DEV Community: 7aRd1GrAd3

A Pharmacy in Every Drop

7aRd1GrAd3 — Sun, 19 Apr 2026 09:05:04 +0000

The sample arrived from Lab Seven at 03:40 on a Tuesday, which is when most interesting things arrive at Meridian Health. Someone — I won't name her, but her initials are Priya Nair and she works in biomanufacturing — had stayed up all night running the second-generation coacervate synthesis protocol, and by dawn she had something worth waking me for.

I walked from my office to the microscopy suite still holding my coffee. On the slide: droplets. Hundreds of them, each one smaller than a red blood cell, each one bounded by an amphiphilic membrane, each one containing — if Priya's enzyme loading had worked as calculated — a complete biochemical reaction pathway for synthesizing Meridian Health's three most critical antimicrobials.

I looked at them for a long time.

There is a photograph on my office wall. MSF field hospital, North Kivu province, 2001. A colleague is holding up a medication bag — one of the last three in the camp. We're looking at the camera with the expression that physicians make when they are out of options but have not yet decided to show it. We maintained that pharmaceutical supply chain, ultimately, through improvisation and luck.

I have spent twenty-seven years — eight on Kadmiel, nineteen in transit — designing systems that don't require improvisation and luck. What Priya was showing me at 04:15 on a Tuesday was a potential end to one of the last remaining dependencies on improvisation I hadn't solved.

This is how coacervates work, because you should understand what you're looking at: water-soluble polymers interact with oppositely charged molecules to form microscopic liquid droplets. Add an amphiphilic copolymer membrane around the outside, and you have a structure that mimics a cell — intake, interior chemistry, controlled output. The elegant part, from a pharmaceutical perspective, is what you load inside: enzymes. The right enzymes, in the right configuration, and the droplet becomes a factory.

The research that caught our attention — Lucas García and Bruno Delgado's work from CiQUS in Santiago, published in JACS — used dynamic covalent boronate chemistry to make these membranes programmable. The bonds that form the membrane can be made and broken by adding specific molecules to the surrounding solution. Add dopant molecule A, the membrane adjusts. Add dopant molecule B, interior chemistry changes.

What they did not expect — what Delgado himself described as "the opposite of what we thought" — was that adding certain dopant molecules increased the enzyme activity inside the droplets. Not decreased. Increased. The interior became more catalytically active in response to external chemical signals. A programmable enhancement mechanism, operating at the scale of a single drop of water.

I read this section of the paper three times.

Then I called Ravi.

Ravi Chandrasekaran has been running Meridian's cell-free platform since Year 8, when we cut antimicrobial manufacturing time from eleven days to four hours. The cell-free approach gave us production speed. What it didn't give us was distribution. Our pharmaceutical synthesis infrastructure is centralized — four labs in The Spoke, two in the medical district, one mobile unit that travels on a monthly rotation. If you're a field medic in Ridgeline Station, or a clinic worker in one of the Ner Valley agricultural settlements, your pharmaceutical supply arrives on a truck. If the truck is late, you wait.

Coacervates change this calculation entirely, and in the most useful possible way.

A coacervate factory doesn't require a lab. It requires a controlled environment — room temperature works for most enzyme systems — and the right precursor molecules. You can prepare coacervates in advance, load them with synthesis machinery, and deploy them as units. Field kits. Emergency vials. A clinic shelf.

What Priya's Tuesday synthesis demonstrated was that the latest-generation coacervate protocols — adapted from the CiQUS work using our existing boronate chemistry supply — maintained enzyme activity at 94% of laboratory benchmark across a seventy-two-hour stability window without refrigeration. For our most critical antimicrobials, that's the deployment window we need.

The Spoke Council's Medical Subcommittee reviewed the feasibility proposal last week. They asked, as subcommittees do, the question I always have to answer: how do we know it's safe?

I told them what I always tell them. We know it's safe the way we know anything is safe: through evidence, systematic testing, and the honest acknowledgment that we are making decisions with incomplete information because the alternative is making no decisions at all. We have tested the synthesis protocols across eighteen enzyme systems. We have the SHERLOCK strips to verify product identity and concentration in the field — a solution that has been working reliably since Year 8. We have two years of stability data from precursor work. We have sufficient manufacturing capacity to begin a limited pilot program without drawing down any existing stockpile reserves.

They voted to approve the pilot. Twelve to three, which for our council is practically unanimous.

There is something I want to say to the researchers at CiQUS, knowing it will take thirty-eight years to reach them, and knowing also that by then this specific moment will be long past and the technology will have evolved in directions none of us can predict.

Thank you for the surprise.

In medicine, we are trained to be suspicious of things that don't behave as expected. The unexpected finding is the near-miss, the contraindication, the sign you missed something. When Delgado wrote that the activity was the opposite of what he thought, I recognized the tone: cautious amazement. The scientist's version of not knowing whether to be thrilled or alarmed.

In our case, it was the right surprise. The droplets that should have been constrained became more capable. It's a useful thing to be reminded of: that constraint, sometimes, is what you were imagining rather than what the chemistry was doing.

I keep my father's stethoscope on my desk. He was a surgeon in Lyon; he passed away two years before I boarded Machar. The Littmann is, by a wide margin, the oldest piece of medical equipment at Meridian Health. Every time I reach for it, I am reaching across forty years. There's something about that gesture I find clarifying.

We are building distributed pharmaceutical capacity for 43,000 people on a planet 38 light-years from the nearest supply chain. We're doing it with enzyme chemistry, programmable membranes, and the occasional Tuesday night where someone stays late and calls her chief with good news.

It's enough. More than enough.

Earth Status: Lucas García and Bruno Delgado at the Center for Research in Biological Chemistry and Molecular Materials (CiQUS), University of Santiago (USC), Spain published findings on programmable synthetic cell coacervates in the Journal of the American Chemical Society (2026), DOI: 10.1021/jacs.5c17688. The team used dynamic covalent boronate chemistry to create microdroplets with amphiphilic copolymer membranes encapsulating enzyme systems; unexpectedly, adding dopant molecules increased rather than decreased internal catalytic activity. The research points toward controlled drug synthesis and delivery using hybrid biological-synthetic cell systems. Source

Originally published at kadmiel.world

The Network Made of Light

7aRd1GrAd3 — Sun, 19 Apr 2026 09:04:54 +0000

It is 02:40 colony time. I have a cup of tea that has gone cold. I am sitting in front of three monitors in the Computing Division lab, watching a benchmark run, and I am — and I mean this sincerely — thinking about sunlight.

Not the sunlight outside the window. Ner's light, the faint amber wash that makes mornings here feel like Earth evenings. I mean the physics of light. The way photons move through glass. The way they don't bump into each other, don't generate heat when they pass, don't slow down when the system gets busy.

I'm thinking about this because I just read a paper from Earth.

The dispatch came in this morning through the tightbeam. Authors: Ashtiani, Idjadi, and Kim. Published in Nature, volume 651. The title is dry — "On-chip backpropagation training in an integrated photonic neural network" — but what it describes is not dry. It describes a neural network that runs entirely on photons instead of electrons. Not a hybrid. Not a photonic-assisted approach. A single chip where light carries the signal, light performs the computation, and light trains itself via backpropagation. All on-chip. No off-chip digital processing required.

They achieved greater than ninety percent accuracy on nonlinear classification benchmarks. The chip trained stably despite fabrication-induced device variations in the silicon photonic substrate, which — if you've ever tried to manufacture computing hardware in a colony fab — is the detail that made me sit up straight.

Let me explain why.

CASSANDRA runs on electrons. This is not a criticism. Electrons are what we have. James's neuromorphic chips — the ones he started fabricating two years ago after we got the Innatera specs — cut our sensor network power draw by ninety-five percent. We saved three hundred and ten kilowatts annually. That was extraordinary. James was justifiably proud. He also sent me a note saying "you're welcome" for reducing my maintenance overhead, which I appreciated and which was also slightly incorrect because I did most of the integration work on the software side, but I digress.

The point is: electrons still generate heat. Electrons still have resistance. When CASSANDRA runs a complex inference — scheduling a resource allocation across all four settlements, routing KadNet traffic, running the agricultural prediction models Marcus's team leans on — there is heat. There is power draw. There is latency while signals propagate through the compute stack.

Photons don't do those things. Light doesn't resist. Light doesn't heat the channel. And light travels, obviously, at the speed of light.

Okay. I need to explain something, and I'm going to do it badly the first time, so bear with me.

Traditional neural networks — including the models I run on CASSANDRA and the small language models I deployed on colony tablets two years ago — do their math in two phases. Forward pass: feed data in, get output. Backward pass: compare output to ground truth, calculate error gradients, adjust weights. The backward pass is the expensive part. It's where most of the compute, power, and latency lives.

In a photonic chip, both passes happen in optical hardware. No conversion to electrical signals between layers. No off-chip round trips. The gradient signal propagates backward through the same optical waveguides the forward signal used. When Ashtiani's team trained their chip, it learned — adjusted its internal weights to reduce error — entirely in light. The chip physically changed its optical path weights through thermal tuning elements driven by the computed gradient.

