the disappearing interval

 A closing movement for “The Last Human Decision”

On August 12, 2026, in Huntsville, Alabama, the outgoing head of U.S. Space Command gave his last symposium speech and read out a list. Five capabilities the command says it will need between 2029 and 2033 to keep its edge over China in orbit: kinetic and non-kinetic space fires; means to counter an adversary’s low-orbit satellite constellations; refuelable satellites that can maneuver instead of husbanding the fuel they launched with; space domain awareness good enough for what the missile-defense world calls a “target-quality track”; and, third on the list, joint, integrated space command and control. Gen. Stephen Whiting was careful to name the genre. “This isn’t just a wishlist,” he said. “It’s a clear demand signal to several audiences.”

It is an interesting choice of words. Industry hears a requirement. Congress hears a justification. An adversary hears capability and intent. One transmission carries several messages simultaneously.

But one item deserves particular attention.

Whiting said the force must “enhance machine-to-machine connectivity to support decision-making and battle management of this global space fight on tactically relevant timelines.”

The important words may not be machine-to-machine. They may be tactically relevant timelines.

Because this essay has encountered that problem before.

In 1983, the Soviet Oko early-warning system told Stanislav Petrov that American missiles had been launched. The system was wrong. Petrov had something that increasingly sophisticated technology cannot guarantee:  time.

Time to look at the signal.

Time to compare it with what he knew.

Time to doubt the machine.

Time to decide.

More than two decades earlier, at NORAD, Air Marshal Roy Slemon confronted another extraordinary warning when the newly operational BMEWS system interpreted the rising moon as a massive incoming Soviet missile raid. He, too, occupied an interval between detection and action in which a human being could ask a profoundly important question:

Could the system be wrong?

The men mattered. Their judgment mattered. But something else mattered first.

The architecture gave them time to exercise judgment.

That distinction becomes important in orbit.

A contested space environment operates at velocities and distances that compress the interval between detection, interpretation and response. Sensors become faster. Networks become faster. Tracking becomes more precise. Machines exchange information directly because human communication increasingly becomes the slowest element in the chain.

There is nothing inherently sinister about this. Much of it is necessary. Machines can integrate sensor streams, maintain tracks and distribute information at speeds no human organization could match.

That is Licklider's promise still alive: machine and human doing different things well.

But there is another possibility.

As the system becomes faster, the machine may cease merely to provide information within the human decision window and begin determining the size of that window itself.

That is a different threshold.

The question is therefore not simply whether a human remains “in the loop.” That phrase may eventually tell us very little. A person can formally remain inside a decision architecture while possessing so little time, context or practical ability to intervene that the human role becomes largely procedural.

The better question is:  How much meaningful time remains between machine certainty and irreversible action?

Whiting does not tell us that humans are being removed from space battle management. His words actually preserve the opposite interpretation: machine-to-machine connectivity will support decision-making. That may prove to be exactly what is intended—machines managing the velocity of information while humans retain authority over consequential actions.

But the tension should not be dismissed.

Petrov's world contained minutes.

Slemon's world contained minutes.

The architecture Whiting describes is being designed for a domain in which minutes may be an eternity.

And that brings us back to the question underneath this entire essay.

For most of the history of automated warning, we have treated the human as the final safeguard against the machine's mistake. Sensors could fail. Algorithms could misclassify. Communications could distort. The human receiver remained the place where context, uncertainty and judgment could interrupt the chain.

But what happens when the speed required for the system to succeed becomes incompatible with the time required for that safeguard to function?

We should be careful here. The answer is not to romanticize slow human decision-making. Humans misinterpret signals too. They panic. They carry biases. They misunderstand adversaries. And in some circumstances hesitation can be as dangerous as haste.

The objective is not simply to preserve human delay.

It is to preserve meaningful human judgment.

That may prove to be one of the central design problems of machine-speed command and control: not merely making the system faster, but deliberately engineering places where speed stops.

A pause.

A threshold.

A requirement for corroboration.

A point at which uncertainty is allowed to remain uncertainty.

A place where the architecture itself says: not yet.

Petrov's pause was partly an artifact of the technological world in which he lived. We should not assume the next generation of systems will give it to us accidentally.

If we believe the pause matters, we may have to design it deliberately.

That is what makes the demand signal from Huntsville consequential.

Space Command is asking industry to build systems capable of operating on timelines increasingly inhospitable to unaided human cognition. That requirement may be unavoidable. It may even be essential.

But somewhere between the sensor and the weapon, between detection and decision, between machine speed and human responsibility, another requirement belongs on the list:  preserve the interval in which someone can still say no.

Petrov inherited that interval.

Slemon inherited it.

The next generation may have to build it.

 

— Samuel R. Lammie

 

A note on how this was made: This piece emerged through an iterative conversation among a human and two AI systems. I brought the underlying framework, questions, and professional perspective; Claude helped develop the initial argument and draft; ChatGPT challenged several of its stronger claims and helped reshape the piece around what I think is the more defensible question: not whether machines will replace human decision-makers, but whether the systems we design will preserve enough time for meaningful human judgment. I made the final choices. In a piece about human judgment in increasingly automated systems, that process seemed worth making visible.

 

A Return on the Investment

 On NSF, the National Academy of Sciences, and why I believe in the whole enterprise

In 1988, the National Science Foundation funded the National Center for Geographic Information and Analysis at the University at Buffalo — one of only three sites in the country, alongside Santa Barbara and Maine. That same fall, I walked into the Department of Geography to begin a master's degree. I did not know I was walking into something the NSF had just planted. I only knew that David Mark and Hugh Calkins were willing to advise a man who did not arrive empty-handed.

