Limen — full roadmap

The historical Evidence Studio tracker was preserved as a tombstone. The active machine-readable tracker now names Novel Biotech, the real macOS host, the BPNI/Limen research scope, separate review and Git-landing axes, and fail-closed external-action boundaries.

Completion evidence: active schema-v2 tracker plus archived historical bytes. Limit: tracker acceptance is administration, not scientific progress.

The project established the broad field boundary: human studies have separately shown peripheral-nerve recording and stimulation components, but broad novelty and universality fail. Chronic direct natural ENG plus afferent return, same-interface simultaneous duplex, durable selectivity, long-term tissue compatibility, meaningful sensory encoding, reversibility, and advantage over lower-burden systems remain open or unknown.

Completion evidence: accepted research-control records. Limit: no target, architecture, or development program follows from component precedent.

The 94-file Product Engineering R3 package decomposes signals, states, interfaces, failures, traceability, possible R0/R1 verification, wiring contracts, and configuration controls. Its acceptance means the documentation is internally controlled and reviewable. It is not a real device blueprint: selected parts, official datasheets, accepted values, calculations, native CAD/EDA, and executed verification are absent.

Completion evidence: exact frozen Product package plus QA and Integrator evidence. Landing: currently the active reviewed Git batch.

Concept V1 explains the layered information path, common versus person-specific functions, maintenance ideas, service boundaries, and architecture-neutral interface concepts. It helps readers reason about the system without pretending geometry, electronics, biology, or implantation has been solved.

Completion evidence: accepted concept snapshot. Limit: concept drawings are explanatory, not fabrication or implantation instructions.

Science, Product, Backend, Business, and Legal/Regulatory/Ethics each supplied the semantic inputs needed to ask a target question consistently. Step 4 froze route identity, source/provenance separation, state priority, comparator logic, governance boundaries, and ten Backend decision classes while selecting none of their concrete implementations.

Completion evidence: accepted owner-lane packets and local Step 4 website explanation. Limit: input closure does not select the input values.

Five possible problem/comparator tuples were evaluated. The closest, S5R2-T01, concerns adults with acquired unilateral transradial limb loss performing vision-occluded contact/release object manipulation using direct natural peripheral ENG forward and afferent peripheral-nerve stimulation return, compared with optimized removable surface EMG plus optimized external feedback.

The complete candidate route and complete comparator route are both UNKNOWN. The exact matched search is OPEN. S5R2-T01 is deferred, not chosen; T02 is deferred; T03–T05 remain on hold.

Completion evidence: accepted Product, Science, Legal/Ethics, and founder-report snapshots. Decision: HOLD_BOTH_UNKNOWN.

Purpose: determine whether an invasive complete route offers a user-valued advantage over an optimized removable complete route. This is the hardest current scientific problem because existing studies mostly prove components in different populations, schedules, endpoints, or route combinations.

Required work:

1. Keep the complete-match search bounded to the frozen population, forward route, return route, endpoint, comparator, and adequate-duration rule.
2. Search candidate and comparator arms symmetrically; do not demand detailed failure accounting from one while accepting marketing-level success language from the other.
3. Extract population, denominator, route identity, function, endpoint, duration, missing periods, adverse events, revision, explant, interruption, category identity, maintenance, and recalibration burden.
4. Keep direct ENG, muscle EMG, evoked CAP/CMAP, stimulation-only return, sequential operation, and simultaneous duplex as different evidence classes.
5. Do not pool separate component studies into a fictional complete-route comparison.
6. Preserve a bounded negative search as OPEN; absence of a located source is not proof that the route fails.

Step 6 closes only when eligible evidence can populate both sides of the frozen decision table—or when the bounded search and independent review justify an explicit continued hold or stop.

The same frozen record must be reviewed independently by affected-community, clinical, disability-rights, bioethics, privacy, cybersecurity, regulatory, and jurisdiction-specific reviewers. Science cannot waive those gates, and founder preference cannot replace them.

Decision rule:

• Candidate MEETS and comparator BELOW: retain the target as a research hypothesis; this is not a superiority or development claim.
• Comparator MEETS_OR_EXCEEDS: invasive uniqueness FAILS for the frozen endpoint.
• Candidate DOES_NOT_MEET: stop the candidate.
• Either side UNKNOWN or evidence mismatched: hold the affected side; no selection.