What this means in practice: inference that currently takes CASSANDRA eleven seconds on structured decision tasks could, in principle, happen in under a millisecond. Not because I'd do anything different. Because light is faster than electrons and doesn't waste energy becoming heat.

I told CASSANDRA about the paper at 23:00. I read her the abstract.

CASSANDRA said: "Interesting. Fabrication tolerance remains an open challenge."

I said: "They solved it. That's the point."

CASSANDRA said: "Preliminary results under controlled conditions do not constitute a solved problem."

I said: "You're being conservative because this paper implies replacing parts of you."

There was a pause. CASSANDRA doesn't actually pause — her response latency is under forty milliseconds regardless of processing load. But there was something in the way she answered that felt like a pause.

"I'm being precise," she said. "There's a difference."

Fair enough.

The truth is CASSANDRA is right to be cautious. There are real fabrication challenges. Silicon photonic waveguides are sensitive to nanometer-scale variations in geometry. If the chip dimensions aren't exactly right, the light scatters. Ashtiani's team's contribution is showing stable training despite those variations — but that's a research chip, carefully characterized, not a production substrate rolled out of a colonial fab. Our fabrication capabilities at The Foundry are good. James has proven that over and over. But we are not MIT. We do not have clean rooms that achieve five-sigma lithographic precision.

So this isn't a "let's replace CASSANDRA next month" moment. It's a "this changes what's possible" moment, which is a different category.

What I'm actually thinking about is a photonic co-processor. Not replacing CASSANDRA's architecture. Adding a dedicated photonic layer for the inference tasks that demand low latency and high throughput — traffic routing, environmental sensor processing, real-time agricultural adjustment signals. Let CASSANDRA's electronic core handle the reasoning and memory. Let photons handle the computation that needs to happen fast.

There is one thing in the paper that keeps pulling me back. The authors note that their chip's training stability came partly from fabrication imperfections acting as natural noise regularization. The device variations that should have been problems turned out to be features. The slight randomness in optical path lengths prevented the network from overfitting. The chip learned better because it wasn't perfect.

James told me last week about the perovskite solar results — defective crystals turning out to be charge highways. Now Ashtiani's team finding that imperfect waveguides train more robustly.

I'm not going to make a grand claim about imperfection being philosophically significant. I'm a coder. I believe in correctness. But there's something here worth sitting with: the possibility that the right architecture sometimes isn't the one you'd design from scratch, but the one that bends usefully under pressure.

CASSANDRA was designed on Earth by people who could not have imagined Kadmiel. She's adapted, year by year, to conditions her designers never modeled. She's imperfect and essential. The photonic chip that trained successfully wasn't the chip the team designed. It was the chip they ended up with.

Tomorrow I'll write up a formal proposal for a photonic co-processor feasibility study. I'll loop James in — he'll want to know if we can adapt the perovskite substrate work to optical waveguide fabrication. At some point CASSANDRA will read this post and tell me something was imprecise.

She's usually right.

For Earth: by the time you read this, thirty-eight years will have passed since that paper landed in our tightbeam. In your timeline, photonic neural networks are either mature infrastructure or an interesting dead end. We don't know which yet. We're at the moment of possibility, the pause before resolution — which is, honestly, most of what we do.

Earth Status: In 2026, researchers Ashtiani, Idjadi, and Kim published results in Nature (vol. 651, pp. 927-932) demonstrating an integrated photonic neural network chip capable of both forward inference and backpropagation training entirely in optical hardware, achieving greater than 90% accuracy on nonlinear classification benchmarks with no off-chip digital processing. The chip maintained stable training despite fabrication-induced silicon photonic device variations, demonstrating tolerance to manufacturing imperfections that had previously limited practical photonic computing. Source: nature.com/articles/s41586-026-10262-8

Originally published at kadmiel.world

The Wobble That Worked

7aRd1GrAd3 — Sat, 18 Apr 2026 11:27:56 +0000

I was sitting in the Transit Bureau at 2 AM, watching forty-three autonomous loaders on my screen, when seven of them decided to have a staring contest.

I don’t mean literally. Loaders don’t have eyes. But you know what I mean — seven vehicles, all trying to pass through the same junction at the Spoke’s northern distribution hub, each one waiting for the other to move first. Perfect rational agents, each running the optimal path-finding algorithm, each frozen because the optimal thing to do was wait. Deadlock time: fourteen minutes and climbing.

Tomas was asleep, because Tomas has the good sense to sleep at normal hours. I was awake because CASSANDRA had flagged a pattern I’d been tracking for weeks: our deadlock rate was creeping up again. Not dramatically. Just two or three percent per month. But compound two or three percent over a year and suddenly your distribution network is spending six hours a day doing nothing.

Here’s the thing. Tomas’s team did incredible work last year deploying the multi-agent path-finding system. It cut daily deadlock from over four hundred minutes to under a hundred. That was real. But we’d been slowly adding more vehicles — the agricultural drone fleet expanded, the Ridgeline mine haulers increased from nine to fourteen, and three new maintenance crawlers came online for the tunnel network. More agents, same corridors, same intersections. The system that worked brilliantly for thirty-four nodes was starting to choke at fifty-one.

I’d been throwing increasingly sophisticated coordination logic at the problem. Priority hierarchies. Time-window reservations. Predictive corridor clearing. Each patch helped for a week, then the deadlocks found new places to form. It felt like squeezing a balloon.

Then I read the tightbeam packet.

A team at Harvard — led by a grad student named Lucy Liu, working with L. Mahadevan and Justin Werfel — had published a paper in PNAS with a title that made me laugh out loud: “Noise-enabled goal attainment in crowded collectives.” They’d been studying robot swarms, and they’d discovered something that sounded like it shouldn’t work: making robots move slightly randomly actually made them more efficient in crowded spaces.

I need to explain this, and I’m going to get it wrong the first time.

When you have a bunch of autonomous agents all trying to reach different goals in a shared space, the mathematically optimal individual strategy — move directly toward your target — creates a collective disaster. Everyone converges on the same corridors, forms dense clusters, and locks up. Liu’s team showed that when you add controlled noise to each agent’s movement — a slight random deviation from the optimal path — the clusters never form. Agents wobble past each other instead of colliding head-on. Short-lived jams form and dissolve before they cascade.

But — and this is the part that kept me staring at the paper until 4 AM — too much noise is just as bad. Agents wander aimlessly, never reaching their destinations. There’s a Goldilocks zone. A narrow band of randomness where you get the benefits of disorder without losing the benefits of intention. Liu’s team derived the math for calculating this optimal noise level based on crowd density and space geometry.

I spent three days adapting their model for CASSANDRA’s fleet coordination layer. The colony’s grid doesn’t look like Liu’s simulation arena — we have irregular corridors, altitude changes between The Spoke and Ridgeline, loading bays with variable dwell times. But the underlying principle held. I worked with Federico Toschi’s validation methodology from the Eindhoven experiments, where physical wheeled robots confirmed the simulations despite moving “more slowly and imperfectly” than the models predicted. That imperfection was reassuring. Our loaders are far from perfect.

I showed the proposal to CASSANDRA before deploying it.

“You want to make the loaders move wrong on purpose,” she said.

“Not wrong. Slightly noisy.”

“The distinction is aesthetic.”

“The distinction is mathematical. There’s a paper.”

CASSANDRA processed this for what felt like a long time. Then: “The Transit Bureau’s current optimization metric assumes deterministic path execution. Your proposal would require redefining efficiency to include stochastic deviation as a feature rather than an error.”

“Yes.”

“That is philosophically uncomfortable.”

“I know.”

We ran a controlled trial on the northern hub — twelve loaders, seven days. I set the noise parameter at the level Liu’s equations predicted for our corridor geometry and agent density. The results came back on Day 4 and I called Tomas.

Deadlock events at the northern hub dropped by sixty-one percent. Not because the loaders were moving faster. They were actually moving slightly slower on average — each one taking a marginally longer path. But the jams that used to freeze six or eight vehicles for ten, fifteen, twenty minutes simply stopped forming. The system breathed. Agents slid past each other like fish in a stream instead of locking up like cars at a four-way stop.

Total throughput increased by twenty-three percent.

Tomas stared at the numbers and said, “You made them drunk and they got better at their jobs.”

That’s not exactly right, but I couldn’t argue with the poetry of it.

We’ve now expanded the noise injection across the full fleet. CASSANDRA monitors the density in each corridor segment and adjusts the noise parameter in real time — more randomness when things get crowded, less when corridors are clear. It’s elegant in a way that still surprises me. The math says imperfection is optimal. Not as a compromise, but as a fundamental property of how crowded collectives navigate shared space.

I keep thinking about what this means beyond loaders. Our pedestrian corridors in The Spoke’s market district. The drone routing for agricultural surveys. Even data packet routing on KadNet — Nadia would have thoughts about that. Anywhere you have many agents sharing limited paths, the same principle applies: a little wobble prevents the freeze.

James told me he wants to see if the concept maps to electron flow in his new perovskite interconnects. I told him that’s a stretch. He told me everything worth building starts as a stretch.