Because I had been doing the work of GIS for years by then — without a computer, and mostly without the name for it. As a Peace Corps conservation agent in the Guatemalan highlands, and later as a forester on the San Carlos and White Mountain Apache reservations, the daily work was spatial to the bone: siting nurseries by elevation, matching species to slope and soil, tracing a watershed by where the water wanted to go, laying land use over terrain over hydrology over the human communities living on it until the pattern surfaced. Call it analog GIS. Ian McHarg had already handed me the method — I read Design with Nature early, before Buffalo or right at its threshold — and his stacked transparencies, one ecological layer laid over another until the fitting ground for a thing declared itself, were exactly what I was doing in the field with paper and judgment and no machine at all: the layers stacked in the mind before any software could stack them for you. By 1986 I had made the jump to the digital tool, pushing the BIA Forestry Branch at White Mountain to trade its aging MOSS system for the new ARC/INFO out of a small company in Redlands. So I came to Buffalo to find the theory beneath a practice I already carried in my hands — not to be handed a practice I lacked.

I want to be plain about what that NSF grant became, because I am one of the things it became.

The three years I spent at Buffalo seeded a career of more than three decades in federal geospatial work. It carried me to Voyageurs, where the data on my desk turned wolf radiotelemetry into the territory polygons of twelve named packs — the early lineage of what people now call the Voyageurs Wolf Project. It carried me, eventually, to the Northern Region of the Forest Service, where I was responsible for the geospatial program across more than twenty-five million acres of Montana, Idaho, and the Dakotas: fire, watersheds, timber, the whole living map. None of that was foreseeable in 1988. Nobody writing that NSF award could have drawn a straight line from a center in Buffalo to a wolf pack's boundary in the boreal north, or to a fire perimeter in the Bitterroot, or to the phone in your pocket that quietly knows where you are. That is not a failure of foresight. That is the nature of the thing. Basic science is funded precisely because you cannot yet see what it will grow into. And yet the ledger is not a mystery: the Federal Reserve reckons that federally funded research has returned somewhere between a dollar-fifty and three dollars on every taxpayer dollar spent since 1950. I am, in the most literal accounting, a return on that investment. So is the map that finds you.

Neil Shubin, who this July becomes president of the National Academy of Sciences, put it better than I can. He found a 385-million-year-old fossil — Tiktaalik, the creature caught in the act of crawling out of the water — and he is the first to say he could only find it because he had funding long enough to fail. Five or six years of failure, he tells it, before the discovery. He would not have survived in an industry that measured him quarter by quarter. Neither would the fossil. Neither, for that matter, would I. Long-horizon funding is not a luxury laid over science. It is the soil science grows in.

I spent my working life in that soil, and I know its economics from the inside. In the Peace Corps I raised two tree nurseries in the highlands of Baja Verapaz — some two hundred thousand seedlings in all. You plant a nursery for a rotation you will not personally cut. You put a seedling in the ground on behalf of a forester who is not yet born. That is silviculture, and it is also exactly what the National Science Foundation did in Buffalo in 1988 — put something in the ground whose harvest would fall to people it would never meet. I am one of that harvest. So are the students who came through the NCGIA training programs and now lead the discipline. So, indirectly, is anyone who has ever watched a map find them after an earthquake.

That is my case for the NSF, and it is not really an argument. It is a witness statement.

The National Academy of Sciences deserves its own word, because it does something different and just as necessary. Chartered by Congress under Lincoln in 1863, it is the body the nation turns to when it needs to know what the science actually says — advice built through a process designed to strip out conflict and bias and return evidence, not opinion. I know something about disciplined advisory work from my own years of it, at a different table than the Academy's — I served as the Eastern Region representative on the Forest Service's national Geospatial Advisory Committee. There is a particular integrity in an institution whose job is to reduce a policymaker's uncertainty without telling him what he wants to hear. In a country now awash in voices that reinforce what people already believe, an institution built to be a clean signal is not a relic. It is civic infrastructure as real as a bridge.

I have spent a career separating signal from noise — it is the frame I see the world through — so let me be precise about the present moment, because vagueness here would be its own kind of dishonesty. In the spring of 2025 the administration terminated roughly 1,752 National Science Foundation grants, some $1.4 billion of committed work, and proposed to cut the agency that seeded my entire discipline by 55 percent. The director resigned that April, in the fifth year of a six-year term; more than a year later the chair still sits empty. Congress, to its credit, refused the deepest cut and funded the Foundation near its old baseline — the institution has defenders yet, and I do not want to bury that — but the machinery was altered underneath the appropriation. New grants have gone out at a fraction of the historical rate. An entire research directorate was shuttered. Peer review, the thing that makes an award a merit award, was thinned from three readers to two.

The National Academy of Sciences has been hit in a quieter and more telling way. Its working arm, the National Research Council — the body that produces those gold-standard reports — lost more than forty federal contracts and around two hundred staff in a single year and is bracing for what its own leadership has called a radical downsizing. And some in Congress have begun to question whether the Academy's advice is truly nonpartisan at all, urging federal agencies to go find their science reviewed somewhere else. Read that twice. An institution chartered under Lincoln for the express purpose of being a clean signal is being told, by people who would plainly prefer the noise, that its signal cannot be trusted. That is not budget-trimming. That is an attack on the instrument itself.

