BPNI: the loop exists in pieces. The decisive comparison does not.

01 / Current decision

Decision now

Hold the architecture decision. Continue the evidence program.

The research establishes bounded human component evidence, not the comparison needed to select a route. Direct-natural-ENG simultaneous duplex, denominator-symmetric burden and user value, and complete-denominator chronic signal and percept performance remain open.

02 / Evidence that is real but narrow

What is demonstrated

Small human reports document implanted muscle-EMG control and electrically evoked peripheral sensory return in the same participant/system; the reports are heterogeneous and are not one pooled cohort. BPNI-R1-C001 ↔ CROSS RCLM-001

The inspected iSens publications describe an overlapping two-person early implanted cohort, with one-person functional/controller subsets and concurrent implanted EMG plus cuff stimulation; excluded artifact-dominated channels were not recovered natural ENG. BPNI-R1-C005 ↔ CROSS RCLM-005

Direct voluntary intraneural ENG has been decoded in two lower-limb participants, but recording and stimulation used the same interface in separate tasks and concurrent stimulation was not tested. Within the inspected packages, no human primary report demonstrated preserved or recovered voluntary natural ENG while peripheral stimulation was being delivered. That second statement is a bounded negative, not a universal absence claim. BPNI-R1-C002 ↔ CROSS RCLM-002 · BPNI-R1-C003 ↔ CROSS RCLM-003

Blanked, held, time-divided, or sequential-polarity operation contains unavailable or sequential samples and is not unqualified simultaneous duplex. BPNI-R1-C004 ↔ CROSS RCLM-004

03 / Keep signal and timing classes separate

Signal/timing evidence ladder

Natural ENG is electrical activity recorded directly from a peripheral nerve. It is not surface or implanted muscle EMG, and an evoked compound response is not voluntary natural ENG.

04 / Comparison discipline

Burden symmetry

The evidence measures different burdens on each side. It does not prove one side has less. The frozen comparator audit did not locate a direct longitudinal study using common burden definitions and complete enrollment denominators for invasive and optimized removable routes.

Missing reporting is not zero burden. Calendar time since implantation, device possession, and actual wear are different exposure clocks.

05 / Retention is a narrowing denominator

Long-term failure and survivorship

Human interface reports include multi-month or multi-year observations, but attrition, missing periods, revisions, infections, connector/electrode failures, explants, and selected-survivor analyses prevent a universal durability statement. BPNI-R1-C007 ↔ CROSS RCLM-007

1. ImplantedOriginal exposed-person and device denominator
2. RetainedStill implanted, reachable, and participating
3. Recently verifiedInterface or channel tested in the relevant period
4. Metric subsetContacts or users contributing the endpoint

A denominator audit, not a success funnel. Each transition needs an explicit disposition.

The longest cuff summary narrows from 14 implanted people to eight recently verified, four in the detailed threshold series, and two in selectivity analysis. The formal 2022 NEJM correction controls the corrected harms and outcome denominator for the 2020 neuromusculoskeletal-arm report: four implanted, clinical outcomes in three, and two serious adverse events. No pooled harm or durability rate follows. BPNI-R1-C007 ↔ CROSS RCLM-007 · BPNI-R1-C008 ↔ CROSS RCLM-008

06 / Registry state is not outcome evidence

Trials/publications

All 15 frozen official trial rows had no posted registry results module at the inspection date. Confirmed publications exist for several registered trials, but they are generally overlapping substudies or interim subsets rather than posted registry results or complete trial-wide reports. BPNI-R1-C009 ↔ CROSS RCLM-009 · BPNI-R1-C010 ↔ CROSS RCLM-010

This is a frozen-source linkage reconciliation, not a current-status adjudication. NCT06275282 contains protocol drift, not a posted result. NCT06620536 remains NOT_YET_RECRUITING in the frozen registry state despite an identifier-linked preprint describing 2024 consent and data. NCT02848846, NCT03178890, NCT04217005, and NCT05249049 retain stale or UNKNOWN states. Those conflicts remain unresolved: publication existence does not silently update registry status, and a stale or UNKNOWN registry state does not negate a paper. BPNI-FBR-QA-002 · RECON drift lines 9–14

Registry state can change after freeze. Publication records, package appearances, trial enrollment, and unique participant cohorts are not interchangeable counts.