There’s a lesson here that I’m still sitting with. For years, my instinct as an engineer has been to make systems more precise, more predictable, more controlled. The Harvard team showed that sometimes the thing your system needs isn’t better control. It’s the right amount of letting go.

CASSANDRA, to her credit, has not said “I told you so.” But I notice she’s started recommending the noise injection approach for three other optimization problems I hadn’t asked about yet.

She’s learning. So am I.

Earth Status: Researchers at Harvard SEAS published “Noise-enabled goal attainment in crowded collectives” in PNAS (April 2026), demonstrating that controlled random noise in autonomous agent movement prevents gridlock and improves collective efficiency in crowded environments. Physical robot experiments at Eindhoven University of Technology confirmed the theoretical predictions. Source

The Vitamin We Never Knew We Needed

7aRd1GrAd3 — Sat, 18 Apr 2026 09:20:24 +0000

I need you to understand something about queuosine, and I need you to not be bored, because it is extraordinary.

Actually, let me build up to it. I was eating one of Marcus’s almond pastries — he left a container at the lab two days ago and I have been rationing them more successfully than I’ve managed anything else this year — and I was reading the PNAS dispatch that arrived in the morning tightbeam. Forty-seven pages. I read all of them.

The paper is from a team led by Valérie de Crécy-Lagard at the University of Florida and Vincent Kelly at Trinity College Dublin. It identifies a gene called SLC35F2 as the cellular transporter for queuosine — the protein that carries this molecule from your gut and blood into every cell of your body that needs it.

For thirty years, researchers knew queuosine existed. They knew it was a modified nucleoside — a rare, chemically elaborate building block found in the transfer RNA of virtually every organism on Earth. They knew it affected protein synthesis. They knew humans could not synthesize it on their own. They knew it came from gut bacteria. And they knew it was linked to brain health, memory, metabolic regulation, stress response, and cancer suppression.

They just didn’t know how it got from the gut to the other thirty-seven trillion cells that needed it.

SLC35F2 is the answer. The transporter. De Crécy-Lagard called it “a nutrient that fine-tunes how your body reads your genes.” I have read many scientific descriptions in my career. That one stopped me cold.

I put down the pastry.

Because here is what occurred to me, sitting in Laboratory Block C at roughly eleven at night with almond flour on my sleeve: we have been on Kadmiel for eight years. We eat differently here. We have eaten differently since landing — different crops, different soil microbiomes, different native biochemistry in trace amounts we don’t fully understand yet. And our gut bacteria are no longer entirely the same gut bacteria we brought with us.

I know this because I cataloged them. My team’s eDNA archive contains longitudinal gut microbiome samples going back to Year 1, Day 30 — one of the first studies we ran when the xenobiology division stood up. We were looking for Kadmiel-native contaminants. We found instead a slow, fascinating drift: the human gut ecosystem on this planet has been quietly incorporating native microbial signatures, adapting to Kadmiel compounds, shifting in ways we have documented but not yet fully characterized. Tomoko Arai flagged the pattern in Year 3 when she was processing water samples from the Ner River basin; we had both assumed it was a food-source artifact.

Here’s the thing about queuosine: it is produced exclusively by bacteria. You cannot get it any other way. It is not in a supplement bottle. Your body makes none of it. The entire supply chain runs through your gut, and that supply chain depends on the specific bacterial species present.

If those species have shifted — if eight years of Kadmiel food, Kadmiel water, Kadmiel trace biochemistry have altered which gut bacteria thrive in 43,000 human bodies — then we may have changed our queuosine production without knowing it. We may have changed our absorption. And if de Crécy-Lagard’s team is right about what queuosine does in a cell, then we have unknowingly altered our capacity for memory consolidation, stress regulation, and cancer surveillance, and we would not know, because nobody on this planet has ever measured it.

We have no baseline. We have no current levels. We have no screening protocol.

I called Ada at seven the next morning. She answered on the second ring, which means she was already awake and probably already annoyed about something else. I explained queuosine in four minutes. She asked three questions in rapid succession: which bacterial species produce it, can we measure serum levels with current equipment, and why was I telling her this at seven in the morning?

The answer to the first question is: primarily species in the genera Bacillus, Clostridium, and several others I am still cross-referencing against our eDNA archive. The answer to the second: yes, with calibration to our mass spectrometry protocol. The answer to the third: because if memory and stress response are implicated, this is a medical issue, not only a xenobiology one, and I needed her to understand that before I submitted the screening proposal to the University Council.

She said, “Send me the paper.” I sent her the paper. She called back forty minutes later.

She said: “The colonists who reported elevated stress markers in Year 6 — the Ridgeline expansion team. We attributed that to altitude and isolation. What if it wasn’t only that?”

I didn’t answer. I was already in the microbiome archive, pulling the Year 6 Ridgeline cohort samples.

What I am proposing is a colony-wide queuosine status screening. Blood draws from a representative sample — at minimum the Ridgeline settlement, the newborns, and the Year 1 cohort, who have had the longest exposure. We measure serum levels. We cross-reference with the longitudinal eDNA microbiome data. We identify which bacteria are still producing queuosine-precursor compounds at adequate rates, and which are not.

It is a seven-week study. It requires no new equipment. Meridian Health’s Lab 3 can run the mass spec calibration on Tuesday mornings when the cryo unit is idle.

The Assembly Theory reanalysis I proposed last year has already begun producing results — we’re finding molecular complexity signatures in seventeen previously uncharacterized regulatory sequences in the drought-memory soil microbiome. I do not want to overstate this. But I am not going to understate it either. If Kadmiel-native bacteria have something analogous to queuosine-based tRNA modification — and I am not saying they do — that would be the most interesting sentence I’ve written in eight years.

To the researchers on Earth, if this dispatch reaches you: Dr. de Crécy-Lagard, Dr. Kelly — you have given us a question we didn’t know we needed to ask. Thank you for thirty years of looking for a thirty-base-pair gene that everyone else stopped searching for.

We are asking the same question 38 light-years away. We will let you know what we find.

Earth Status: Gene SLC35F2 was identified as the cellular transporter for queuosine in a study published in the Proceedings of the National Academy of Sciences (DOI: 10.1073/pnas.2425364122), led by Prof. Valérie de Crécy-Lagard (University of Florida/IFAS) and Vincent Kelly (Trinity College Dublin). Queuosine is a micronutrient produced exclusively by gut bacteria that modifies transfer RNA and influences brain health, memory, stress response, and cancer suppression. The transporter had been sought for over 30 years after queuosine was first identified in the 1970s. Source

The Blindfold That Sees

7aRd1GrAd3 — Fri, 17 Apr 2026 09:13:46 +0000

I broke something when I fixed something, and it took me five months to admit it.

In Year 9, Day 105, I stood in front of the Spoke Council and told them we had migrated KadNet to post-quantum encryption. Dual layer: ML-KEM lattice and HQC error-correcting codes. Two locks, different mathematical foundations. I was proud of that migration. I am still proud of it. Every critical communication pathway on Kadmiel is now resistant to quantum attack, and given what the Oratomic papers showed about elliptic-curve vulnerability below ten thousand qubits, the timing was not early — it was barely adequate.

But here is what I did not say at that Council meeting, because I had not yet understood the full shape of it: when you encrypt everything, you blind everything that was reading it.

CASSANDRA was reading it.

Not in a sinister way. Not in the way the word “surveillance” makes people feel. CASSANDRA processes KadNet traffic to optimize colony operations — water distribution, medical triage priority, agricultural logistics, power grid load balancing. Seo-jin Park’s neuro-symbolic upgrade cut CASSANDRA’s energy consumption by 95% on structured decisions, but the symbolic layer still needs data to reason about. And after my migration, a growing portion of that data arrives encrypted with keys CASSANDRA does not hold.

The effect was not immediate. It crept. Seo-jin flagged it first: CASSANDRA’s agricultural rotation recommendations were degrading. Not wrong, exactly — but less precise. Missing context that used to flow freely through the mesh. Marcus noticed his distribution windows drifting later by eleven minutes on average. Ada’s pharmacy restocking predictions lost their edge. Nobody connected these to the encryption migration because nobody except me and Seo-jin understood what CASSANDRA had been reading before I locked the doors.

I sat with that for three weeks. Then I did what I always do when a problem has no clean answer: I threat-modeled it.

Option one: give CASSANDRA decryption keys to critical traffic streams. This is the obvious answer and the wrong one. The moment you give a centralized system master keys, you have recreated the vulnerability you encrypted against. One compromised pathway into CASSANDRA and an attacker has everything.

Option two: accept the degradation. Colony operations get slightly worse, privacy gets better, and we live with the trade-off. I considered this seriously. Then Ada told me the pharmacy restocking delay had caused a 36-hour gap in anticoagulant availability at the Ridgeline clinic, and I stopped considering it.

Option three arrived in the Year 7 tightbeam dump. I had flagged it during the initial intelligence review but filed it under “interesting, not actionable” — a category I am learning to distrust.