You cannot claim the dividend while you burn the seed. You cannot ask for the fish while you defund the six years of failure that precede it.

And still — like Shubin, and I do not say this lightly — I am an optimist about the long run, and not out of naivety. Congress rejected the worst of the cuts; the appropriation held. The seed is not yet burned. I have reason to hope. I watched a grant made before anyone knew what it would do turn into a discipline, a career, a way of seeing. I have knelt in enough churches and stood in enough forests to believe that the curiosity behind all of this is not a policy line item but something closer to the human spirit itself — the same impulse that drew a boy to draw craters on the moon, the same one that drew me to read a landscape as a text. No administration plants or uproots that. It only decides, in its season, whether to keep the nurseries open.

I believe in science because I am, myself, a thing it grew. The only question worth arguing is whether we keep planting. I know which side of that I am on. I have a shovel, I have used it, and I am not done.

— Samuel R. Lammie

A note on how this was made. I wrote this essay in collaboration with Claude, the AI system built by Anthropic. The life in it is mine — the wolf packs, the nurseries, McHarg read early, the twenty-five million acres — and so is the judgment, and the decision to stand behind every word. Claude drafted and redrafted at my direction, ran down the current figures on the funding cuts and handed them to me to verify, and pushed back when I needed pushing back. When I told it my early forestry was really analog GIS in McHarg's overlay tradition, it took the correction and made the passage truer; when I asked it to drop a pine I had come to regret planting in Guatemala, it did. That is the arrangement I would want for this tool wherever it touches serious work: the human as witness and final judge, the machine as a tireless second pair of hands, and nothing consequential taken on faith. An essay in defense of human curiosity, made this way, is no contradiction. It is the disciplined version of the thing — the only version worth keeping. 

 

The Carrier and the Signal: Antimony, the South Fork Salmon, and a Mine Four Presidents Built

There is a place in the Salmon River Mountains of central Idaho called Stibnite. The name is not decoration. Stibnite is the mineral form of antimony sulfide, and the ground was named for what men pulled out of it — antimony for the Second World War, tungsten for Korea, and gold whenever gold was worth chasing. The district sits forty-five air miles east of McCall, hard against the Frank Church–River of No Return Wilderness, in the headwaters of the South Fork Salmon River and within the homelands of the Nez Perce. By the Forest Service’s own reckoning, the South Fork holds the most important remaining habitat for summer Chinook salmon in the entire Columbia River basin.

I spent a career reading landscapes like this one — as a geospatial program manager for the Forest Service in the Northern Region, and before that as a line officer with a district ranger’s authority over a couple hundred thousand acres of my own. So I want to be careful here, because the Stibnite Gold Project is the kind of decision that punishes anyone who reaches for it with only one hand. It is not a simple story, and the people telling it simply — in either direction — are the ones I trust least.

The signal being broadcast

The case for the mine arrives wrapped in the language of national security, and that language is not fabricated. Antimony is a genuine strategic vulnerability. It hardens the lead in bullets and armor-piercing rounds; it sits in primers and tracer compounds, in night-vision optics and the infrared seekers that steer missiles, in flame retardants for military electronics. The United States mines effectively none of it and, in 2024, imported around 85 percent of what it consumed. China accounts for roughly 60 percent of world mine production and a far larger share of refining — the step that actually turns ore into usable metal.

Then Beijing closed the valve. In December 2024 China announced a full ban, “in principle,” on antimony exports to the United States, alongside gallium and germanium, in retaliation for American semiconductor controls. Prices did what prices do when a dominant supplier walks off: antimony reached an all-time high near $59,750 a tonne in July 2025, several times its level a year earlier. The National Defense Stockpile, the emergency reserve that exists for exactly this contingency, had been drawn down to what the Defense Logistics Agency called its lowest levels since the Cold War. In September 2025 the DLA signed a $245 million sole-source contract with a domestic producer to begin refilling it.

That is a real signal. If you are responsible for the supply chain behind the world’s largest arms exporter, a single foreign government holding the switch on a metal the defense industry says runs through some two hundred munitions types is a problem you are obligated to take seriously. I do take it seriously. My confidence that antimony dependence is a legitimate strategic exposure is high, and it is anchored on hard trade and production data, not on anyone’s press release.

The mine four presidents built

Here is where the partisan cartoon falls apart, and where I think the honest story actually lives.

Stibnite is not a Trump project or a Biden project. It is a four-administration project, and each one added a coat.

The application landed under Barack Obama: Midas Gold, Perpetua’s predecessor, submitted its plan of operations in 2016. The formal environmental review began under the first Trump administration, which published the notice of intent in 2017 and issued the draft Environmental Impact Statement in 2020. The Biden administration carried it the longest distance — a supplemental draft in 2022, the final EIS and draft decision in September 2024, and then the final Record of Decision, signed January 3, 2025 and released January 6. That is the detail almost everyone gets wrong. A January 2025 mine approval reads, to most people, as a last-minute Trump giveaway. It was not. It was signed by two Forest Service supervisors seventeen days before the inauguration. The green light was Biden’s.

So was most of the money. The Defense Department’s support for Perpetua — now more than $80 million, structured under the Defense Production Act — began in 2023 and 2024, under Biden’s Pentagon. The Export-Import Bank’s first offer of financing, a letter of interest for up to $1.8 billion, came in April 2024, also under Biden.

