Preface: The Shape of the Problem — Expanded
Introduction: The Five Pillars — Adding Repair
PART I: THE WOUND
Chapter 1: The Queue That Kills
Chapter 2: The Genome as a Book That Cannot Be Rewritten
Chapter 3: The Administrative Recidivism Loop
Chapter 4: The Disembodied Queue — Disability Beyond Paperwork
PART II: THE GHOST IN THE GENOME
Chapter 5: The Cryptographic Envelope
Chapter 6: Neuroprosthetic Data Sovereignty
PART III: SEE — Detection & Biomarkers
Chapter 7: HSPA v3.0 Homomorphic Scoring
PART IV: REMEMBER — The Immutable Ledger
PART V: REASON — Triage & MORTAR Engine
PART VI: REPAIR — The New Pillar (Neuroprosthetics)
Chapter 15: From Limb Replacement to Limb Reintegration
Chapter 16: Osseointegration & Antagonistic Muscle Interfaces
Chapter 17: Neuromorphic Hardware for Natural Sensation
Chapter 18: Closed-Loop BCIs & Bidirectional Neural Dialog
Chapter 19: Algorithmic Fairness in Prosthetic Access
PART VII: ENDURE — Institutional Architecture
COPERNICAN INTERLUDE
Why COBOL Still Matters: The Democratization of Machine Language
REFERENCES
There is a river that runs through the American administrative state. It was designed to carry people from injury to remedy, from diagnosis to support, from need to the means of living. For most of the twentieth century, it moved with a certain bureaucratic sluggishness—slow, but forward. In the twenty-first century, the river has backed up. It has become a pond, then a swamp, then a place where people go to wait and some of them do not come out.
But the river is only half the wound. When benefits are finally approved, when the queue spits someone out after years of waiting, the system considers its job done. It hands a check and walks away. For the person who lost a leg to diabetes, who lost an arm in an industrial accident, who lost motor function to spinal cord injury, the check is not the end of the injury. It is only the beginning of the rest of their life in a body that no longer works the way it used to.
This second edition expands the MERCY Protocol from four pillars to five. The original four—see (detect risk), remember (create immutable record), reason (fair triage), endure (outlast political cycles)—were necessary but incomplete. They could get people through the queue. They could not give them their hands back. They could not give them the ability to walk up stairs, to feel the texture of a child's hair, to play the piano again.
The fifth pillar is repair. Not the cosmetic repair of the "wooden hand" or the passive socket that fits over a stump, but the deep biological repair enabled by 2025–2026 breakthroughs: osseointegrated bionic knees anchored directly to bone (Shu et al., Science 2025), neuromorphic tactile sensors that distinguish textures at 98.38% accuracy (JHU, Science Advances 2025), antagonistic muscle interfaces that restore natural proprioception (MIT Media Lab), fully implanted brain-computer interfaces that restore both movement and sensation (Neuralink N1, 博睿康 NEO-ONE 2026), closed-loop deep brain stimulation that listens before it stimulates (Southern Medical University 2026). These are not science fiction. They are published, peer-reviewed, FDA-cleared, in human patients today.
Mercy, in the administrative sense, is seeing a person where the system sees a case number. But mercy in the full sense—mercy that does not stop at the check—is giving that person back the ability to participate in the world. To hold a cup. To walk without pain. To feel.
A community health clinic in Baltimore. A veteran who lost his right leg in Afghanistan walks in wearing ordinary pants. He does not limp. He shakes hands with his doctor using his right hand—the hand he lost in the same explosion, now restored with a tissue-integrated bionic prosthesis. He does not look at his hand when he shakes. He feels the pressure, adjusts his grip automatically. The check from the VA arrived in 2028, but the repair arrived in 2031. That is three years of his life he will never get back. MERCY exists to close that gap.
The first four pillars address the failure of the queue. The fifth pillar addresses the failure of the remedy. A system that approves benefits but provides no access to restorative technology is a system that pays people to be disabled, rather than investing in their ability to not be disabled—where technology makes that possible.
When the Social Security Administration approves a disability claim, it makes a determination: this person cannot engage in substantial gainful activity. It does not ask: can this person's ability to engage be restored? It treats disability as a permanent state rather than a potentially modifiable condition.
This is not a failure of malice. It is a failure of temporal horizon. The SSA adjudication system was designed in an era when amputation meant a wooden leg, when spinal cord injury meant permanent paralysis, when the line between "disabled" and "not disabled" was a binary that technology could not cross. That era is over.