Output required: one exact problem, population, forward/return function, endpoint, comparator, exclusions, schedule/duration basis, decision state, and accepted review chain.

Product converts the accepted target into bounded requirements without changing Science claim states. It freezes intended information routes, accepted source classes, functional states, common versus person-specific responsibilities, lifecycle expectations, exclusions, safe-state semantics, and the matched comparator obligation.

Completion condition: exact target tuple, Product translation, applicable Legal/Ethics review, fresh QA, Integrator acceptance, and explicit founder authority for the next bounded scope.

Backend proposes the concrete choices needed to implement a local deterministic simulator. Science separately defines and accepts synthetic reference vectors that test route identity, artifact, provenance, invalidity, calibration, command, acknowledgement, and fail-closed behavior without pretending the records are biological measurements.

Completion condition: all ten technology decisions selected and accepted; exact synthetic vectors accepted by Science; no dependency or implementation begins before authorization.

R0 is an isolated, offline, non-actuating software system. It processes only accepted synthetic records, preserves source and route identity, rejects ambiguous or invalid inputs, executes deterministically, denies network and device access, and produces append-only auditable outputs.

What it may prove: conformance to its frozen software contract. What it cannot prove: neural detection, stimulation delivery, biological validity, chronic stability, safety, efficacy, feasibility, selectivity, encoding, or comparator advantage.

Independent verification must cover golden vectors, negative vectors, deterministic replay, serialization identity, resource limits, path containment, device/network denial, audit-chain integrity, crash and cancellation behavior, state priority, stale calibration, provenance loss, aliasing, and command inhibition.

Completion condition: frozen implementation, executable test matrix, reproduced results, Independent QA, Integrator acceptance, and a decision to stop, correct, or consider R1.

The rejected R1_BENCH_V1 is not a shortcut. A new R1 question may be opened only if accepted R0 evidence shows a non-biological bench can answer a specific remaining engineering question and the founder explicitly authorizes that scope.

Kill condition: if simulation already answers the question, or the bench would imply unsupported biological behavior, do not create R1.

A legitimate R1 package would require selected non-biological functions, official component datasheets, calculations and uncertainty budgets, interface contracts, power and signal containment, native editable CAD/EDA, bill of materials, fixtures, passive loads or phantoms, test points, configuration control, and a verification plan. No anatomy, implant geometry, biological stimulation prescription, or surgical design belongs here.

Blueprint gate: only exact accepted inputs, native sources, calculations, and executable verification can justify the term “engineering blueprint” for R0 or a strictly non-biological R1.

Purchasing, assembly, energizing, and testing require separate authorization after the engineering package and its QA close. Results must remain limited to the passive non-biological bench. Failures, rework, substitutions, calibration, and uncertainty stay in the evidence record.

Limit: a passing electronics bench does not establish compatibility with a nerve, tissue, person, implant, or clinical use.

Families remain neutral: A conventional electronic, B biohybrid/living intermediary, and C combined biological/electronic. Every family must face the same accepted target, route identity, artifact, calibration, failure containment, monitoring, service, replacement, burden, reversibility, comparator, and falsification requirements. Missing evidence is never a favorable score.

Selection threshold: one measurable, source-separable contribution that survives comparison and has defensible containment, withdrawal, and failure attribution.

This step would require a qualified scientific team, institutional facilities, current regulatory and ethics advice, a bounded biological question, validated non-biological prerequisites, formal governance, funding, quality systems, and explicit authorization. The present roadmap contains no biological protocol, material prescription, cell/tissue method, animal plan, surgical procedure, or human-testing plan.

Stop rule: no biological work begins because a document, simulation, bench, patent, or founder preference appears promising.

A development program would require an accepted unmet need, evidence-backed target, comparator advantage hypothesis, selected architecture, qualified team, funding, IP/FTO work by qualified counsel, regulatory strategy, quality system, risk management, preclinical evidence, and a separate founder decision. Even then, program start would not mean the device is safe, effective, approved, available, or commercially viable.

Current state: no development program exists.