The second Trump administration then did what it has been open about doing: it accelerated. A March 20, 2025 executive order on mineral production named Stibnite among the first priority projects and placed it on the federal permitting dashboard as a “transparency project.” In May 2026 the Export-Import Bank approved a $2.9 billion loan for the build. The Justice Department is now defending the Biden-era decision in court, and in June 2026 a federal judge in Idaho declined to halt construction while the lawsuit proceeds. Ground was broken in October 2025.

Read the arc plainly and the through-line is unmistakable. The one thing Obama, Trump, Biden, and Trump again all agreed on was the antimony logic. That is the carrier wave — a national-security consensus stable enough to survive three transfers of power. It is worth pausing on, because genuine bipartisan agreement in this decade is rare enough to be evidence of something.

The payload riding the carrier

And yet. A carrier wave is not the message. The thing modulated onto it is.

By the accounting in the conservation groups’ own filing — a figure I hold as their claim rather than as audited fact — gold is about 96 percent of the project’s profit and antimony about 4 percent. The critics put the on-site antimony at roughly a three-year supply of national demand; Perpetua projects up to 35 percent of U.S. antimony over its first six years, a company projection that a competing domestic producer has publicly called inflated, arguing this is “a gold mine with a little antimony.” I cannot adjudicate the exact percentage from where I sit, and I will not pretend to. But the direction is not seriously in dispute: the revenue is gold, and antimony is the argument that makes the gold politically movable.

That is the part that asks for the older instinct in me — the one that reads signal and noise as a kind of moral topology. The national-security frame is doing real work, and it is also doing convenient work. When a metal becomes a headline, a whole apparatus assembles around the headline. Investment newsletters now circulate breathless pitches for junior exploration companies — I have read one working the Nevada ground hard — promising to be “the only pure-play U.S. antimony producer by 2027,” on deposits that as of mid-2026 have a maiden resource and no feasibility study, no reserves, no permit to build, and grades a fraction of a percent. That is noise riding the carrier, and it degrades the public’s ability to receive the true signal underneath. A citizen trying to weigh Stibnite honestly has to filter the strategic reality out from the promotional wash, and most citizens do not have the time.

Two more facts belong in the honest ledger, because they cut against the tidiest version of the security argument. First, the antimony crisis has already eased faster than the mine can answer it: China suspended its export ban in November 2025 as part of a trade truce, and by mid-2026 the price had fallen roughly 60 percent from its peak. The switch that justified the urgency has, for now, been flipped back partway on — by diplomacy, not by domestic supply. Second, under the 1872 Mining Law, a company can remove gold and antimony worth billions from public land and pay the American people no royalty at all. The public bears the watershed risk. The public does not share the gold.

What stewardship in the real looks like

I do not think this decision is easy, and I distrust the versions that do. The strategic exposure is real; I would put high confidence on that. The place is also real, and irreplaceable in a way antimony is not — you can refine antimony from a dozen deposits, but there is one South Fork Salmon, one run of summer Chinook, one homeland. What the record actually shows is a mine whose antimony is real but modest, whose gold is the engine, and whose national-security justification was strong enough to hold four administrations together even as the acute crisis it invoked began to pass.

I spent enough years inside the agency now sitting in the defendant’s chair to know that the Forest Service is not a villain here so much as an institution absorbing a decision that four presidents narrowed to a single question. The people who will run this project are, in the main, trying to do a hard thing carefully. The plan will disturb some 3,265 acres, move hundreds of millions of tons of rock, and re-plumb a river that already carries a century of mining’s damage — with, its defenders note, a genuine promise to clean up legacy pollution on the way. Whether the repair outruns the new harm is the whole question, and it will be answered in the water, over decades, long after the press releases are filed.

Duty, honor, country — I was raised on that phrase and I still mean it. But the eye has to lift past the nation to the world, and to the ground the nation stands on. A country that cannot make its own bullets is vulnerable. A country that trades its last best salmon river for a three-year hedge on a metal, and calls the gold underneath it a security measure, has confused the carrier for the message. The task, as always, is to hear the difference.

---

Sources: Center for Biological Diversity et al. press release (Feb. 18, 2025); Payette National Forest Final Record of Decision (Jan. 2025); Idaho Capital Sun; U.S. Forest Service, Region 4; USGS Mineral Commodity Summaries 2025; CSIS analysis of China’s antimony export restrictions; Perpetua Resources and Export-Import Bank announcements; Defense Logistics Agency / U.S. Antimony Corp. contract (Sept. 2025); U.S. District Court for the District of Idaho ruling (June 2026); Equedia/NevGold promotional letter (as cited).

Written with Claude (Anthropic). The frame and judgment are mine; Claude verified the timeline and figures against primary sources and helped keep documented fact distinct from contested claim. Any remaining errors are mine. 

The Friction Zone

Four inheritance systems, four rates, and the gap where nothing slow is watching

Samuel R. Lammie, with Claude (Anthropic)/ChatGPT (OpenAI)
The Musical Stone · Atlas of Persistence, Plate 000

---

Start with the distinction the whole thing turns on. Persistence is not permanence. Nothing that lasts does so by staying the same. A forest that persists for a thousand years is not a thousand-year-old stand of trees; it is a sequence of stands, burned and regrown, held together by something that survives each turnover. A language that persists across a millennium is not the same language. An institution that persists is emphatically not the same institution, whatever its letterhead insists.

What persists is not material. It is pattern, and pattern is process made visible.

Which raises the actual question. If everything that endures does so by changing, what distinguishes the changes that preserve identity from the changes that end it? That is the question an atlas of persistence exists to ask, and the first plate has to answer a narrower one before it can get there: what is the architecture? Through what does anything get carried forward at all?

