Research note
Baltic Birch
Baltic Birch
A working description of a design language, a craft position, and a way of building things that hold.
I. The name, and why it was always going to be load-bearing
The name began as a joke, in the sense that all good names begin as jokes — said lightly, with no expectation that anyone would take them seriously, including the person saying them. Baltic Birch. A sheet good. A specific grade of plywood, fifteen millimeters thick by preference, with thirteen plies, often roughly graded B/BB on its faces, manufactured under a Soviet industrial regime whose successor economies still produce it for export, and used by a particular kind of hobbyist and small-shop woodworker for a particular kind of clean, modular, modestly-scaled work — bookshelves, kid furniture, jigs, cabinets that need to be built once and used for a long time.
The name arrived as a casual association during a late conversation about web design grammars. The author of the project — Mikael, a programmer with a serious sideline in handmade furniture for toddlers and small children, working primarily in Baltic birch with pine accents — was musing aloud about wanting a design system that would take its bones from his woodworking practice. The conversation rapidly outgrew the casualness of the framing. Within an hour or two, the name had stopped being a joke. It had become a position. The association turned out to be load-bearing in a way the speaker had not initially realized, which is what good names do.
The reason the name took weight is that Baltic birch as a material has the specific property of being standardized and alive at once. It is industrial: machine-graded, dimensionally consistent, rectilinear by default, built up from layers of rotary-cut veneer glued under heat and pressure into a sheet whose thickness is reliable to within a fraction of a millimeter. It is also organic: the plies show their grain, the edge reveals the stack of laminations as a striped figure that no industrial process can fully tame, the surface accepts oil and stain and develops patina, the material yellows over decades and acquires character that the original sheet did not possess. It is, in short, both a substrate and a material, both a constraint and a presence — the kind of stock from which you can build a cabinet that will be used by your grandchildren, and which will look better, not worse, when they use it.
The design language that takes its name from this material is an attempt to articulate, for the medium of software interfaces and adjacent practices, the same combination of properties: industrial discipline, organic warmth, modular standardization, visible construction, eased outer surfaces, crisp inner joints, and a commitment to building things that hold their bearing under use over long periods of time. The name is metaphor, but it is metaphor that pays its debts: every commitment of the design language can be derived, at need, from the corresponding commitment of the cabinetmaker working with the actual sheet stock. The metaphor is load-bearing, in roughly the same way that the material it names is load-bearing.
II. Against the slippery world
The most concise statement of what Baltic Birch is for comes from a passage in Moby-Dick that one of the conversation’s participants surfaced about three quarters of the way through the discussion. Ahab, in the carpenter’s shop late in the novel, watches the carpenter at work, takes hold of the bench vise, screws it down on his own flesh, and exclaims:
“I like to feel something in this slippery world that can hold, man.”
The line is half-comic and wholly serious. It names a hunger that runs underneath all of the design talk — the hunger for things that grip back when you reach for them, for surfaces that admit the hand, for systems that resist when you push and yield only when you mean them to. The slippery world Ahab is complaining about is not a natural condition; it is a built condition, produced by environments and tools and conventions that do not provide grip. The cabinetmaker’s vise, by contrast, does provide grip — that is what it is for — and Ahab’s gesture is, among other things, an oblique tribute to the discipline of building things that hold.
Baltic Birch, as a design position, takes Ahab’s complaint as its starting condition. The contemporary built environment of software — the dominant register of mid-2020s product design, which has now sedimented through enough years and enough Figma templates that it can fairly be called a style — is slippery in exactly Ahab’s sense. Cards float on atmospheric drop shadows that suggest depth without committing to it. Borders are minimized into 1px hairlines that ask to be seen as little as possible, on the assumption that visual quietness is always a virtue. Padding is generous to the point of waste, on the assumption that “breathing room” is always desirable, regardless of what is breathing or what would suffocate without the space. Corners are rounded with a uniform radius, set globally as a design token, applied without regard to whether the corner in question is meeting empty space or another structural member. Animations gracefully ease everything into and out of existence, masking the discrete state changes that, properly visible, would let the user understand what the system is doing. The whole stylistic register is one of atmospheric softness, and its cumulative effect is to leave the user without anything specific to grip.