As of 2026, the gap between what is technologically possible and what is administratively available is measured in decades. Osseointegrated bionic knees with direct neural control (Shu et al., Science 2025) allow above-knee amputees to climb stairs, walk at natural speed, and avoid obstacles—capabilities that would have reclassified them from "disabled" to "able-bodied" for many occupations. But these devices cost approximately $150,000 and are not covered by standard Medicare or private insurance for most indications. Fully implanted brain-computer interfaces (Neuralink, 博睿康 2026) have restored cursor control and fine grasping to quadriplegic patients, but they are available only in clinical trials to fewer than 50 patients worldwide as of this writing.
If genomic data is sensitive—and it is, as established in Chapter 2—neuroprosthetic data is an order of magnitude more intimate. A brain-computer interface does not merely record static biological information. It records your intentions before you act on them. It records the neural patterns that precede speech, before the words reach your mouth. In closed-loop systems that both read and write stimulation (the new generation of bidirectional BCIs and responsive neurostimulators), it also modulates your neural activity. It can change your mood, your memory, your perception of pain.
The post-quantum cryptographic envelope described in Chapter 5 is not merely desirable for neuroprosthetic data. It is a prerequisite for deployment. The "harvest now, decrypt later" threat model takes on new meaning when the data in question is your motor cortex output from 2026, decryptable by a quantum computer in 2040—decades after the implant, when the data is still biologically yours and still reveals everything about how your brain works.
The same five properties that make genomic data uniquely sensitive apply, intensified, to neural data:
| Property | Genome | Neural Implant Data |
|---|---|---|
| Immutable | Yes (cannot change your genome) | Partially (plasticity changes signals, but core motor/speech representations are stable for decades) |
| Uniquely identifying | Yes (50 SNPs suffice) | Yes (motor intent patterns are individually distinctive) |
| Predictive | Future disease risk | Imminent intent, emotion, cognitive load, seizure risk |
| Relational | Reveals relatives | Can reveal what you are about to say/do before you do—affecting everyone you interact with |
| Persistently sensitive | More revealed over time | More revealed as decoding algorithms improve (a 2026 motor recording may decode speech in 2035) |
Prosthetics have gone through four generations. The first generation was cosmetic: a wooden hand that looked like a hand but did not move. The second was body-powered: a hook that opened and closed when you shrugged your shoulder, operated by cables. The third was myoelectric: electrodes on the skin of the stump detected muscle activity and opened or closed a mechanical hand. The fourth generation—today's frontier—is tissue-integrated, neuromorphic, and bidirectional.
The difference between third- and fourth-generation prosthetics is not incremental. It is the difference between a tool you wear and a body part you inhabit.
Xu Jialing, a Paralympic swimmer who lost her arm in a fireworks accident at age 12, stands at the cauldron of the Asian Para Games. She reaches out with her right hand—her BrainCo neural prosthetic hand—and lights the flame. Every finger moves independently. She does not look at the hand. She thinks about closing her fingers around the torch, and they close. For the 85,000 people in the stadium, this is magic. For the 24 million people in China with limb disabilities, this is the future arriving on schedule.
The 2025 Science paper by Shu and Herr (MIT) marked a threshold moment. Their tissue-integrated bionic knee for above-knee amputees did three things no previous prosthesis could do reliably:
The result: subjects could walk faster than with traditional prosthetics, climb stairs, navigate obstacles, and position the prosthesis in free space without visual feedback. One subject exceeded the maximum walking speed of their intact biological leg. Herr, himself a double amputee, put it plainly: "It is not just a tool a human uses. It becomes part of the self."
The socket is the most hated component in all of prosthetics. It must be fitted precisely to the residual limb, but the residual limb changes volume throughout the day with fluid shifts, temperature, and activity. Too loose and it rubs, causing blisters and infection. Too tight and it cuts off circulation. It must be donned and doffed daily. It traps sweat. It cannot transmit sensory feedback because there is no direct interface with the nervous system through it.
Osseointegration eliminates the socket entirely. A titanium fixture called an abutment is surgically implanted into the bone (femur for above-knee, tibia for below-knee, humerus for above-elbow). Over several months, the bone grows onto the titanium surface—a process called osseointegration, the same biological mechanism that makes dental implants work. A connector passes through the skin via a stoma (a carefully maintained skin opening), and the prosthesis attaches directly to this connector.
The advantages are not merely comfort. The direct skeletal connection means that forces from the prosthesis transmit directly to the skeleton, allowing the user to feel weight and pressure through the bone itself (osseoperception). Combined with AMI muscle reinnervation, this provides proprioceptive feedback accurate enough to walk without watching one's feet.
Movement without sensation is control without competence. You can hold an egg without breaking it only if you can feel how hard you are squeezing. You can walk across uneven ground only if you can feel the terrain through your feet. You can feel a loved one's hand only if pressure and temperature signals reach your somatosensory cortex.