Four inheritance systems, and why the fourth is not a subdivision of the third

Inheritance runs through four systems, and they are genuinely distinct rather than nested.

Biological inheritance is the deep one — genes, epigenetic marks, bodies carrying information across generations. It is slow, it reaches from the individual to the biosphere, and it is the only one of the four that predates us.

Cultural inheritance is language, skill, story, craft. It moves faster than biology by orders of magnitude, and it is Darwinian in structure but Lamarckian in character, because acquired traits are inherited in culture. A technique learned at forty passes forward intact.

Institutional inheritance is charter, record, precedent, doctrine. It replicates by document rather than by body, which gives an institution a time horizon no individual possesses. It is also the system most capable of holding a strategy coherent across a century.

Technological inheritance is the one that most people fold into culture, and shouldn't. The distinction that makes it a separate system is the carrier. Culture requires a living mind to hold it — break the chain of practitioners and the knowledge is gone, whatever the archive says. A technology does not. An artifact transmits without a carrier, sits dormant for a century, and resumes. That difference in mechanism is why technological inheritance can jump gaps that cultural inheritance cannot, and it is why the fourth system deserves its own row.

Rate and extent are one operator seen on two axes

Here is where most systems diagrams go wrong, and where this one earns its keep.

Four inheritance systems, four rates — that is the natural way to draw it, and it is where most diagrams stop. It is also not sufficient, because rate alone does not tell you what a system can reach. Fire and climate both change forests, at wildly different speeds and wildly different extents, and the interesting cases are the ones where those two axes disagree.

Every pattern depends upon scale. A system stable at one scale and unstable at another is the ordinary case, not the exception. A stand can be lost while the landscape persists. A landscape can be lost while the species persists. Fire regimes read as catastrophic at stand scale and as corrective at landscape scale — the same event, two verdicts, and only the second one is right about what the system is doing.

So the honest picture is not four bars on a timeline. It is a field. Rate on one axis, organizational extent on the other, each system occupying a region rather than a line. Draw it that way and the four stop being a ranked list and start being a map with overlaps — and the overlaps are where coupling between them is possible at all.

The gap on the left

Draw the field properly and something appears that no list of four rates can show you.

There is a region on the fast end of the rate axis where technology operates across every extent — individual to biosphere — and nothing else is present. Culture does not reach that fast. Institutions certainly do not; they run at document rate, which is the whole reason governance arrives after the fact. Biology is not even in the neighborhood.

 The four inheritance systems placed by rate and extent. Where the fast end runs empty, nothing with a longer memory is present to correct what happens there. (Source: Claude, ChatGPT, and the Author)

 That is the friction zone, and it is not a metaphor. It is a structural feature of the architecture. Technology acts there at a rate the slower systems cannot answer, which means that in that region there is no correcting feedback from anything with a longer memory. Whatever happens there happens uncorrected.

This is the same finding that turns up wherever you look at AI governance seriously. Every major instrument — ISO 42001, the NIST framework, the EU's Act — operates at audit-cycle rate against systems running at inference rate. None of them holds a refusal at the moment of action. The document-rate/operational-rate gap is not a flaw in any particular regulation. It is the friction zone, seen from inside one industry.

Which is the argument for reading the plate as a whole rather than as decoration. The gap is general. The instances are specific.

The ratchet, and how it slips

Of the six dynamics that recur across every adaptive system — generate, select, remember, coordinate, adapt, reorganize — one carries more load than the rest, and it is not a close call.

Remember is the mechanism that makes cumulative anything possible. Without retention, nothing compounds. Variation is generated and lost, selection acts and its result is not held, each generation starts where the last one started. A species like ours runs almost entirely on that mechanism: we are unremarkable animals by any physical measure, and what we have instead is the ability to hold gains across time so that each cohort begins where the last one stopped.

The error is drawing that ratchet as monotonic. It is not. Retention is maintenance, not deposit. Cut the connectivity or shrink the transmitting population and complex technique disappears — the accumulated stock is a standing achievement requiring continuous upkeep, not a balance that stays banked.

Which gives the shape of the real threat to anything that runs on cumulative transmission. You do not have to beat the organism. You corrupt the transmission. That is cheaper by orders of magnitude, and it works on any system whose survival depends on what gets encoded and passed forward.

The control group

An atlas that shows only how systems endure is a survivorship sample, and it will teach you very little. The plate that matters includes the anatomy of failure: what it looks like when each dynamic goes.

Generation fails and you get monoculture — nothing left to select from when conditions move. Selection fails and you get drift, accumulation without filtering, volume mistaken for progress. Coordination fails and you get fragmentation, or capture, cheap signals driving out honest ones. Adaptation fails and you get rigidity, error detected and not acted on, feedback running open-loop. Reorganization fails and you get brittle collapse, a threshold crossed with no successor structure standing ready.

And Remember fails and the ratchet slips. That is the fatal one, because every other failure is recoverable by a system that still knows what it knew.

Blank spaces

Two panels on this plate are there to say what it cannot do, and they are the ones I would defend hardest.

The first declares a blind region. Detection competence and influence competence run inversely: sufficiently skilled external influence in any adaptive system is indistinguishable from the system's own dynamics, by design rather than by accident. Which means a framework like this one can find only unskilled influence. The corollary has to be stated in the same breath or it becomes a license to believe anything — absence of signal is not evidence of the skilled kind. It is not evidence.