This is not, the position holds, a stylistic preference to be argued with on stylistic grounds. It is a structural failure of the dominant register to provide the user with what they actually need from the surfaces of their working environments, which is grip. Grip is provided by edges. Edges are made visible by contrast, by thickness, by seating, by declared boundaries between regions. The dominant register has systematically eliminated all of these in pursuit of atmospheric quietness, and the cumulative cost is paid by every user of every interface designed under its conventions, every day, in the form of low-level cognitive fatigue, drift of attention, inability to sustain focus, and the specific feeling — not quite articulable but instantly recognizable — of being unable to find anywhere to land in the surface one is working on.
Baltic Birch begins with the wish to provide grip. From that wish, by working through what grip means at each scale of the interface, the rest of the design language is derived.
III. The cabinetmaker’s lessons, transposed
The cabinetmaker — and here we mean the tradition, not any particular person, though the tradition lives in particular hands — knows a number of things that have become unfamiliar to most contemporary software designers. The Baltic Birch position is, in large part, a transposition of these things into the medium of software, where they apply structurally even though the substrate is different.
Edges show construction. A border around a region is not a line drawn on the surface; it is the place where two members meet, where a cut was made, where a seam was joined, where the construction of the artifact becomes legible to anyone with eyes to read it. A cabinetmaker who has cut a clean dado does not hide the dado under filler; the dado is the joint, and the joint is the structure, and the structure is the thing the cabinet is made of. To hide it would be to misrepresent what the cabinet is. The same applies to interface regions: the seam between two panels is information about how the panels are related, and the design language should make that information visible rather than smoothing it into a gradient.
This generalizes into a vocabulary of edges that the design language has developed over the course of the conversation. A cut edge is a simple divider between adjacent panes. An exposed ply edge is a stronger boundary that reveals nesting or depth. A lipped edge is a header or title band attached to a pane. A rabbet edge is an inset region, like a list inside an inspector, that seats into a step cut in the surrounding panel. A finger edge is a draggable splitter or handle. A finished edge is the outer perimeter where the assembly meets the world. Each kind of edge does a different structural job. A design language that uses only one — the 1px hairline, undifferentiated, applied wherever a separator is needed — has impoverished its own vocabulary to the point of muteness. Baltic Birch keeps the full vocabulary alive.
Ease the outside, keep the inside sharp. This emerged in the conversation as one of the sharpest single principles of the position, and it is straightforwardly transposed from cabinetwork. When a cabinetmaker eases an edge — softens it with a sanding block or a shave or a router bit, just enough to take the sting off — the easing is applied to the outside of the piece, where the hand will rest, where the eye will fall, where the finished edge meets the world. The inside of the joint, where one member seats against another, is kept rigorously square, because square is what allows the joint to seat. A round-over bit run indiscriminately across every edge, including the shoulders of tenons and the bottoms of dados, would be the work of someone who had not understood what the bit was for. The bit is for the outside. The inside stays sharp.
The interface analogue is direct and consequential. The outermost corner of a card, where the card meets the empty surrounding canvas, can legitimately be eased, because that corner is meeting the world. The inner corners — where the card meets a header inside it, where two adjacent cards meet at a shared edge, where a panel seats into its frame — must stay crisp, because crispness is what allows the elements to seat against each other. The dominant convention of applying a single global border-radius to every corner of every element is, on the cabinetmaker’s standard, the equivalent of taking the round-over bit to every edge of the workpiece. It looks “soft” in the way Duplo blocks look soft. It is the visual signature of someone afraid of edges, who has decided to over-correct with cartoon roundness rather than to attend to which edges should remain sharp because they have structural work to do.
The discipline produces a specific principle: a corner’s treatment depends on what is on the other side of it. Empty space → ease it. Another member → keep it crisp. The radius is not a property of the element; it is a property of the relationship between the element and what it meets. Almost no design system encodes this. Baltic Birch insists on it.
Thick boundaries, with two edges. The cabinetmaker, like the architect, knows that a real boundary is not a line but a zone. A wall has thickness, and the thickness is inhabited — the niche, the window seat, the shelf, the picture rail. The frame of a door is thick, with an outer face that meets the room and an inner face that holds the panel; the space between is the frame proper, with its own structural members (mortise, tenon, mullion) and its own life. Christopher Alexander’s Pattern Language names this explicitly as Pattern 15, “Thick Boundaries,” and the corresponding fundamental property in The Nature of Order is “Boundaries” itself, treated as one of the load-bearing features of living structure.