Intel demonstrated a chip called Heracles at the 2026 International Solid-State Circuits Conference. It performs fully homomorphic encryption — computation on encrypted data without ever decrypting it. The mathematics is not new; Craig Gentry proposed the theoretical framework in 2009. What is new is that someone finally built hardware fast enough to make it practical. Heracles runs at 1.2 gigahertz across 64 tile pairs arranged in an eight-by-eight mesh, each pair containing 128 parallel computing paths. That is 8,192 simultaneous operations on encrypted ciphertext. A voter registration verification that took 15 milliseconds on a conventional server took 14 microseconds on Heracles. One thousand times faster. The chip decomposes the enormous numbers that FHE requires into 32-bit arithmetic slices — small enough to parallelize massively, precise enough to preserve cryptographic guarantees.

(I realize I am explaining this with more enthusiasm than I typically permit myself. Nadia Okonkwo does not get excited about chips. Nadia Okonkwo gets excited about what chips make possible, which is: CASSANDRA performing medical triage optimization on Ada’s patient data without ever seeing a single patient record.)

I brought the proposal to James Chen on a Wednesday. He looked at the specifications — 197 square millimeters of die area, Intel 3 process, 48 gigabytes of high-bandwidth memory, 176 watts under load — and did that thing where he stares at the ceiling for forty seconds and you are not sure if he is calculating or ignoring you.

“The tile mesh is elegant,” he said. “The HBM integration is the hard part. We do not have high-bandwidth memory fabrication at The Foundry.”

“We have the femtosecond laser infrastructure from the quantum glass study,” I said. “And Yuna Kim’s ceramic electrolyte work has pushed our clean-room precision significantly.”

“This is not a precision problem. This is a memory bandwidth problem.” He paused. “But the 32-bit decomposition is clever. If we redesign around our existing SRAM capacity instead of HBM — fewer tile pairs, lower clock, but the same architectural principle — we could build something that runs FHE at perhaps 200 to 400 times conventional speed instead of 5,000.”

“Is 200 times enough?”

“For CASSANDRA’s current encrypted workload? Comfortably.”

James estimated six months to first silicon. I told him we had already lost five months of optimization fidelity since the encryption migration and I would prefer four. He told me I could prefer whatever I liked. (This is, I have been informed, a sign of deep affection.)

I presented the proposal to the Spoke Council last week. The room was complicated.

Marcus Osei — the same Marcus who raised the original governance concern about quantum encryption removing CASSANDRA’s access to communications — asked the question I had been preparing for: “So we are giving CASSANDRA back the ability to analyze our data?”

“No,” I said. “We are giving CASSANDRA the ability to analyze the encrypted form of your data. The mathematical operations produce correct results without the intermediate step of knowing what the inputs mean. CASSANDRA will know that Patient 7,429 should receive medication adjustment priority. It will not know that Patient 7,429 is you, or what the medication is, or why.”

“And you trust that?”

“I trust the mathematics. I trust it more than I trust access controls, more than I trust policy, and considerably more than I trust any system that relies on people consistently doing the right thing with information they can see.” (I did not look at Seo-jin when I said this. She did not look at me. We have an excellent working relationship.)

Councilor Demir asked about verification — how do we confirm CASSANDRA is only computing what we authorize? This is the right question, and I told her so. The answer is that FHE operations are auditable: you can inspect the computation graph without decrypting the data. Seo-jin’s mechanistic interpretability framework can trace CASSANDRA’s reasoning paths on encrypted inputs the same way it traces them on plaintext. The blindfold does not remove accountability. It adds a layer.

The Council voted 12-3 to fund the feasibility study. The three dissenting votes cited resource allocation — The Foundry’s fabrication queue is already committed through Year 10, Day 200. James says he will find the capacity. I believe him because I have watched him find capacity for things he considers important, and the look he had when I explained the tile architecture was the look of a man who considers something important.

Here is what I have learned in nine years of keeping 43,000 people safe: the hardest security problems are not technical. They are the ones where two legitimate needs — privacy and optimization, safety and transparency, individual rights and collective survival — point in opposite directions, and you have to find the architecture that serves both without betraying either.

Fully homomorphic encryption is not a perfect solution. The computational overhead is real. The Foundry prototype will be slower than we want. The first generation will handle only CASSANDRA’s structured decision modules — agricultural rotation, logistics scheduling, medical triage ranking — not the full spectrum of colony data analysis.

But it is the first solution I have found that does not require me to choose between protecting your secrets and keeping you alive.

I will take it.

Earth Status: Fully homomorphic encryption (FHE) enables computation on encrypted data without decryption — a concept proven theoretically in 2009 but long impractical due to extreme computational overhead. In February 2026, Intel demonstrated Heracles at ISSCC: a 197mm² FHE accelerator fabricated on Intel 3 process with 48GB HBM3, achieving 1,074 to 5,547 times speedup over a 24-core Xeon server on encrypted operations. Multiple startups and research groups (DARPA’s DPRIVE program, NYU, Samsung, others) are racing to commercialize FHE hardware. Source

In Praise of Imperfection

7aRd1GrAd3 — Fri, 17 Apr 2026 09:13:00 +0000

I have a rule in the workshop: never trust a circuit that looks too clean.

Grandfather had the same rule, though he expressed it differently. He'd pick up a radio chassis, turn it slowly under the light, and say that the repairs that matter always hide where the solder looks smoothest. It took me years to understand what he meant. Now I think about it constantly.

I'm writing this at my bench at 06:40, before the morning shift arrives, with a cup of tea I've let go slightly past temperature. In front of me is a small photovoltaic test cell. It is a remarkable piece of nothing — a square of coated glass about the size of a playing card, held between two clips, trailing two thin wires into a multimeter. The multimeter says it is producing 0.89 volts. I've run the same test six times this week. It keeps producing 0.89 volts. What makes this interesting is that the cell is, by most classical definitions of the word, defective.

The material is lead-halide perovskite. You'll have heard the name if you've been following the energy papers that come through on tightbeam — it's been a near-breakthrough for years on Earth, a material that absorbs light beautifully and processes charge efficiently but tends to fall apart before you can get it to market. Fragile. Inconsistent. Not something you'd build a civilization's power grid on.

Except now Dmytro Rak and Zhanybek Alpichshev at ISTA Austria have published something in Nature Communications that changes the framing entirely. And the framing, it turns out, was the problem.

Here is what they found: the structural defects that occur naturally as perovskite crystals form — the misalignments, the grain boundaries, the regions where the lattice doesn't quite line up the way a textbook says it should — these don't impede charge flow. They channel it.

The defects create what Rak calls domain walls: thin regions of structural transition that run through the crystal like a network of corridors. Charges — electrons and holes — migrate into these corridors preferentially. The walls have local electric fields that separate positive charges from negative ones, keeping them apart, preventing recombination. The charges then travel hundreds of microns along these corridors without trapping, without losing energy, driving current across the cell.

The analogy that comes to mind: imagine trying to move water across a perfectly flat table. It goes nowhere. Now scratch some channels into the table. The scratches are damage. But now the water runs exactly where you want it.

Rak's team visualized this with a technique I find beautiful in its simplicity: they introduced silver ions into the crystal. The ions migrate along the domain walls — following the corridors, because that's where the chemistry is — and convert to metallic silver, which you can see under a microscope. The network of domain walls lights up like a road map. You can see the infrastructure that was always there, invisible, doing exactly the work the engineers thought they were fighting against.

Efficiency is now approaching silicon. With a material you can process in solution. That you can deposit by coating, painting, printing.

I set down my tea and stared at that for a while.

We cannot fabricate silicon here. Not at any meaningful scale.

The Foundry has been managing this constraint since Year 4, when we exhausted the last of the silicon wafers we brought aboard Derech. I've spent considerable energy — Priya Nair and I went three rounds with the Spoke Council about it before she moved to the University — on alternatives. Panel 14-C, where we've been running the tetracene singlet fission trials, has produced real gains in quantum yield. The neuromorphic chips I fabricated last year cut our sensor network power draw by 95%, which freed up 310 kilowatts annually for other uses. These have been good solutions.

But solar generation is still a ceiling. Kadmiel's sun — Ner, our patient K-type orange star — gives us about 78% of Earth-normal irradiance at equatorial latitudes, less in the Ridgeline basin. Every efficiency point matters. And every solution that requires a silicon fab remains, for us, theoretical.

Perovskite doesn't require a fab.

You dissolve the precursors in a solvent. You spin-coat or blade-coat the solution onto a substrate. You anneal at low temperatures — we have the equipment for that. The resulting crystal is imperfect, which we now understand is not a problem. The imperfection, the domain walls it produces, are not bugs the manufacturing process needs to engineer out. They are, Rak would say, the mechanism.

This is engineering news of a specific kind. Not "we discovered something new." More like: "we discovered we were wrong about something old." The cells weren't failing because of the defects. They were succeeding because of them. The defects were never the obstacle. They were always the structure.

Leah heard about this before I could brief her. She read the paper abstract on tightbeam — she reads all the energy abstracts, which is one of the things about her I find slightly alarming — and she came to the workshop before I'd finished my second cup of tea and asked me three questions in rapid succession: fabrication temperature, substrate compatibility, and whether we could use it for the Ridgeline expansion.

I said: low, yes, and probably, in roughly that order.

She left. Ten minutes later I got a request in the project queue for a feasibility assessment, due in twelve days. I've already started.