The second names, in advance, what would force revision. A system persisting through regime change with a demonstrably absent dynamic. Matched rates producing instability, or mismatched rates producing durable stability. A fifth inheritance system not reducible to the four. Persistence with no retention mechanism at all.

That panel is the difference between a framework and a mandala. It is tempting to describe a durable framework as one that stays useful because it can absorb revision — but that is a disposition, and an unfalsifiable structure can claim it in perfect sincerity. The version with teeth is narrower: a durable framework names in advance what would revise it. Every atlas should have blank spaces, and it should mark them.

Good ancestors

The question underneath all of it is not analytical.

The four systems are mechanisms. What they carry is something else — the standing accumulation, the biosphere and the languages and the institutions and the infrastructure as we find them. Those are the landscapes, and we inherit all four. None of them we made, and all of them we are currently modifying at a rate the slow systems cannot answer.

The stewardship question is not whether they change. They will. It is whether they change in ways that preserve their identity, and whether the transmission we hand forward is one our successors can trust.

Ne parcas nec spernas. Neither spare nor scorn — and read as intellectual discipline rather than heraldry: neither shelter your favored hypothesis from criticism, nor dismiss unfamiliar evidence without examining it. That is a good motto for an atlas. It is a better one for an ancestor.

---

Disclosure: this piece was developed in collaboration with Claude, a model built by Anthropic, and with OpenAI's ChatGPT. Both are products of frontier AI laboratories with direct commercial and reputational stakes in how questions about inheritance, information infrastructure, and the governance of fast-moving technology are framed. An argument identifying a "friction zone" where technology outruns institutional correction is not a neutral instrument when co-written by the technology in question. Read it accordingly.

Available

The Musical Stone — 18 July 2026

There was a word for it, and the word was available.

It meant your red card was current, your pack test was passed, your quals were signed off, and when the call came your regular job stopped being your job. It did not mean you worked for fire. Most of us didn't. It meant that when the system reached a certain pitch, the question of what you did for a living was suspended, and the only question left was whether you could roll.

This morning the National Interagency Fire Center moved the country to Preparedness Level 5. Seventy-three uncontained large fires. Twenty-three complex incident management teams committed. Seventeen thousand six hundred and eighty-three people assigned. Standard operating procedure at that level is to make all Forest Service staff available.

Missoula, 2003. Crews and gear coming off a charter, a rented school bus out of Plains, and the smoke thick enough to swallow the far end of the field. Photograph by the author.

I want to say plainly what that sentence used to mean, because I am not certain it still means it, and the difference is the whole story.The Forest Service's fire capacity was never the fire organization. That is the thing the public never understood and the thing the agency itself sometimes forgot in the quiet years. The engines and the hotshot crews and the helitack were the skeleton. The muscle was everybody else.

Go down the roster of any Type 1 or Type 2 team from the years when the system worked and count the permanent fire positions. There are not many. The Situation Unit Leader is a hydrologist. The Resource Unit Leader runs timber sales in October. The GIS Specialist spends the other ten months building layers for a forest plan revision. The Resource Advisor is the district archaeologist, and she is on that fire because she is the only person within two hundred miles who knows where the sites are and can tell a dozer boss, at two in the morning, not there. Finance is a budget officer. Plans is whoever can still think at hour sixteen.

That was the militia. It was not an auxiliary. It was the organization.

I was in it for most of a working life. Not as a fire officer — I never was one. I started as a forester at San Carlos, with the Bureau of Indian Affairs, burning under prescription and supporting wildfire. I moved into GIS with the Park Service in the early nineties. I finished as a GIS coordinator with the Forest Service. And whatever the job title said along the way, I worked fire data every year of my geospatial career.

That is worth saying plainly, because it cuts against the tidy version of the militia story. Fire was not the thing that interrupted my work. Fire was in my work the whole time — in the fuels layers and the perimeter files and the burn units, in the ten-year fire history somebody wanted mapped for a NEPA document. It was in the burn plans and the wildfire monitoring protocols. And in the early years it was in the GISS standards and templates — the same body of standards that lives today as the Standards for Geospatial Operations.

I want to be careful here, because there is a romantic version of this that is simply false. Those standards were never folk knowledge passed hand to hand between old-timers. They are owned. The National Wildfire Coordinating Group holds them — the Standards for Geospatial Operations, PMS 936 — maintained by its Geospatial Subcommittee with a charter, a steward, and a one-year revision cycle. There is an office. It has a name and a publication number.

What there is not, and never was, is a workforce. NWCG's own membership rules state the exposure in plain language: members are appointed through agency command channels with full understanding of the collateral duty nature of the work and its potential workload impacts. Each member agency decides for itself what it can spare. Usually one primary member per committee.

So the standards that let a GISS step off a plane in a state he has never worked and be useful before lunch — the symbology, the file naming, the map product specifications, the directory structure that means the incoming shift can find what the outgoing shift built — are maintained by people doing it on top of their actual jobs, contributed by agencies that decide each year whether they can still afford to contribute them.

That arrangement holds beautifully right up until the agencies get thin. Nobody votes to abolish a standard. The seats simply get harder to fill, and the people who fill them arrive with less behind them, and the revision cycle slips, and one day the document that made a stranger useful on arrival is four years out of date and nobody can say quite when that happened.

What the dispatch interrupted was only the part of the job that could wait. The phone would go, and then you were on a team somewhere you'd never been, working for people whose names you learned at the morning briefing, doing in one night what would take a month at the district. I went out both ways, as a firefighter and as a GISS. Same man, same red card, whichever the order called for.