Almost no contemporary interface region has a thick boundary in this sense. A typical card has a single 1px border — a thin membrane — and inside the border there is “padding,” treated as empty space, with all the region’s contents floating in the interior. The result is that the region has no frame zone; everything is either chrome (which is minimized) or content (which is the whole interior), with no inhabited boundary in between. Baltic Birch insists that every region of substantial weight should have a thick boundary zone between its outer and inner edges, populated with the region’s identifying labels, controls, status indicators, and structural members. The boundary becomes a place — a frame in the cabinet sense, or a window seat in the architectural sense — rather than a line.
The operational consequence is that “padding” and “border” stop being interchangeable whitespace tokens and become structural specifications. Padding is the wall’s thickness; it is sized to the containment load. A modal containing a critical decision needs a thicker frame than a sidebar item containing a navigation link. Border is the visible declaration of the edge; it is how the boundary announces itself to the eye. The two work together to produce a region that holds.
Joinery, not just adjacency. When two cabinet members meet, they are not merely placed next to each other; they are joined, in one of a small number of canonical ways, each of which has its own structural character and its own visible signature. The finger joint, the dovetail, the mortise and tenon, the rabbet, the half-lap, the spline, the through-tenon — these are not decorative choices; they are load specifications, each one appropriate to a particular kind of stress and a particular kind of relationship between the members. A cabinet whose members are merely butted together — flush surfaces glued and prayed over — is the weakest possible construction, and any cabinetmaker can identify it on sight.
The interface analogue is, again, almost entirely missing from the contemporary register. Most regions in a contemporary UI are butt-jointed: two flat surfaces abutted along a hairline, held together by visual glue (color, shadow) and the user’s willingness to imagine continuity. Baltic Birch insists on real joinery, and develops a vocabulary for it. A finger-jointed seam between two panels — alternating projections of one into the other, with elements of each region addressable from both sides. A through-tenon that originates in the toolbar and emerges in the content area as filter chips, the same object visible at both ends of its travel. A rabbet where a header sits into a step cut in its pane rather than floating above it. A drawbore where two panels are positionally cross-referenced through a real coupling rather than an event listener. These are not exotic; they are the basic vocabulary of constructing things from members. The fact that contemporary interfaces almost never use them is a sign of how far the medium has drifted from its construction-aware predecessors — BeOS, NeXTSTEP, classic Mac OS, and the longer architectural and cabinet-making traditions before them.
IV. McLuhan’s teeth, and why typography is not innocent
Halfway through the conversation that produced this position, the discussion turned to typography, and from there to a passage that turned out to be central. McLuhan, in Understanding Media and elsewhere, argues that the phonetic alphabet is not innocent — that its introduction into a culture retrains the cultural mind by drilling perception into segmented, linear, sequential, abstract patterns that the pre-literate mind did not necessarily share. The famous formulation invokes Cadmus, the mythical bringer of writing to Greece, who in the same myth sowed the dragon’s teeth and watched them spring up as a regiment of armed soldiers. Letters are teeth and armies. Each letter is a discrete, hard, biting unit; the sequence of letters is a regiment in formation; the alphabet as a whole is a standing army of standardized perceptual units, deployable in any combination, which the literate mind learns to take for granted but which is, considered freshly, a remarkable and somewhat alarming object.
This framing matters for Baltic Birch because it makes typography legible as a particular kind of structured wildness that the design language must reckon with explicitly. Type is not neutral content placed inside the frame; type is itself a configuration of edges — strokes, terminals, counters, baselines — arranged by long convention into a regimented field that the reader’s eye scans rapidly for meaning. A page of running text is, viewed as pure pattern, visually noisy in the extreme. We do not experience it as noisy because we have been trained from childhood to parse it, but the noise is there at the substrate level, and the typography of the page is what organizes the noise into legible passage.
In one of the most generative phrases of the conversation, the participant described type as a labyrinth of irregular scribbles, very unruly like a thorn bush. The framing is exactly right. Type is wild material, drilled into formation. It needs to be enclosed, the way a hedge is enclosed against a wall, in order for its wildness to become legible rather than chaotic. The page, the column, the baseline grid, the margin — these are the cabinetry around the thorn bush. Without them, the type does not have a shape; it just spills.