Traditional myoelectric prosthetics restore movement but not sensation. The user must watch the hand constantly. They crush plastic cups and drop fragile objects because they cannot feel grip force. They cannot feel texture, temperature, or pain.
The March 2025 breakthrough from Johns Hopkins University (published in Science Advances) addressed this with a biomimetic prosthetic hand featuring a three-layer neuromorphic tactile sensing system:
The system achieved 98.38% accuracy in texture discrimination, compared to 83.02% for traditional rigid prosthetic hands and 82.31% for soft robotic hands. It could distinguish rough from smooth, hard from soft, and detect slip fast enough to reflexively adjust grip force before an object fell.
But sensing on the prosthesis is only half the problem. The other half is delivering that sensory information back to the nervous system in a way the brain can interpret—this is the "write" direction of the bidirectional neural interface. The January 2024 Nature Communications perspective from the University of Electronic Science and Technology of China mapped the technical roadmap: neuromorphic hardware that uses analog/mixed-signal CMOS circuits to directly emulate the spiking behavior of mechanoreceptors, rather than simulating them in power-hungry software on digital processors. This allows the sensory encoding to run at microwatt power levels, suitable for implantation.
2026 was the year bidirectional closed-loop brain-computer interfaces moved from laboratory curiosity to clinical reality. The Chinese Ministry of Science and Technology launched the first multi-center clinical trial of fully implanted BCIs on May 18, 2026. On May 19, the world's first patient with simultaneous invasive BCI and timed spinal cord stimulation was reclassified from complete spinal cord injury (ASIA A) to incomplete injury (ASIA C)—meaning restored sensory and motor function below the injury level. That is not a prosthetic replacing a limb. That is the original limb being reconnected to the brain via a digital bridge.
At the Guangdong-Hong Kong-Macao Greater Bay International BCI Industry Conference, the Shenzhen Institutes of Advanced Technology demonstrates non-invasive ultrasound BCI. No craniotomy, no electrodes implanted in the brain. Focused ultrasound through the skull activates mechanosensitive ion channels on neurons at 10 cm depth with millimeter precision. In pilot studies with chronic disorders of consciousness, 13 of 24 patients showed clear behavioral improvement; some regained speech and the ability to eat independently. This is not the far future. It was presented two days before this paragraph was written.
The philosophical shift represented by closed-loop systems was captured by Dr. Jiang Hongjie of Zhejiang University Second Affiliated Hospital, speaking at the 2026 World Science and Technology Development Forum: "We are not sending commands to the brain. We are listening to the brain first, then having a conversation."
Open-loop deep brain stimulators for Parkinson's disease delivered constant stimulation 24 hours a day, regardless of whether the patient was experiencing tremor at that moment. It was like playing the radio at full volume all day because you wanted to hear the news at 6 PM. Closed-loop systems (such as the Chinese-made DBS system that entered multi-center clinical use in May 2026) continuously monitor local field potentials, detect the specific neural signature that precedes symptom exacerbation, and deliver targeted stimulation only when and where it is needed—reducing total stimulation time from 1440 minutes/day to approximately 30 minutes/day for some indications, with fewer side effects.
| System | Route | Channels | Indication | Status (2026) |
|---|---|---|---|---|
| Neuralink N1 | Intracortical | 1024 | Paralysis, Blindsight (visual) | 12 patients implanted; Blindsight human trials initiated |
| 博睿康 NEO-ONE SCI | Intracortical | 256 | Spinal cord injury | NMPA Class III medical device approval (March 2026); world's first commercial fully implanted BCI |
| Synchron Stentrode | Endovascular (stent) | 16 | ALS | COMMAND trial; 10 patients implanted; first permanently implanted vascular BCI |
| Paradromics Connexus | Intracortical | 768 | Voice decoding for locked-in patients | 200+ bps; FDA clinical approval; first human recordings 2025 |
| Precision Neuroscience Layer 7 | Epidural (thin film) | 1024 | General BCI | FDA short-term human use authorization (2025); film thinner than human hair |
| CASIA Ultrasound BCI | Non-invasive (transcranial ultrasound) | 10,000+ | Disorders of consciousness, epilepsy, depression | Pilot clinical results (2026); no implantation required |
The technologies described in this part are life-transforming. They are also extraordinarily expensive and unevenly distributed. As of 2026, the global population of amputees who could benefit from fourth-generation neuroprosthetics is estimated at 20–30 million people. The number who have actually received a tissue-integrated neuromorphic device worldwide is fewer than 100.