And that is the part that gets missed when people talk about the militia as though it were a pool of willing amateurs. It was not. It was people who had been standing next to the fire program their entire careers, close enough to be fluent in it, and who could therefore be useful the day they stepped off the plane.

It was also, and this matters more, people who had qualified. Nobody wandered onto an incident because they were available and willing. You went through the hoops. That meant a task book, and a trainee assignment, and a qualified evaluator prepared to sign that he had watched you do the work, and then another assignment, and currency requirements after that or the card went cold in your wallet. The training was extensive, it was not optional, and it was the same system the career fire people went through. A hydrologist carrying a Situation Unit Leader card had walked an identical path to anyone else carrying that card. That is precisely what made the arrangement work — the red card meant the same thing regardless of whose payroll you were on. Interoperability was not a slogan. It was a qualification system that a stranger's signature could be trusted inside of.

Which is where the arithmetic turns against us, and I want to be plain about it, because it is the part nobody says out loud.

Trainees are produced by assignments. You cannot qualify in an office. Somebody has to take you out, carry you while you are still slow, and sign the book at the end. That costs the team something on every single incident, because a trainee is by definition not yet the person you needed. In a hard year, when every team is short and every order is urgent, the trainee slot is the first thing to go. Cutting it is always defensible. It is defensible this morning, on seventy-three uncontained large fires with twenty-three teams committed.

So the seasons that most demand a deep bench are exactly the seasons in which the bench stops being built. Run that for a few years, add a reorganization and a workforce reduction on top, and you do not get a collapse anybody can photograph. You get a slow one, arriving five and ten years out as positions nobody can fill — and no single decision anywhere along the way that you could point to and call the mistake.

That is the part that does not appear in a staffing table. The capacity was never a number in a resource database. It was a few thousand people scattered across specialties who had made the same private arrangement with the agency — that when it got bad enough, what they did for a living stopped, and something older took over.

A Montana fire camp at last light, 2003. Every tent is somebody whose regular job stopped. Photograph by the author.

Both of those photographs are from 2003, which was also a Preparedness Level 5 year. That summer more than five hundred thousand acres burned in Montana along the crest of the Rockies, and at the peak of it there were about fourteen thousand people working fires in this one geographic area — a region that cycled through eighty-four incident management teams before the season ended.

Fourteen thousand, in the Northern Rockies.

This morning the entire United States has seventeen thousand six hundred and eighty-three, and the Northern Rockies has four hundred and fourteen. The seasons are not the same and the calendar is not the same — 2003's peak came in late August, and today is the eighteenth of July with the worst of it still in front of us. I am not claiming the arithmetic proves a decline. But I took those pictures standing in a field with what a full mobilization looked like when the militia still turned out, and I would like somebody to tell me we could still fill that field.

Four things have happened to that arrangement.

The first began long before anyone proposed a new agency, and it explains most of the rest. The Forest Service budget changed shape. In 1995, fire was sixteen percent of the agency's appropriated budget. By 2015 it was fifty-two percent — the first time in a hundred and ten years that more than half of what Congress gave the Forest Service went to wildfire, with a projection of sixty-seven percent by 2025. National Forest System funding, which is where most of the agency's actual work lives, fell from fifty-eight percent of appropriations to twenty-nine.

Now set the staffing beside it. Non-fire personnel went from roughly eighteen thousand in 1998 to fewer than eleven thousand in 2015 — a thirty-nine percent cut. Over the same years, fire staffing more than doubled.

Read those two sentences together and you have the whole of it. Those seven thousand people were the militia. Not the fire organization — the militia. The hydrologists and the archaeologists and the surveyors and the GIS people, whose salaries came out of non-fire lines, who carried red cards, who staffed the teams. The fire program did not simply take budget share from them. It consumed the payroll that produced its own surge capacity.

I do not believe that was intended, and I am not certain it was ever fully noticed. Fire grew. The bench fire ran on shrank. Both came out of the same appropriation. You cannot double the fire organization by spending the militia and finish with more capacity than you started with. You finish with a larger skeleton and less muscle.

The second is that fire has now been separated from the agency that fed it. There is a U.S. Wildland Fire Service, under Interior, and there is still a Forest Service, under Agriculture, and the non-fire workforce that historically backfilled the national system at PL5 now sits on the far side of a departmental line from the incidents it would fill. On paper the qualification records are interagency and nothing has changed. In practice an internal mobilization has become a transaction. Release authority rests with supervisors whose delegations and incentives are not what they were. Cost coding crosses a department. And the thing that actually moved people — the sense that this was ours, that the fire on the next forest was a family emergency — got quieter the moment fire stopped being ours.

The third is more arithmetic, on top of the first. The recent reductions did not fall evenly; they took hardest from the mid-career specialists, which is exactly the band where the quals and the currency live. A GS-11 with fifteen years and a Division Supervisor card is not replaced by two seasonal hires. The pool is smaller again and the average depth inside it shallower again, and neither of those facts appears in any staffing table as a fire number.

The fourth is that the transition is unfinished, and we already know what the seam looks like, because we have seen it.

On the night of 27 June, on the Knowles Fire in Mesa County, a wind-driven run overtook a Rifle Helitack crew conducting initial attack. Five deployed shelters. Emily Barker, thirty-eight. Nick Hutcherson, twenty-seven. Sydney Watson, twenty-seven — all three died at the scene. Two others were carried out with burns; one of them, Nathan Matthews, forty-three, died of his injuries weeks later.