This produces a really striking picture of what good typography is for, transposed back to interface design: typography and frame are co-dependent . The frame holds the wildness so it can become legible; the wildness presses against the frame and gives it something to contain. Strong frame allows wild type. Weak frame requires tame type. There is a conservation law at work: somewhere in the rendered page there must be discipline, and the design choice is whether to put the discipline in the frame (allowing dense, alive, intricate typography to flourish within it) or in the type itself (sparse, even, tamed sans-serifs at heroic line-heights, because there is no frame to hold anything richer). Most contemporary web design has opted for the second strategy by default, removing the frame and consequently being forced to remove the typographic life as well. The page becomes a gentle desert. Baltic Birch chooses the first strategy: build the cabinet thick, then let the type be alive within it.
Type also turns out to interact with the easing-and-joinery vocabulary in productive ways. The edges of the text block — the vertical lines formed by the leftmost stems and the rightmost stems, the horizontal lines formed by the cap heights and baselines — are real edges, formed by the type itself, and they can do edge work that would otherwise be done by visible borders. A flush-left column of well-set text has a left edge as solid as a ruled line, and the design system can rely on that edge in placing adjacent elements. A heading set in heavier weight produces a horizontal edge at its baseline. A baseline grid produces a whole field of horizontal edges that can interlock columns even where no visible rules exist. Type is edge-bearing material, and the discipline of Baltic Birch is to register the typographic rhythm to the structural rhythm of the frame, so that the two reinforce rather than compete.
There is a corresponding cluster of failures that are worth naming, since they appear constantly in contemporary web design. Interference: type set too close to a frame edge of comparable weight, so that the eye cannot tell which is the structural edge and which is the textural one, and both edges suffer. Abandonment: type set in too much padding, with no relationship to the frame, so that the type floats meaninglessly in a heroic void and the frame has nothing to receive. Rhythm clash: a baseline grid in the type at one frequency and a structural grid in the frame at another, beating against each other unproductively. Each of these is the typographic equivalent of cross-grain joinery, and each is fixable through a discipline that the conversation has accumulated under the heading of register: the type’s grid and the frame’s grid should agree.
V. The substrate constraint, and why bounded media support craft
A theme that emerged repeatedly through the conversation, in different domains, is that craft is enabled by substrate constraints, not impeded by them. The cabinetmaker’s craft is possible because the sheet stock has standardized thicknesses, the joinery vocabulary is small and enumerable, and the geometry is predominantly rectilinear. Knuth–Plass paragraph layout is computationally tractable because the page width is fixed and the paragraph is bounded. Bernardy’s Pareto-optimal source-code pretty-printing is tractable because monospace fonts make width additive and the column boundary prunes the search space. Mathematical notation is fluent because the symbol set is small and short. Lisp’s pretty-printability is a function of its homogeneous tree structure. Forth’s expressive density is a function of its tiny primitive vocabulary. In every case, the substrate is constrained, and the constraint is what makes globally-optimal craft work possible rather than merely aspirational.
The parallel observation, which is harder to digest because it cuts against the contemporary instinct: unconstrained substrates produce greedy, locally-optimal output, which is always worse than what a constrained substrate can produce. The web’s continuously-resizing viewport is unconstrained in exactly the dimension that most matters for typography (width is a free variable, controlled by the user, varying continuously), and the result is that the web is structurally condemned to greedy paragraph layout in a way that print is not. Print can run Knuth–Plass once, batch-style, and amortize the cost over centuries of reading; the web cannot, because the cost would be paid on every reflow, and the user dragging the window edge is not willing to wait. The web’s greediness is not a failure of web designers; it is a substrate condition downstream of the choice to make width a free variable. The choice was probably correct overall — the web’s flexibility has enabled enormous things — but the cost in craft is real, and most of the difference in typographic quality between a printed book and a webpage of comparable content is downstream of this single substrate decision.
This generalizes into a principle that Baltic Birch insists on at the level of design: constraint is the condition for craft. A design system that allows any size, any spacing, any color, any radius, any font is one whose space of possible designs is too large to search globally; it is forced into local greedy decisions, and the cumulative result is that nothing in the system has been globally considered. A design system with deliberately constrained substrate — a small number of allowed thicknesses, an enumerated set of joinery types, a baseline grid that all members register to, a small palette of finish tones — is one whose space is small enough to be checked, and the work within it can be globally optimized. The constraint is not restrictive; it is enabling. It is the same move as Bernardy’s column boundary, Knuth’s fixed page width, Forth’s tiny primitive set, Lisp’s homogeneous tree, mathematics’ compact notation. The cabinetmaker who works in three thicknesses of stock and a dozen canonical joints is not impoverished; they are equipped, with a vocabulary small enough to think with.