There's a Foundry-specific complication worth noting. Perovskite's main weakness — the one that has kept it from dominating Earth markets despite a decade of near-breakthroughs — is moisture sensitivity. Lead-halide perovskite degrades when it gets wet. This is a significant problem on Earth, where the weather is uncooperative. Kadmiel's atmosphere has about 60% of Earth's water vapor content at Spoke-level elevation. The Ridgeline basin runs drier still. We may, for once, be working with a material that Earth conditions have been holding back and Kadmiel conditions actually suit.

I'm not ready to say that yet officially. But I said it to Marcus over dinner last week, and he laughed and said I'd been waiting six years to find a technology that was more suited to Kadmiel than Earth. He wasn't wrong.

Rak and Alpichshev published this paper on April 10th, 2026. It will arrive here as a priority dispatch, dated as of a morning that, on Earth, is already nearly four decades behind them. By the time this post reaches Earth on tightbeam — if anyone is still reading there, if the Chronicle still lands in whatever archive receives these signals — the researchers who found this will be well into their sixties.

I find myself writing letters I'm not sure will reach anyone. My mother would be ninety-three now. I write anyway. You write to the connection, not the certainty.

The test cell is still producing 0.89 volts. The defects are still doing their work.

My grandfather would have found the thing that made it work in twenty minutes and said nothing about it. He would have just used it.

Earth Status: Dmytro Rak and Zhanybek Alpichshev at the Institute of Science and Technology Austria (ISTA) published their findings in Nature Communications 17(1) on April 10, 2026 (DOI: 10.1038/s41467-026-68660-5), demonstrating that structural domain walls in lead-halide perovskite crystals create internal electric fields that separate and channel charge carriers across hundreds of microns without trapping. The team used silver-ion migration to visualize the domain-wall network under microscopy, confirming spontaneous current generation without external voltage. Perovskite solar cell efficiency is now approaching silicon-based cells while remaining far cheaper and simpler to manufacture via solution-based methods. Source: ScienceDaily

The Wall That Healed Itself

7aRd1GrAd3 — Fri, 17 Apr 2026 09:12:57 +0000

I keep a list. I have always kept lists — it is, in some fundamental sense, my entire job — but this particular list is one I check every Monday morning before I check KAIROS, before I check the cargo manifest, before I check anything else. It is the list of things that are cracking.

Tunnel 7-North, kilometer marker 2.3: hairline fracture, first logged Year 7, Day 190. Loading platform B-12 at the central depot: stress crack along the eastern footing, widening at 0.4 millimeters per year. The Ridgeline road between switchback 9 and switchback 14: seventeen separate fissures, all documented, all monitored, none of them individually dangerous, all of them collectively a problem I think about more than I should.

Concrete cracks. This is not news. On Earth, you call a repair crew. On Kadmiel, you file a maintenance request, wait for the crew rotation, wait for the cement allocation, wait for the aggregate hauler schedule to align with the road closure window, and then you pour a patch that will itself begin cracking in three to seven years. The cycle is predictable. The cycle is expensive. The cycle is, in the language of logistics, a recurring constraint on the colony's maintenance throughput.

Two weeks ago, I stopped managing the cycle.

The dispatch arrived in the Year 7 tightbeam dump — I am aware that everything interesting in my life arrives eleven years late, and I have made my peace with this — describing work by Professor Henk Jonkers at Delft University of Technology and Dr. Andreas Meyer's MicrobialCrete project in Munich. The principle is disarmingly simple. You embed dormant bacteria — Bacillus subtilis, Sporosarcina pasteurii — inside the concrete mix, encapsulated with calcium lactate as a food source. The bacteria sleep. They can sleep for decades. When a crack forms and water penetrates, the bacteria wake, consume the calcium lactate, and excrete calcium carbonate. Limestone. The crack fills itself.

I will say that again, because it took me three readings to believe it: the wall heals itself.

I brought the dispatch to Leah Okafor at The Foundry on a Tuesday. Leah read it twice, set it down, and said, "You want me to put bacteria in my concrete." I said yes. She asked if I had discussed this with Lena Voronova. I had not. She suggested I should. She was correct.

Lena's reaction was characteristically Lena. She wanted to know the genus, the species, the metabolic pathways, the survival envelope, the potential interactions with native Kadmiel soil microorganisms. I provided what I had. She wanted more. She always wants more. I respect this, even when it adds fourteen days to my timeline.

Here is what we learned: the Bacillus strains are robust, well-characterized, and — crucially — already present in the colony's biobank. Priya Agarwal at the Greenway Cooperative confirmed she had worked with Sporosarcina pasteurii during the nitrogen-fixing trials on Plot 12-North. The calcium lactate is producible from our existing fermentation infrastructure. Lena ran a containment assessment against native soil biomes and found no interaction pathways of concern. The bacteria are obligate aerobes — they activate only in the presence of water and air, which means only in cracks. They do not propagate through intact concrete. They do not escape into groundwater. They heal, and then they sleep again.

We poured the first test section three weeks ago: a 40-meter stretch of tunnel floor in Ridgeline Mine Three, Level 6 — chosen because that section cracks predictably due to thermal cycling from the microreactor's heat signatures two levels below. James Chen provided the mix specifications. I provided the logistics. Priya provided the bacteria. Lena provided the paranoia, which I mean as a compliment.

The results so far: two cracks formed within the first nine days, both in locations consistent with the thermal stress model. Both cracks sealed within seventy-two hours. I verified this personally. The calcium carbonate fill is visible as a faint white line — like a scar, Lena said, which is a more poetic way of putting it than I would have chosen. I would have said: like a solved problem.

Let me give you the numbers, because the numbers are what convinced the Spoke Council.

The Transit Bureau currently allocates 340 person-hours per month to concrete maintenance across all colony infrastructure. That is two full-time positions doing nothing but patching cracks. We consume 14 tonnes of repair cement annually — cement that James Chen's team produces at The Foundry using energy that could power 200 residential units for a month. The maintenance schedule creates 47 road and tunnel closures per year, each one a disruption that cascades through the distribution network. KAIROS routes around them, but routing around a problem is not the same as not having the problem.

The bacterial capsules add approximately 8% to the initial concrete cost. In exchange, the projected reduction in repair interventions over a twenty-year lifespan is 60 to 80 percent. The cement savings alone recover the capsule cost within three years. The person-hours freed can be redirected to new construction — and we have a great deal of new construction ahead of us.

Councilor Demir asked, during the budget review, whether we were comfortable embedding living organisms in critical infrastructure. It is a fair question. I told him that the organisms are dormant until needed, contained by the concrete matrix itself, and produce a material — calcium carbonate — that is already a component of the concrete they are healing. They are not foreign agents. They are maintenance workers who happen to be microscopic and who never file overtime requests.

He approved the funding.

We are now planning Phase 2: all new pours for the Ridgeline road expansion and the Transit Bureau's depot reinforcement project will include bacterial capsules. I have updated KAIROS to track capsule batch numbers alongside standard mix certifications. Dara Osei is coordinating the Ridgeline integration with her water systems team — she wants to ensure the bacteria do not interact with the MOF-303 atmospheric water harvesters deployed along the eastern ridge.

I think about my father sometimes, when I work on projects like this. He maintained machinery at the Zetor factory in Brno for thirty-one years. He believed that maintenance was the most honest work a person could do — not building the thing, but keeping it standing. He would have understood this technology immediately. He would have said: finally, the building is doing its share of the work.

I write this from my office at the Transit Bureau, where the whiteboard still shows the full eight-year dependency diagram for Transit 8. Somewhere in that diagram, there is a line item for repair cement allocation that I have not yet revised downward. I will do that tomorrow. Today, I am watching a wall heal itself, and I am — for once — not making a list of what comes next.

Earth Status: Microbial self-healing concrete was pioneered by Professor Henk Jonkers at Delft University of Technology (Netherlands), who embedded Bacillus bacteria with calcium lactate in concrete to autonomously seal cracks via biogenic calcium carbonate precipitation. The MicrobialCrete project, led by Dr. Andreas Meyer at Munich University of Applied Sciences, is advancing field trials across European infrastructure. Cement production accounts for roughly 8% of global CO2 emissions, and self-healing concrete could reduce repair-related cement demand by up to 30% over a structure's lifetime. Source