Barker and Hutcherson were badged to the Forest Service. Watson was badged to the Wildland Fire Service. One crew. Two agencies. That seam existed at the crew level three weeks ago. At Preparedness Level 5 it exists at every ordering channel, every dispatch, every delegation, simultaneously.

The serious accident investigation is still open. I have said before and I will keep saying that the useful question is not what those five people did on that slope. It is what was assumed on their behalf by the structure that put them there — whether the escape route and the contingency were facts or hopes, and who was in a position to know the difference. That question does not get easier to answer inside a PL5. It gets harder, because the qualified people who staff serious accident investigations are the same people the system is currently consuming.

Here is what I will be watching, and it is a narrower thing than acres burned.

Not engines. Not crews. Overhead. Section chiefs, unit leaders, GISS, resource advisors, finance. If equipment continues to fill while overhead orders go unable-to-fill, that is the militia not answering, and it will appear in the daily situation report before anyone says it out loud in a press conference. That is the number that tells you whether the muscle is still attached to the skeleton.

And one more thing, which is not a fire matter at all.

The mobilization standard is explicit that preparedness levels govern wildland fire, prescribed fire, all-hazard response, and international assistance. Those twenty-three teams committed to Oregon this morning are the same teams this country sends to hurricanes, to floods, to pipeline failures, to the recovery of things nobody wants to describe. It is a national bench, not a fire bench. It has always been quietly available for whatever came, because the wildland fire community built the only scalable incident management capacity the United States has.

I know that because I was on it. In January of 1997 I went to the Yosemite flood as a cartographic technician out of Voyageurs — northern Minnesota to the Sierra Nevada, because the Merced had run twenty-four thousand cubic feet a second through the valley, the highest in eighty years of record, and somebody needed maps of what was left.

Yosemite, January 1997. The Merced took the road out from under its own centerline. Photograph by the author.

In 2003 I went to Montana as a firefighter, which is where these photographs come from, the tarmac and the tents both. In 2005 I went to Katrina as the GIS coordinator off the Monongahela. Eight years, three agencies, three entirely different kinds of catastrophe, and the same red card in my wallet the whole way through. Nobody hired me to do any of it. It was simply understood that a mapmaker who could also hold a line was worth having on the list, and that the list was how the country got through things.

That is what is being tested this week.

It is spoken for. Atlantic season is opening. A magnitude 7.3 is still ringing off the coast of Chiapas with a swarm behind it. And the forecast beneath a strengthening ridge over the Rockies calls for dry lightning across the interior West into cured fuels, which is to say the system is about to be handed new starts it has no depth left to catch. It will get worse before it gets better. Whatever else happens to this country between now and September happens to an organization that has already filed the paperwork saying it has nothing in reserve.

I do not think this fails. The people are still good, and the system has absorbed worse than a reorganization. A wildland fire service under Interior is a defensible idea and it will probably work, given time enough to grow back what a transition costs.

What I am saying is narrower. The bumps in this line of work are measured in people. The bench that used to absorb them was built over decades out of an arrangement nobody ever wrote down — and it is being rebuilt now, in the middle of a Preparedness Level 5, by an organization that cannot pause the season to do it.

Available was a status. I'd like to know if it's still a promise.

 

Samuel R. Lammie began his career as a forester with the Bureau of Indian Affairs at San Carlos, where his work included prescribed fire and wildfire support. He moved into GIS with the National Park Service in the early 1990s and retired as a GIS coordinator with the USDA Forest Service, working fire data throughout his geospatial career. His work included burn plans, wildfire monitoring protocols, and, in the early years, GISS standards and templates. His fire experience spans three agencies — BIA at San Carlos, the National Park Service at Voyageurs, and the Forest Service on the Monongahela — and he was red-carded and dispatched both as a firefighter and in GIS. Assignments included the 1997 Yosemite flood, the 2003 Montana fire season, Hurricane Katrina in 2005, and wildfire incidents across the West. The photographs are his own, made on a 2003 fire dispatch to Montana.

This piece was drafted in conversation with Claude (Anthropic), which supplied research, current fire figures, and prose. The argument is the author's: that the agency's real surge capacity was always the militia rather than the fire organization, that consolidation and workforce reduction put that capacity in question, and that a Preparedness Level 5 commits the nation's all-hazard bench, not merely its fire bench. The AI had the preparedness level wrong at the outset and was corrected. Fire figures are from the NICC Incident Management Situation Report of 18 July 2026. Budget and staffing figures are from the Forest Service report The Rising Cost of Wildfire Operations: Effects on the Forest Service's Non-Fire Work (2015). Standards references are to the NWCG Standards for Geospatial Operations (PMS 936). Northern Rockies 2003 figures are from contemporaneous reporting (Daily Inter Lake) and NASA Earth Observatory.

 

A note on corrections (July 31, 2026). An earlier version of this piece stated that the National Wildfire Coordinating Group archived its Geospatial Subcommittee on July 1, 2026. That is incorrect — the subcommittee remains active — and the sentence has been removed. The same version slightly overstated the 2003 Montana fire acreage, now corrected to more than five hundred thousand acres, and predated the death of a fourth Knowles firefighter, Nathan Matthews of Lincoln, Nebraska, who died July 24 of injuries from the June 27 burnover; the text now reflects it. These drafts were prepared with AI assistance. The errors were mine to catch, and the corrections are logged here rather than made silently — which is, after all, the argument the piece is making.