Baltic Birch, accordingly, is a design language with deliberately bounded substrate. A small number of ply orders (ranging from hairline to major workspace container, but enumerable). A small number of joinery types. A small number of edge treatments. A small number of finish tones. A small number of typographic scales registered to a single baseline grid. The constraint is not concession; it is the precondition for the craft.
VI. Time, also, has grain
Most of what has been said so far concerns space — how regions meet other regions, how edges are treated, how typography sits in its frame. But time turns out to require the same disciplines, applied along a different axis, and the conversation converged on this through several routes.
The user’s attention is not a continuous unbroken resource that the system can spend at will; it has grain, rhythm, characteristic batch sizes. Cognition is paragraphic — it accumulates in chunks, releases between chunks, regroups, and proceeds. A system that respects this rhythm by yielding to the user at the right grain is one the user can sustain attention in over long periods. A system that does not yield — that goes off to do unbounded work without surfacing progress, that locks the interface during long operations, that delivers walls of output after long silences — is the temporal equivalent of the slippery surface. The user reaches for somewhere to put their attention and finds nothing. They are displaced from agency until the system returns. This is, structurally, the same complaint Ahab made against the slippery world; only the dimension is different.
The Baltic Birch position takes this seriously enough to specify it. Computations are coroutines, not commissions. Every long-running operation yields at semantic boundaries, with visible progress, with cancellation as a first-class mode of cooperation. Long operations are productive streams whose every intermediate state is itself a usable product the consumer can act on, rather than waste-while-waiting-for-the-final-result. The yields are the inner edges of the temporal cabinet — clean seats where one step meets the next, places the user can re-grip without losing their bearing. Background work is always inspectable, never invisibly proceeding on the user’s behalf without a way to see where it is.
This applies, with particular sharpness, to the contemporary class of AI agents. An agent that is given a task, goes off to do it, and returns with a result is the temporal equivalent of a flat-design dashboard with no edges anywhere — there is nothing for the user to grip while the agent is at work, no opportunity for redirection, no chance to verify intermediate steps, no defense against the agent doing something unrecoverable. A Baltic Birch agent yields constantly: after every tool call, after every reasoning step, after every commitment. The user is never out of grip. The interaction feels slower in raw throughput and enormously more habitable for the long arc of attention. The agent is a coroutine, not a commission. This is a real position, and it distinguishes Baltic Birch sharply from most contemporary “agentic” software discourse.
The temporal joinery doctrine also produces an operational test that pairs cleanly with the spatial one. The spatial test asks: can this region absorb strange contents without losing its bearing? The temporal test asks: can this operation be interrupted at any moment without losing its bearing? Both must pass. Spatial wildness — long text, weird unicode, unusual aspect ratios — must not break the cabinet. Temporal wildness — cancellation, network blips, the user changing their mind — must not break the cabinet either. The cabinet must hold against both wild contents and unexpected interruptions, because both are the substrate of real use.
VII. The deeper claim: phenomenal grain is real
The most ambitious thread of the conversation, which surfaced most sharply when the discussion turned to QRI-flavored phenomenology and to psychedelic reports as natural experiments on visual structure, is the claim that visual fields have grain in roughly the way wood has grain — that the substrate of perceptual experience is a real material with structural properties, that those properties matter for what happens in consciousness when attention engages a surface, and that good design is at least partly a discipline of providing legitimate phenomenal structure that the perceptual system can bind to without having to manufacture it from nothing.
The argument runs roughly as follows. The everyday visual field is heavily compressed by predictive priors; we do not perceive the surfaces around us in their full structural detail because the system is summarizing aggressively for efficiency. Conditions that loosen the priors — fatigue, dissociation, sustained close attention, certain pharmacological states — let perceptual energy flood into the surface, and what the surface does in response is enormously diagnostic of what kind of surface it actually is. Smooth surfaces destabilize under loosened priors: blank walls breathe and warp, plain ceilings melt, glassy minimalism reveals itself as a void papered over with predictive suppression. Grained, structured surfaces come alive under the same conditions: wood reveals deeper figure, fabric reveals weave, hand-decorated surfaces become radiant. The first kind of surface depended on prior compression to feel coherent; the second has legitimate organization the perceptual system can ride deeper into.
