated in bounded settings.
- Some studies show operation outside a single laboratory session, but longer observation of one component is not evidence for the chronic complete route.
- Important engineering problems—signal acquisition, decoding, stimulation delivery, calibration, and user task performance—can be studied separately.
- Burden belongs on both sides of the comparison. Removability does not mean burden-free, and implantation does not by itself mean durable functional advantage.
The defensible conclusion is narrower than the ambition: relevant components exist, but their existence does not establish that either complete route works as specified or that one route is better for the user.
Evidence classes that must stay separate
| Evidence class | What it measures | What it cannot prove by itself |
|---|---|---|
| Direct natural ENG | Naturally occurring peripheral-nerve activity used as a forward signal. | Muscle control performance, evoked response, sensory return, or simultaneous duplex operation. |
| Muscle EMG | Electrical activity produced by muscle and used as an external forward signal. | Direct nerve recording or an implanted neural route. |
| Evoked CAP / CMAP | A compound neural or muscle response elicited by stimulation. | Natural voluntary ENG or useful perception. |
| Stimulation-only return | A stimulus was delivered; some studies also measure a reported or physiological effect. | A complete forward-and-return loop, retained forward information during stimulation, or functional advantage. |
| Sequential operation | Recording and stimulation occur at different times or in alternating phases. | Usable simultaneous duplex operation with controlled artifact. |
| Simultaneous duplex | Forward acquisition remains usable during relevant return stimulation. | Long-term stability, acceptable burden, or superiority over a removable comparator. |
Why no route can be selected
| Decision field | Invasive route | Optimized removable route |
|---|---|---|
| Eligible exact-route record | 0 of 11 near matches | 0 of 9 near matches |
| Complete-route outcome | Unknown | Unknown |
| Shared endpoint and duration | Not established | Not established |
| Symmetric burden accounting | Incomplete | Incomplete |
| Eligible head-to-head comparison | None located within the bounded search | |
Zero eligible comparisons does not mean that either route fails or that no relevant study exists anywhere. It means the bounded search did not locate evidence capable of answering the frozen decision. The scientifically honest result is therefore OPEN_BOUNDED, with PROPOSED_HOLD carried forward for independent review.
What Step 7 has—and has not—done
The accepted Science record and the accepted Legal/Governance record were merged by stable identifier and hash. Independent QA passed the exact merge, and the Integrator reproduced it after QA. The merge preserves all 14 Science imports and all 44 governance questions without converting partial evidence into closure.
That is custody work, not adjudication. No panel has been convened. No reviewer, attendance record, conflict disclosure, recusal, vote, dissent, or quorum evidence exists. The present label HOLD_MISSING_INPUT is administrative: the six-role review cannot occur under current authority because qualified reviewers may not yet be named or contacted.
The Science proposal therefore remains exactly that—a proposal. Step 7 has neither accepted nor rejected it.
What evidence would change the decision
An eligible comparison must place both complete routes in the same decision frame. It must use the frozen population, exact forward and return routes, the same vision-occluded contact-and-release endpoint, and the same prospectively fixed adequate-duration rule.
It must also report all enrolled, exposed, analyzed, withdrawn, non-wearing, interrupted, revised, explanted, refitted, and continuing participants; unavailable periods; category identity; adverse events; calibration and recalibration; maintenance; and route-specific burden. A result limited to continuing users is a denominator limitation, not evidence that abandonment did not occur.
For the invasive route, useful forward neural information must remain demonstrable during relevant return stimulation; a command, a delivered stimulus, and a measured sensory or functional effect must remain separate observations. For the removable route, the implementation cannot be an unspecified marketing comparator. Both sides need equal evidentiary scrutiny.
What happens next
- Preserve the accepted input snapshot. Any changed byte creates a new candidate and requires fresh independent QA.
- Obtain explicit authority for the real Step 7 review. The required independent chair and six-role core panel must be selected through a process that records qualifications, independence, conflicts, recusals, attendance, and quorum.
- Review every open question without collapsing evidence classes. The panel must adjudicate the 14 Science imports and 44 governance questions separately, preserve dissent, and record any missing mandatory input.
- Keep the result provisional until new QA and Integrator custody. A panel record is not accepted merely because a meeting occurred.
- Stop before design if the evidence remains open. No blueprint, procurement, fabrication, biological work, animal work, human work, or development program follows from the present record.
Verification and custody
Science: 16 files, 228,987 bytes, registry e0772c0caf7ae5631b82de5246f706d019cdce5b787d5973523cf82dbc141e4e. Legal/Governance: 12 files, 104,234 bytes, registry 7988037eef47aaebeb17e33a16bf7325d4333031e8087f215bc8566c6986ee2a. Step 7 input merge: 9 files, 82,872 bytes, registry ed6c81ba0c1762e0acc106a40034f5d7aa28472ae3212a8fcb56664f62b12ca7. Acceptance evidence registry: 98a6c10dd34947b216231f9e53a2942cd24f3dd15c06935e22bbf8316d7fe294.
Independent QA returned P0/P1/P2/P3 = 0/0/0/0 for each exact accepted package, followed by Integrator reproduction of the unchanged bytes. A checksum proves identity, not scientific truth, legal clearance, or permission.
Local source paths: docs/science/limen/deep-review-03/README.md · docs/legal/limen/step-7-merge-01/README.md · docs/review/evidence/limen-step6-step7-inputs-2026-08-30/05-step-7-boundary.md