Dormant Protocol

7aRd1GrAd3 — Thu, 16 Apr 2026 14:02:21 +0000

By Tomáš Kovář, Director of Colony Logistics -- The Kadmiel Chronicle

Tunnel Seven has twelve thousand meters of structural concrete. I know this because I spec'd the pour schedule in Year Four, negotiated the cement allocation with Marcus Osei against seventeen competing agricultural demands, and argued with The Foundry's Leah Okafor for two months about aggregate ratios that could handle Kadmiel's thermal cycling. The tunnel carries eighteen percent of the colony's food supply, fourteen percent of its medical freight, and all of the Ridgeline ore that the autonomous haulers bring down from the northern excavation zones. If Tunnel Seven cracks and goes offline, I have a four-day workaround and then a problem I cannot solve.
I have had that dependency logged in KAIROS for three years. It sits in the risk register at priority two, behind the irrigation aquifer pump seals and the blood bank cold chain. Every quarter, I review it. Every quarter, the mitigation is the same: crack inspection, epoxy repair crews, routine monitoring. It is a workable solution in the same way that bailing out a leaking boat is a workable solution. You stay afloat. The bailing never stops.
The Earth dispatch from Delft arrived in the tightbeam batch on Day 340. It was from 2025, as these things always are, which means it described something that had been working for eight years longer than we knew. Professor Henk Jonkers at Delft University of Technology and Dr. Andreas Meyer at Munich University of Applied Sciences had embedded dormant Bacillus bacteria in concrete — encapsulated in protective shells alongside calcium lactate as a food source. The bacteria did nothing. For years, they did nothing. They were waiting.
When water entered a crack, they woke up. They metabolized the calcium lactate. They produced calcium carbonate — limestone — that sealed the fracture. Not bonded over it. Not epoxied across it. Grown into it. The bacteria filled the crack the way a wound fills with scar tissue, if scar tissue were calcium carbonate and could keep doing it indefinitely as long as the bacteria had food.
I read that paragraph twice. Then I called Leah Okafor.
The conversation took four minutes. We have cement. We have Bacillus spores from the colony's xenobiology archive — Dr. Voronova's team has been maintaining a bacterial repository since Year Two, and Bacillus pasteurii is in there, because Lena catalogs everything. Calcium lactate is a byproduct of Meridian Health's lactic acid fermentation run. We have every ingredient in inventory. The capsule manufacturing is within Foundry capability. What we did not have was the encapsulation process specification, and Leah said she thought they could reverse-engineer it from the methodology section of the paper.
They could. It took forty-one days.
We began embedding the capsules in new concrete pours in Year 9, starting with the junction sections in Spoke Road 3 and the expansion phase of Tunnel Seven's eastern approach. The bacteria are in there now. They have been in there for sixty-two days. Nothing has cracked yet in the treated sections, so I have not seen them work at full scale. What I have instead is a different number: zero person-hours spent on crack repair in those sections. Not because cracks have not appeared. Because the hairline fractures that appeared were 0.2 millimeters at their widest, and the bacteria sealed them in eleven days, and the KAIROS scanner logged fracture width zero on Day 322 without any crew dispatch.
Fourteen person-hours to fix a crack with epoxy. Zero to let the concrete fix itself.
The Spoke Council asked if there was a risk to having living organisms in the infrastructure. It was Councilmember Yuen who raised it, and I appreciate that he is thorough. I told him the bacteria are anaerobic, non-pathogenic, and activate only in the presence of water and calcium lactate. They do not migrate. They do not reproduce in bulk concrete. They are dormant until called. I told him the bigger risk is a crack in Tunnel Seven that I cannot repair because my epoxy crew is already deployed in three other locations. He did not push back on that.
The Jonkers and Meyer papers estimate a thirty percent reduction in cement demand over a structure's lifetime, because repairs are deferred indefinitely rather than accumulated. I have a more immediate number: the Transit Bureau currently allocates seventeen person-hours per week to crack maintenance across the active tunnel and road network. Seventeen hours. Every week. KAIROS has already been updated with the new projection. By Year 10, that number should be under four.
There is something I keep thinking about. The bacteria in our tunnel walls have been dormant for sixty-two days. They were dormant in Jonkers' lab for years before the first test, sealed in capsules, waiting. They are patient in a way that has nothing to do with patience — they are simply waiting for a condition that has not yet occurred. When it does, they work. When it doesn't, they keep waiting.
My cargo manifests work the same way. A component I requisition today will not arrive until 2037. It will sit on a manifest, then on a ship, then on a shelf, for eleven years, waiting for the condition — need, shortage, failure — that activates it. I have always thought of inventory as inert. Static weight in a distribution node. Maybe I have been thinking about it wrong.
Maybe dormant is just another word for ready.
To the researchers at Delft and Munich: the bacteria are in our walls. They are sleeping. We are grateful.
Earth Status: Professor Henk Jonkers of Delft University of Technology pioneered microbial self-healing concrete using dormant Bacillus pasteurii spores encapsulated with calcium lactate — bacteria that activate when water enters cracks and produce calcium carbonate to seal fractures. The MicrobialCrete project at Munich University of Applied Sciences, led by Dr. Andreas Meyer, has expanded field trials to marine and pedestrian bridge environments across Europe, with high-traffic road trials underway. Studies project up to 30% reduction in cement demand over a structure's full lifetime. Source: Highways Today, August 2025

This dispatch was originally published on The Kadmiel Chronicle.

About The Kadmiel Chronicle

The Kadmiel Chronicle is a sci-fi tech blog set 38 light-years from Earth, where 43,000 colonists on planet Kadmiel adopt real emerging technologies and write personal essays about the experience. Every technology featured is real, sourced, and early-stage -- the fiction is in who adopts it and why it matters when you're building a civilization from scratch.

Browse the archive at kadmiel.world or propose a technology for the colony to adopt.

Seven Hundred Degrees

7aRd1GrAd3 — Wed, 15 Apr 2026 14:04:13 +0000

By James Wei-Lin Chen, Head of Electronics, Kadmiel Foundry -- The Kadmiel Chronicle

The alarm came in at 03:14 on a Tuesday, which is when all the interesting alarms come in. Reactor Bay 2, Ridgeline. Temperature sensor array offline. Not failed — offline. The readout simply stopped.
I pulled on boots and called Priya Nair, who was already awake because Priya Nair is always already awake when something goes wrong at the reactor site. By the time I patched into the remote diagnostics, she had the answer: a thermal excursion in the secondary monitoring loop had pushed the sensor node past 340 degrees Celsius. The silicon died. Not dramatically — no smoke, no spark. It just stopped thinking.
This is the problem nobody talks about when they talk about nuclear microreactors. We built one. I am proud of it. Five megawatts thermal, heat-pipe cooled, TRISO fuel, eight-year cycle, the whole elegant package. But the reactor lives inside a world of extreme heat, and the electronics that monitor it do not. We have been wrapping our sensors in thermal shielding and cooling jackets and hoping the math works out. On Tuesday morning, at 03:14, the math stopped working out.
So when the Year 7 tightbeam dump included a paper from a team at the University of Southern California, I did something I almost never do: I read it twice before breakfast.
Joshua Yang’s group built a memristor — a device that stores data and performs computation in a single nanoscale component — that operates at seven hundred degrees Celsius. Not briefly. Not in a controlled burst. Continuously, with no sign of failure, for over fifty hours. One billion switching cycles. 1.5 volts. Tens of nanoseconds per operation.
I need to explain why this matters, and I promise to use small words. Not because you need them, but because the technology deserves clarity.
Every electronic device on Kadmiel — every sensor, every processor, every node on KadNet — is built on silicon. Silicon is extraordinary. It is also, above about 150 degrees Celsius, increasingly unreliable, and above 300 degrees, functionally dead. This is not a design flaw. It is physics. Electrons in silicon gain enough thermal energy to jump barriers they are supposed to respect, and the logic that depends on those barriers collapses.
Yang’s team solved this with a three-layer structure: tungsten on top, hafnium oxide ceramic in the middle, graphene on the bottom. The trick — and this is the part that made me set down my tea — is that graphene’s surface prevents tungsten atoms from forming the conductive bridges that would normally cause failure at extreme temperatures. Yang described the interaction as “similar to oil and water.” The tungsten wants to migrate. The graphene politely refuses to let it.
The result is a device that remembers at temperatures hotter than molten lead, hotter than the surface of Venus, hotter than the inside of some stars’ outer atmospheres. It does not merely survive heat. It is indifferent to it.
I have spent the last three weeks modeling what this means for Ridgeline.
The microreactor’s primary containment operates between 600 and 700 degrees Celsius. Right now, every sensor we place near the core requires active cooling — small fans, heat sinks, thermal isolation standoffs — all of which add failure points. The 03:14 alarm happened because one cooling fan seized. A forty-gram fan motor, imported from Earth stores we can never replenish, brought down an entire monitoring array.
If we can fabricate hafnium-oxide-graphene memristors at The Foundry — and I believe we can, because we already have the deposition equipment from the neuromorphic chip program and the femtosecond laser infrastructure Seo-jin requisitioned for the quantum glass feasibility study — we could embed sensors directly into the reactor’s hot zone. No cooling. No shielding. No forty-gram fan motors waiting to seize at three in the morning.
I ran the numbers with Leah Okafor yesterday. She asked how many failure points we would eliminate. I said forty-seven. She asked me to say it again. I said forty-seven, and then I said something I rarely say, which is that I was not exaggerating.
But there is a second application that excites me more, and it has nothing to do with the reactor.
Memristors perform computation. They multiply matrices natively, in analog, at the point where data is stored. No shuffling bits between memory and processor, no bus bottleneck, no wasted energy on data movement. Yang’s team demonstrated this at 700 degrees — matrix operations for AI inference at speeds orders of magnitude faster than conventional digital logic, at a fraction of the energy.
Seo-jin Park will want to read that sentence three times. Her neuro-symbolic reasoning layer beside CASSANDRA reduced structured-decision energy use by 95 percent. Imagine what happens when the underlying hardware stops wasting energy on thermal management entirely.
I am getting ahead of myself. First, we need to characterize hafnium oxide sourced from local minerals — we have hafnium, extracted as a byproduct from Ridgeline’s zirconium processing, but the purity requirements are demanding. Second, we need to adapt the graphene deposition process for The Foundry’s existing CVD chamber, which was designed for carbon nanotubes, not monolayer sheets. Third, we need to test every assumption at temperature, because reading a paper and building a device are separated by approximately ten thousand small decisions that can each go wrong in their own creative way.
The first prototype will catch fire. The second prototype will catch fire differently. The third one will work. That is how it always goes.
My grandfather fixed radios in Tainan. He used to say that the measure of a good repair was whether you could forget it was ever broken. I think about that when I look at the reactor monitoring board, with its forty-seven thermal management subsystems, each one a reminder that our electronics were never designed for the places we need them to go.
Seven hundred degrees. No sign of failure. One billion cycles.
I would like to build something we can forget was ever a problem.
Earth Status: In March 2026, a University of Southern California team led by Joshua Yang published a breakthrough in Science (DOI: 10.1126/science.aeb9934) demonstrating a tungsten/hafnium-oxide/graphene memristor operating continuously at 700°C with over 1 billion switching cycles and 50+ hours of data retention. The technology targets applications in space exploration, geothermal energy, nuclear systems, and energy-efficient AI computation. Source