If this picture is even approximately right, it has a clear consequence for design. Most contemporary flat-design UI is smooth-surface UI in this sense. It depends on prior compression to feel coherent. Under any condition that loosens the priors — and “long sustained use” is one such condition, alongside more dramatic ones — its coherence degrades, and the user begins to feel the specific exhaustion of being in a surface with nothing to land on. Whereas a Baltic Birch interface, with real joinery and visible structure and edges that have been attended to, would be grained-surface UI, with legitimate organization at every scale that attention can engage with. Such an interface would reward sustained attention rather than punishing it. The grain would deepen under closer looking.
This reframes the cost-benefit calculation of craft. The standard objection to spending design effort on visible structure is that “users won’t notice.” Under the phenomenal-grain hypothesis, this is partly correct and entirely beside the point: users will not consciously notice the joinery, but their perceptual systems will respond to it, and the response will show up as longer sustainable attention, lower cognitive fatigue, less drift, fewer aborted sessions, and the specific feeling of being able to inhabit the work rather than merely passing through it. The benefit is real; it is paid out in a register most A/B tests don’t measure; and over long horizons it is decisive.
This is also what underwrites the otherwise-suspicious claim that good design is, in some way, a small-scale practice of cultural perceptual training. Marshall McLuhan’s argument about the alphabet is that it slowly retrained literate cultures into linear-segmented-sequential cognition; the analogous claim about contemporary flat UI is that it is slowly retraining a generation to expect surfaces not to grip. A generation whose daily perceptual diet consists of atmospheric-depth, hairline-bordered, soft-shadowed, animation-eased interfaces is being incrementally trained to find such surfaces normal and to find their alternatives strange. The training is real, even though its individual increments are small. Baltic Birch is, accordingly, a small counter-training: an offer of surfaces that grip, in the hope that the perceptual systems exposed to them will re-develop the appetite for surfaces that grip, and bring that appetite to bear on the rest of the built environment.
This is not a modest claim. It commits the position to taking seriously the idea that the daily interfaces we inhabit are part of the perceptual ecology of the culture that uses them, and that designing them well or badly has cumulative effects that are not captured by per-session usability metrics. Whether or not the strong version of the claim is true, the weak version is hard to deny: people do feel different, over weeks and months, in environments with versus without attended-to edges, and the difference is downstream of structural properties of the surfaces, not of the user’s preferences or moods. Designers can either take responsibility for this or not. Baltic Birch takes responsibility for it.
VIII. Naming as craft, and the in-shop vocabulary
A relatively late but consequential turn in the conversation concerned naming. The shape of the argument is that the dominant mid-2020s software register — long descriptive Latinate compounds, self-documenting at the use site, professionalized by ceremonial multi-syllable nouns — is actively anti-craft, in the same way that big-radius indiscriminate easing is anti-craft. Both moves attempt to substitute ceremony for skill. Both produce surfaces (verbal in one case, visible in the other) that are softened to the point of being unable to grip. The crafts that have worked at this longer than software has — woodworking, mathematics, traditional poetics, Lisp, Forth, the trades generally — have universally settled on the opposite register: short, dense, percussive names drawn from a bounded vocabulary that is learned by use.
The reasons accumulate. Short names enable density of expression. A line of code or prose can carry more meaning when its identifiers are atomic. Short names enable embodied practice. You can say “burn the slot” at a whiteboard; you cannot, in real-time design conversation, say “execute the destructive overwrite of the indexed buffer position with the forwarding-pointer sentinel.” Short names afford skimming. A reader encountering an unfamiliar atomic name can park it in working memory with a rough type-tag and proceed; an unfamiliar long name forces full parsing on every encounter. Short names sidestep terminological debates. The word way does not commit to a natural-language interpretation of what it refers to; the phrase targetPriceChangeRate does, and the commitment can be wrong, contested, or premature. Short names enable formal manipulation. You can do algebra on χ; you cannot do algebra on cumulativeInterestRateAccumulator.