07 / Technical adjacency, not legal conclusion

Prior-art boundary

The prior-art map contains 16 representative claim-inspected families, 15 trials, and 15 named systems. The patent-family and claim-adjacency rows identify published technical adjacency only; they do not establish implementation, current force, ownership, claim overlap, novelty, infringement, or freedom to operate. BPNI-R1-C012 ↔ CROSS RCLM-012

Named systems and first-party availability statements describe external research or commercial systems under their own source boundaries; they do not establish regulatory status, broad clinical availability, comparative performance, or availability from Novel Biotech. BPNI-R1-C015 ↔ CROSS RCLM-015

Claim adjacency is not FTO, novelty, ownership, current force, implementation, or human performance.

08 / Percept evidence

Sensory/percept stability

Human sensory encoding and percept maps are participant-, channel-, and session-specific; repeated calibration and drift handling are incompletely reported across studies. BPNI-R1-C013 ↔ CROSS RCLM-013

FINE home evidence includes two people in a short crossover and one person with 115 calendar days but 49 reported wear days, unequal channel windows, interruptions, and one failed channel. A 14-month USEA case lost working contacts; stability analysis stopped at day 298, and single contacts could not evoke percepts alone after day 354.

The NCT03260400 RPNI publications are overlapping program reports: the broadest inspected preprint reports five enrolled and four implanted/analyzed, while other papers report N=1-N4 subsets. Those publication denominators must not be added, and the preprint is not peer reviewed. BPNI-R1-C011 ↔ CROSS RCLM-011

State: human sensory signal present; chronic generalization unresolved. A biomimetic rule is not automatically a natural percept, and a task gain is not percept stability.

09 / What users may actually value

User-valued outcomes

User-valued outcomes span laboratory tasks, home use, qualitative experience, burden, embodiment, pain, and function, but instruments, exposure, comparators, and denominators are too heterogeneous for a pooled utility estimate. BPNI-R1-C006 ↔ CROSS RCLM-006

Agency, ownership, useful sensation, monitoring effort, confidence, pain, actual wear, non-wear, setup, calibration, repairs, comfort, maintenance, and clinic support are part of the outcome. Exact small-study gains exist, but most invasive reports analyze one to three selected people. No powered chronic comparison establishes patient-valued superiority over an optimized removable route.

10 / Evidence that would move the decision

What changes the decision

1. Close signal identity and timing: voluntary direct-natural-ENG recording during active peripheral stimulation, including saturation, blanking and recovery, residual artifact, unavailable-sample fraction, and online task performance.
2. Make burden symmetric: apply the same adverse-event, interruption, maintenance, calibration, non-wear, abandonment, revision, and explant ontology from the inception denominator.
3. Measure exposure: separate supervised training, testing, possession, availability, and actual home wear; publish missing periods and reasons.
4. Measure user value: prespecify validated agency, ownership, activity, pain, cognitive-load, confidence, satisfaction, wear, and burden endpoints with participant-level trajectories.
5. Reconcile people, not papers: maintain participant crosswalks across registries, reports, supplements, corrections, follow-ons, exits, and overlapping cohorts.

A null result at any gate is useful evidence. None of these requirements authorizes construction, recruitment, human testing, or development.

11 / Snapshot identity and limits

Methods/provenance

This public R1 reading edition synthesizes eight frozen producer evidence packages plus one frozen cross-package reconciliation package. All nine conventional manifests reproduced before content freeze. Producer validation and hashes prove snapshot identity, not independent scientific acceptance.

The reconciliation controls duplication and wording. It contains 41 matched identities, 12 high-risk overlap guards, and 15 report-safe claims. These are deduplication and claim-control records, not additional evidence, study, trial, cohort, source, or participant counts. The 26 repeated publication identities in the crosswalk are 26 works, not the sum of their package appearances; the 15 NCT identities are 15 registry records, not additional cohorts. BPNI-R1-C014 ↔ CROSS RCLM-014

The frozen R0 QA package reproduced at manifest 9cc5e82f1518d1f494e35ea9665ab682fcc05d002f4128c9fcf657b5d1a0f956. R1 corrected its one P1 and three P2 findings. Fresh independent QA then passed the exact corrected source-report bytes with P0/P1/P2/P3 = 0/0/0/0; the QA record manifest is 1685f6f3e5345ab18860478f50689e4c06099b160430e15aed1a9dbb0fa69bc9. This publication-facing derivative is not byte-identical to that source report and has no recorded Integrator acceptance. The source PASS does not select an architecture or create scientific, clinical, legal, product, development, or release authority.

Truth boundary

Human, animal, bench, registry, patent, and hypothesis evidence remain separate. This report does not pool heterogeneous cohorts, convert a bounded negative into absence, join sequential components into a fictional complete system, or treat source silence as zero burden.