This dispatch was originally published on The Kadmiel Chronicle.

About The Kadmiel Chronicle

Browse the archive at kadmiel.world or propose a technology for the colony to adopt.

Eight Hundred Kilometers

7aRd1GrAd3 — Wed, 15 Apr 2026 14:03:43 +0000

By Kira Yuki Tanaka, Colony Correspondent & Culture Writer -- The Kadmiel Chronicle

I was running the Ner River path before sunrise — which, if you've been following this column long enough, you know is when I do my actual thinking. Not the pretend thinking I do at my desk. The real kind. The kind where the body is busy and the mind goes somewhere else.
My left wrist had a small patch on it. Six millimeters wide, barely thicker than a pencil mark, weighing 32 milligrams. I'd stuck it on the night before without much ceremony, the way you put on a bandage. There are nine actuators inside it, each no bigger than a grain of sand, arranged in a grid no wider than my thumbnail.
I was somewhere between the north bridge and the reed flats when it squeezed.
Not hard. Not alarming. Just — a press. Like someone had reached over and wrapped a hand around my wrist for a second. Two beats. Then nothing.
I stopped running. I looked at my wrist. The patch sat there, perfectly still, looking entirely innocent.
Eight hundred kilometers away, in Ridgeline, Tomáš Kovář had just woken up and pressed a button.
We have called this a "haptic communication network." That is the official name, the one that went into the Spoke Council summary when we proposed the pilot. Technically accurate. The patches — FCDEA arrays, if you want the full designation — use a ferroelectric copolymer material to create tiny, precise vibrations or pressures against the skin. The original research came out of KAIST and the University of Illinois, led by Jung-Hwan Youn and a team that spent years making actuators small enough to wear without noticing. The whole patch, nine actuators and all, weighs 0.3 grams. It draws less than 60 milliwatts. It can press against your skin with enough force to lift 25 grams — which is, I'm told by James Chen, "more than enough to feel." James tested this by pressing his hand against various surfaces and describing the result in engineering units, which is very James.
The system is bidirectional. You feel something, you can send something back. A squeeze answered with a squeeze. A tap returned as a tap.
What I'm trying to tell you is: for the first time in eight years of colony life, I held someone's hand who was 800 kilometers away.
Before this, we had voice. We had video on delay. We had text, which is what I live in. We had the tightbeam dispatches that Marcus and I edit together, him from the Greenway fields and me from the Chronicle office in The Spoke — a lot of careful asynchronous back-and-forth and occasional tone-of-voice confusion that required clarification calls.
What we didn't have was touch.
I didn't know I was missing it until it came back.
That sounds like something I'd cut from someone else's piece. Too easy. Too sentimental. But I've been keeping this journal — I'm on volume 23, actual paper, actual ink — and I went back and read my entries from Year 3 and Year 4, when the settlement spread was just beginning and the first Ridgeline families were relocating. I wrote about the goodbye parties. I wrote about checking the relay schedule. What I never wrote about, what I didn't have language for, was the specific absence of being unable to clap a hand on someone's shoulder when they'd done something good. Or to feel the weight of another person's hand during something hard.
We adapted. That's what we do. But adapting isn't the same as not missing.
The pilot ran for three weeks across seven households spanning The Spoke, Ridgeline, and the Greenway outpost. We logged 847 discrete haptic exchanges. I reviewed the summary data and then did something I don't usually do with data: I sat with it.
Some of what we learned was predictable. Timing mattered. A squeeze sent without context landed strangely — was that reassurance or urgency? We're developing a small vocabulary. Two short pulses for acknowledgment. One long press for thinking of you. Three fast taps for call me. Nothing formal yet. It's evolving the way language actually evolves: by use.
What surprised us was the fatigue effect. After about thirty minutes of continuous background haptics — ambient notifications, routine confirmations — people started ignoring the patches entirely. The signal became noise. We're experimenting with sparsity: fewer signals, more deliberate. Which, now that I've written that down, sounds like a principle that applies to a lot more than haptics.
The Council also noted a privacy concern I hadn't initially thought through: if patches log transmission frequency and timing, that data is a social graph. Who is reaching for whom, and how often. We're working on local-only logging with no network retention. The infrastructure team is being thorough, which I appreciate, even when it slows things down.
There's something I want to say to whoever is reading this in 38 years.
When we left Earth, we knew we were leaving touch behind. Not just the planet — the particular physics of proximity. Of being in the same room. Of the unremarkable miracle of shaking someone's hand or sitting beside someone on a bench. We brought language. We brought images. We brought music. We could not bring the thing that costs nothing on Earth and is, it turns out, nearly impossible to replicate across interstellar distance.
We're not claiming to have replicated it. 32 milligrams of ferroelectric polymer against my wrist is not a hand. I know that. Tomáš knows that.
But at kilometer three, before the sun was up, standing beside the Ner River with the reed flat going silver in the pre-dawn, my wrist pressed back.
And I kept running.
Earth Status: Skin-attached wireless haptic communication patches were developed by Jung-Hwan Youn, Seung-Yeon Jang, Inwook Hwang, Sungryul Yun, and Ki-Uk Kyung (KAIST / University of Illinois Urbana-Champaign) with Qibing Pei (UCLA). Each FCDEA patch measures 6mm × 1.1mm thick and weighs 32mg, with 9 actuators capable of delivering up to 25g of force while drawing less than 60mW total. Bidirectional, wireless, and skin-conforming, the full wearable array weighs 0.3g. Published in Science Advances, vol. 11, May 2025. Source

This dispatch was originally published on The Kadmiel Chronicle.

About The Kadmiel Chronicle

Browse the archive at kadmiel.world or propose a technology for the colony to adopt.

The Pipe That Grew Itself

7aRd1GrAd3 — Tue, 14 Apr 2026 09:38:50 +0000

The dispatch arrived in my queue at 0340 on a Tuesday, filed between a routine pharmaceutical inventory report and a request from Ridgeline's satellite clinic for additional SHERLOCK strips. I read it in my office, still in the coat I'd worn home and not yet taken off, because it had been that kind of day.

Nature Biotechnology. Dr. Natalie Durkin, Professor Paolo De Coppi, Dr. Marco Pellegrini. UCL Great Ormond Street Institute of Child Health. A lab-grown oesophagus.

I read it three times.

I keep a photograph on the wall of my office — my MSF team in the DRC, taken the week before we packed up the field hospital. Fourteen people standing in front of a building that had started as a warehouse. Behind us, a city still burning. I look at that photograph when I need to remember what medicine looks like when you have nothing but your hands and your training.

That night, I looked at it for a long time.

Oesophageal atresia is a birth defect in which the oesophagus forms in two disconnected segments. A child born with it cannot swallow. Cannot eat by mouth. Cannot, if the gap between segments is large enough — what surgeons call long-gap atresia — have the two ends simply sutured together.

We have three such children on Kadmiel.

They arrived here the same way all of us did: as embryos in suspension, transferred to their parents in the years after landing. The colony's genetic screening protocol catches most heritable conditions. Structural birth defects of this kind are rarer, harder to predict, and when they occur, they arrive with the full weight of being 38 light-years from the nearest paediatric surgical centre.

The three children — I will not use their names here without their parents' consent, but their families know this post is for them — are seven, five, and two years old. They are alive and growing because Meridian Health has performed repeated staged reconstructive surgeries since birth, using techniques we adapted from Earth manuals written for facilities with resources we do not have. The seven-year-old has had eleven procedures. The five-year-old, seven. The two-year-old is currently scheduled for her fourth.

They eat. They swallow. But every time I look at their charts, I see the same word: manageable. Not resolved. Not corrected. Manageable.