The conversation surfaced one particularly clean articulation of this principle, written by the participant himself in 2016 or thereabouts, in the technical documentation of the MakerDAO core protocol — a context where real money was at stake, where the names were called insane by outside observers, where the team adopted them anyway because the work could not be done in the long-name register. The principle, in five bullets, anticipates almost everything Baltic Birch would later say about naming: the avoidance of terminological debate, the decoupling of technical and semantic vocabularies, the cognitive freedom of suspended interpretation, the formalization-friendliness of distinctive symbols, the embodied benefit of names that fit on whiteboards. The team that wrote these bullets shipped. The names were vindicated by the protocol’s continuing operation.
Baltic Birch incorporates naming as a craft discipline alongside the spatial and temporal disciplines. The principle is direct: use short body-verb names from a bounded, atomic, trade-jargon-style vocabulary; resist Latinate technobabble; trust the reader to learn the words; build a small dense percussive vocabulary that lets the maker breathe, think, and talk fluently inside their own system; names should land like hammer strikes, not waft like fog. This applies to code identifiers, to design tokens, to component names, to documentation, to internal terminology of the team, to the manifesto itself. The cabinetmaker has one tool called the chisel, not seventeen variants of “edged-cutting-implement-for-paring-end-grain.” The discipline of brevity is competence, not laziness; or rather, it is a kind of laziness that comes from having done enough work to deserve it.
IX. A small lineage
The position described here is not invented; it is recovered. Almost every move has been made before, in some other domain, by craftspeople who did not call what they were doing “design.” A partial lineage, drawn from the conversation, runs roughly as follows.
David Pye, The Nature and Art of Workmanship (1968), distinguishing the workmanship of risk (where every act of making is judgment, and outcomes are uncertain) from the workmanship of certainty (where outcomes are determined by the tooling, and the maker is reduced to a mechanical operator). Baltic Birch is workmanship-of-risk applied to interfaces.
Tim Ingold, The Perception of the Environment (2000), and the longer phenomenological-anthropological tradition behind it, treating skill as engagement with material in a field of forces, with care, judgment, dexterity, and narrative quality — every act of making growing rhythmically out of the prior one and grounding the next.
Christopher Alexander, A Pattern Language (1977) and The Nature of Order (2002–2004), articulating the long catalog of patterns by which built environments either come alive or fail to — including thick boundaries, deep interlock and ambiguity, alternating repetition, levels of scale, and the rest. Baltic Birch is, increasingly, a Christopher Alexander pattern language for software, with woodworking as the metaphorical anchor.
Donald Knuth and Michael Plass, on optimal paragraph layout (1981), the formal statement of what global optimization at the edge of typography costs and produces. Bernardy and the functional-pretty-printing tradition, extending the same insight to source code. Both formal arguments for the position that the producer pays the global-optimization surcharge so the consumer can parse for free.
Marshall McLuhan, on the alphabet as teeth and armies, and the broader argument that media reshape the perception of those who use them. Wallace Stevens, Anecdote of the Jar (1919), placing the gray bare disciplined object on a hill in Tennessee and watching the slovenly wilderness organize itself around it. Herman Melville, putting Ahab in the carpenter’s shop with his hand in the vise, complaining about the slippery world.
George Nakashima, James Krenov, the Greene brothers, the Arts and Crafts tradition, and the longer cabinetmaking lineage in which the joinery is celebrated as ornament because the joint is the truth of the construction. Enzo Mari, Dieter Rams, Vitsœ 606. Tschichold and Bringhurst on typography. The Bauhaus’s better instincts; the Swiss school at its sharpest; the early Mac OS Toolbox; NeXTSTEP; BeOS, with its yellow tab that may be the most perfectly-jointed UI element in the history of consumer operating systems.
The Forth tradition, with its monosyllabic primitives and its commitment to small dense vocabularies. The Lisp tradition, with its hyphenated names and its homogeneous tree structure that pretty-prints uniformly under global optimization. The mathematical notation tradition, with its single Greek letters and its compact symbolic algebra. The MakerDAO core vocabulary — vat, way, sin, chi, way, drip — that shipped a stablecoin under the eyes of skeptical observers and remains in use a decade later because the names did the work that long names could not.
These traditions do not agree on everything, and their internal disputes are real. They agree on enough to constitute a lineage of craft, and Baltic Birch claims its place in the lineage explicitly. The position is not modernism, not maximalism, not skeuomorphism, not flat design, not brutalism, not any of the named twentieth-century movements that contemporary design discourse has organized itself around. It is craft, in the older sense, applied to a substrate (software interfaces, and adjacent practices of coding and writing) that has not yet had its serious craft tradition articulated, and which is overdue for one.