This gap is not a market failure that will resolve itself. These devices will follow the standard trajectory of medical technology if unguided: they will be available first to wealthy patients in wealthy countries, covered by premium insurance plans, and will trickle down to the general population over 20–30 years. That is three decades during which millions of people will be denied the ability to walk, to grasp, to feel, because they happened to be born too early or too poor.
The MERCY Protocol addresses this through the Endure pillar's capital structure, but the algorithmic fairness question runs deeper. Even when devices are available, the machine learning algorithms that decode neural signals are trained on data from patients who received implants early—predominantly male, predominantly white, predominantly young, predominantly healthy enough to qualify for clinical trials. These algorithms may generalize poorly to women, to older patients, to patients with comorbidities, to patients whose neural anatomy differs from the training distribution.
Input: Multicenter neural recording dataset D,
demographic attributes A (age, sex, etiology, comorbidities),
minimum equalized odds threshold ε
Output: Decoder f with performance parity across groups
1: Group-stratified train/validation/test split preserving
demographic proportions (no data leakage across subjects)
2: Pretrain base decoder f0 on all data using cross-entropy loss
3: for each demographic group g in A do
4: Evaluate per-group decoding accuracy Acc_g
5: end for
6: while max(Acc_g) - min(Acc_g) > ε do
7: Identify worst-performing group g_worst
8: Apply group-specific adversarial debiasing head
9: Augment training with weighted sampling from g_worst
10: Re-evaluate all Acc_g
11: end while
12: Validate that overall accuracy does not degrade by >5%
13: Deploy with continuous monitoring of per-group performance drift
Result: Decoding accuracy parity within ε across all demographic
groups, preventing algorithmic discrimination in who gets
"good" control of their prosthesis.
Grace Hopper did not invent COBOL because she loved verbose syntax. She invented it because she believed that the people whose lives were governed by computer programs should be able to read those programs. In 1959, programs were written in assembly language—cryptic, opaque, accessible only to specialists. COBOL used English verbs: MOVE, ADD, IF, PERFORM. A business manager could read a COBOL program and understand what it was doing with payroll, with inventory, with benefits.
That principle—the rules that govern people must be legible to the people they govern—is the direct ancestor of the MERCY Protocol's transparency requirements. The algorithms that triage disability claims, the control systems that move bionic limbs, the cryptographic systems that protect genomic and neural data—all of these must be legible. Not just to specialists. To the people whose lives depend on them.
Seventy years later, COBOL still runs 70–80% of the world's financial transactions. It processes an estimated $3 trillion per day in the United States alone (Thomson Reuters, 2021). The Social Security Administration still runs primarily on COBOL. State unemployment systems still run on COBOL. Banking core systems still run on COBOL. There have been dozens of attempts to replace it over the decades. None have fully succeeded—not because COBOL is technically superior, but because it is reliable, it is understood, and the cost of error in replacing it is catastrophic.
The lesson for the MERCY Protocol is not that we should write new systems in COBOL. The lesson is about continuity. The systems that govern social benefits must work reliably for longer than any political cycle, longer than any technology generation, longer than any individual career. They must be maintainable by ordinary programmers. They must not require exotic expertise to debug. When we build the post-quantum audit ledgers, the neural decoders, the MORTAR triage engine—we must build them with the same Hopper principle: verifiable, comprehensible, maintainable, and fundamentally legible.
A glass-walled data center on the outskirts of Baltimore. Inside, rows of post-quantum HSMs sit next to racks of COBOL mainframes—the old and the new running side by side, because both are required. The COBOL systems calculate benefits with the same English-readable verbs Grace Hopper specified in 1959. The HSMs protect genomic data and neural implant telemetry with lattice-based cryptography that will remain secure through the quantum era. The MORTAR triage engine continuously reorders the queue, ensuring that no one dies waiting while a lower-risk claim is processed first.
A few blocks away, at a community clinic, a 62-year-old woman who lost her left leg to diabetes in 2028 walks in for a follow-up appointment. She is wearing ordinary shoes on both feet. Her osseointegrated bionic leg adjusts automatically as she climbs the three steps to the entrance—she does not think about it, just as you do not think about how your biological leg adjusts to stairs. She feels the ground through the implant, just as you feel the ground through your foot.
Outside, by the river that used to be a swamp, children are playing. A public kiosk allows anyone to check their claim status in plain language, to see exactly where they are in the queue and why, to verify that their data has not been accessed without authorization. No passwords are required for public status inquiries; no biometrics; no knowledge of bureaucratic procedure. The system sees people. It remembers them. It reasons fairly about them. It repairs what it can. And it endures.
The river is flowing again.
Open source repository: huggingface.co/caustino
Second Edition · November 2026 · Released under open license for humanitarian implementation