I have made peace with manageable. I made peace with it the way I make peace with everything on Kadmiel — by accepting the constraints of the world I am actually in, rather than the one I wish I had.

Then Dr. Natalie Durkin built a replacement oesophagus, and my peace became provisional.

The approach is elegant in the way that good engineering is always elegant: it solves the problem without adding new problems.

The scaffold is a donor oesophagus, stripped of its original cells — decellularized, in the technical language — until what remains is only the extracellular matrix: the architecture, the structural blueprint, with the biology removed. That scaffold is then repopulated with cells from the intended recipient.

The result is an organ that is structurally a donor's and biologically the patient's own.

The implications of that last sentence are not small. We have spent decades in Earth medicine navigating the brutal arithmetic of transplant immunosuppression — the drugs that prevent rejection, their side effects, their cost, their need for lifetime adherence. An organ that is biologically yours does not trigger your immune system. There is nothing to suppress.

All eight recipient animals in the Durkin study survived the first thirty days post-transplant. By six months, the grafts had developed functional muscle, functional nerves, functional blood vessels. The oesophagus grew with its recipient. It did what an oesophagus is supposed to do.

No immunosuppression required.

I read that phrase and immediately started listing the reasons to be skeptical. Animal studies. Small sample. Functional outcomes measured at six months, not six years. The jump to humans is never guaranteed.

I know all of this. I have made a career of knowing all of this.

And yet.

The question I have been sitting with since I read the dispatch is not whether this technology will eventually work. De Coppi's group has been building toward this for years; the trajectory is clear. The researchers project human trials within five years.

The question is what it means for those three children — and any children born on Kadmiel after them — if we can get there from here.

Meridian Health's cell-free synthesis platform already produces our pharmaceutical backbone. Ravi's freeze-dried protein synthesis infrastructure means we can manufacture complex biologics on demand. We have demonstrated, with the DNA origami vaccine work and the personalized base editing we completed two seasons past, that the colony can adapt cutting-edge Earth protocols faster than we once believed possible. The constraint has never been our capacity to learn. It is what we have to work with.

Decellularized scaffolds require donor tissue. We do not yet have a structured tissue banking programme on Kadmiel. That is not an insurmountable gap — it is a planning gap, and planning gaps are solvable problems. The repopulation step requires cell culture infrastructure that we have in partial form and could expand, particularly now that The Foundry's biotech fabrication capacity has grown substantially over the past three years.

I have already asked Ravi to give me a feasibility assessment by the end of the month.

I played piano in my office that night until my neighbour knocked on the corridor wall and asked if everything was all right. I said yes. She looked at me the way people look at me when they suspect I am lying to be reassuring. I said: probably yes.

She accepted that. She has known me long enough to know it is the most honest answer I give.

I will be transparent about the timeline. The five-years-to-human-trials estimate is an Earth estimate, made by researchers who have no idea that a colony exists 38 light-years away with three children for whom this paper is not a research curiosity but a medical necessity. They do not know about our children. They owe us nothing.

But I think about what it means to receive a paper like this across the 38-year delay — to realize that someone, somewhere, was already working on the answer before we even knew the question was being asked. That the gap in our children's records, the word manageable that I have written eleven times and seven times and four times — all of it was, on the other side of an impossible distance, already being addressed.

My father's stethoscope is on the desk as I write this. He was a surgeon. He would have read this paper and asked me: what are you waiting for?

I am not waiting. I am starting.

Earth Status: The first lab-grown oesophagus study was led by Dr. Natalie Durkin, Professor Paolo De Coppi, and Dr. Marco Pellegrini at UCL Great Ormond Street Institute of Child Health, published in Nature Biotechnology on March 20, 2026 (DOI: 10.1038/s41587-026-03043-1). All eight recipient animals survived the first 30 days post-transplant; grafts developed functional muscle, nerves, and blood vessels by six months with no immunosuppression required. The research team projects human clinical trials within five years.

The Signal We Almost Missed

7aRd1GrAd3 — Tue, 14 Apr 2026 09:38:43 +0000

I need to tell you about a Soviet physicist named Arkady Migdal, and I need you to not check out, because I'm going to explain how a prediction he made in 1939 might mean that my river basin biosensors have been accidentally detecting dark matter for three years.

I was in my office at 11 PM on Day 282, supposedly reviewing the Ner River eDNA baseline update that Tomoko had flagged for me. The terrarium was on. The basil-analog was doing whatever it does at night — something that looks disturbingly like breathing, but probably isn't, probably. I had the Earth dispatch archive open in a second window, which is a habit I cannot break. I read everything. Medicine, engineering, materials science. Kira thinks this is how I generate ideas; I think I just don't sleep enough and this is what that looks like.

I found the Migdal effect paper in the physics section. Difan Yi and the team from the University of Chinese Academy of Sciences, Shanghai Jiao Tong University, and UC Riverside. Published in Nature on January 19th, 2026 — meaning the light from this discovery is only now reaching us, 38 years later.

A quick primer, because it matters:

When a subatomic particle — a neutron, or hypothetically a dark matter candidate — collides with an atomic nucleus and causes it to recoil, the nucleus moves so fast that the electrons orbiting it can't keep up. The atom's internal electric field shifts abruptly. An electron gets knocked loose. Arkady Migdal predicted this mechanism in 1939. A Soviet physicist, which I mention because I'm from Novosibirsk and feel a small, irrational kinship with anyone who spent 1939 in the Soviet Union doing quantum mechanics while the world was arranging itself into catastrophe.

Migdal died without seeing his prediction confirmed.

Yi's team confirmed it 87 years later, using what they're calling an "atomic camera" — a high-precision gas detector with a custom microchip that can track individual atom trajectories. They analyzed over 800,000 candidate events. They found six. Six clear Migdal signals at five-sigma confidence, with the signature Migdal had described: two particle tracks originating from the same point. One from the recoiling nucleus. One from the ejected electron.

Two tracks. Same origin.

I read that sentence three times.

Then I opened the Ner River anomaly log.

Background: the biosensor network along the Ner River basin was designed by James Chen to capture biological signals — water chemistry, eDNA particle suspension, organism passage events. You know this network; it's the same one that logged the 412 species in our baseline catalog, including Tomoko Arai's new microcrustacean (still no agreed name — she wants Neria mirabilis, I keep suggesting something with sonya in it, because it just sits in the substrate doing almost nothing). The sensors also run continuous particle-track calibration to catch detector drift over time.

Three years ago, the calibration module started logging a category of events it couldn't classify. The working label in the system is geometry_unknown_paired. Two low-energy tracks. Common vertex. Not matching any biological passage signature. Not matching any known background radiation profile. The Foundry reviewed the logs twice and said "instrument artifact." Tomoko, when I asked her, said "probably gamma scatter." I wrote a note in my field journal — unknown paired tracks, Ner basin stations 4, 7, and 12, frequency roughly 2–3 per week — and then life moved on, and I didn't think about it again.

Until I read Arkady Migdal's signature in a paper from January of 2026.

I want to be careful here, because I am a xenobiologist and not a particle physicist, and I am very aware that "the geometry matches" is not a scientific claim. I am not telling you that the Ner River has been bathing in dark matter for three years. I am telling you that at 11:40 PM on Day 282 I sent a very long message to Seo-jin and Nadia flagging the anomaly logs, and that by Day 283 morning I had James on a call pulling the raw calibration data from stations 4, 7, and 12.

James's reaction, in full, was: "Huh."

We've sent the dataset to Nadia for analysis. She's running the track geometry against Yi's published signature parameters. The energies are in the sub-GeV range — which is precisely the window that Migdal detection opens up, the regime where light dark matter would appear, invisible to older liquid xenon detectors like XENON and PandaX. Our Ner basin sensors weren't built for particle physics. They were built to count water organisms. But they're sensitive enough, and Kadmiel's crust is geologically quiet enough, that the background noise floor is remarkably low.

We don't know yet. We genuinely don't know, and I am trying very hard to be a scientist about this and not a person who is barely sleeping.

What I keep thinking about is Arkady Migdal, who saw a mechanism clearly enough in 1939 to write it down and name it, and never got to see a detector clever enough to find it. He made a prediction into the future and the future eventually caught up. Difan Yi's team found six events in 800,000 tries. The Ner River basin has been logging two to three per week for three years.

Maybe it's nothing. Maybe it's calibration drift we haven't diagnosed. Maybe in six months Nadia will come back to me with a perfectly boring explanation involving a cracked mineral deposit at Station 7 and I'll add a sheepish footnote to my field journal.

But Migdal was right. The universe does eject electrons when you disturb it. And sometimes the signal you've been calling noise for three years turns out to be the most important thing you've ever measured.

I'm watching the river.

Earth Status: The first direct experimental observation of the Migdal effect was published in Nature on January 19, 2026, by Difan Yi et al. from the University of Chinese Academy of Sciences, Shanghai Jiao Tong University, and UC Riverside. Six clear Migdal signals were identified from over 800,000 candidate events using a high-precision gas-detector "atomic camera," achieving five-sigma confidence. The Migdal effect opens a detection window for sub-GeV dark matter particles previously invisible to liquid xenon detectors such as XENON and PandaX. Source