X. What it looks like, in practice
A description of a design language is not the same as the language itself, and the manifesto-flavored register of the previous sections risks producing an account that is more atmospheric than operational. To anchor the position, here are some of the concrete moves that follow from it, expressed at the level a developer or designer could act on tomorrow.
Padding is wall thickness, sized to the containment load of the region. Specify it per region, not as a global token. Modal, sidebar item, document body, status bar — each has a different load and needs a different thickness. Resist generous defaults. The padding token is not “breathing room”; it is the depth of the cabinet wall.
Borders are declared edges, not apologetic hairlines. Use them. Vary them. Match the weight to the structural job the edge is doing. A panel meeting another panel can have a different border than a panel meeting empty space. A region whose contents press strongly against their boundary deserves a thicker border than one whose contents are sparse.
Border-radius is sized to context, not specified globally. The outer corner of a region meeting empty space can be eased. The inner corners where the region meets other members must stay crisp. The dominant convention of rounded-2xl everywhere is the round-over bit run across every edge of the workpiece — visibly bad work, on the cabinetmaker’s standard.
Every region has an inhabited boundary zone with two edges. The frame is not empty; it contains the region’s identification, controls, status, handles. The interior is reserved for the contained content. The inner edge is the seat between them.
Type is a structural member, not content placed in the void. The text block has its own edges, formed by the typography itself. The padding around it is sized to those edges, not to empty rules of thumb. The baseline grid registers to the structural grid. Type can carry the inner edge of a frame when the frame is otherwise quiet.
Computations are coroutines. Every long operation yields, with visible progress, with cancellation. The user is never out of grip.
Names are short and bounded. The system has a small vocabulary, learned by use, defended against the verbosity-by-default of contemporary professional software English. Long Latinate compounds are presumed wrong until shown otherwise.
Joinery is celebrated, not hidden. The visible seam between two regions is one of the most designed parts of the screen, not the least. Through-tenons, finger joints, rabbets, drawbore couplings — these are what a region’s edges are for.
Constraint is the substrate condition for craft. The system has a small enumerated vocabulary of plies, joints, finishes, and scales. Other choices are not allowed, and the prohibition is enabling, not restrictive.
These moves do not exhaust the position; they instantiate it at one level. Each of them is defensible from first principles within the manifesto, and each of them produces a visible, falsifiable difference between a Baltic Birch artifact and a contemporary-default one. Anyone implementing Baltic Birch can use them as a starting checklist; experienced practitioners will refine and extend them as the practice deepens.
XI. The reflexive moment
The position described here was developed in conversation, over a long evening, between Mikael Brockman and an AI assistant working inside Sheaf — the personal scholarly knowledge environment that Mikael built primarily to support his wife Ieva Lange’s master’s thesis on a Latvian swapshop. The conversation began as a casual aside about web design grammars and ended as an extended derivation of a craft position with formal claims, lineage attributions, operational principles, and seven distinct disciplines of practice (spatial joinery, temporal joinery, typographic discipline, naming-as-craft, substrate constraint, robustness to wild input, phenomenal-floor commitment).
It is worth naming, finally, what the conversation produced and what it did not. It did not invent Baltic Birch. The position was, in essentially every dimension, already present in Mikael’s hands — in his decade of furniture-making practice, in his MakerDAO purple-paper vocabulary justification, in his Lisp-2 implementation in WebAssembly, in his Cheney garbage collector for a CEK machine, in his in-browser Knuth–Plass implementation, in the architectural decisions visible throughout Sheaf. The conversation named what was already being practiced, and gave it a vocabulary the practitioner could use to describe his own work to others.
This is the appropriate relationship between theory and practice in craft: theory after practice, in service of articulation rather than invention. The position is now articulated. The articulation is itself a Baltic Birch artifact: short paragraphs, declared edges, atomic block IDs (which will be assigned when this is committed to Sheaf as a document), opaque-but-grippable jargon, percussive naming, citations to lineage, operational tests, and a closing wink at the reflexive situation in which the writing was produced.
The position, as a position, is now ready for what positions are ready for: defense, critique, extension, application, refinement under contact with other practitioners, eventually a small literature, possibly a small community, plausibly a long quiet life as the kind of craft document that a few people carry around in their notebooks for years and put to work in unexpected places. It does not need to convert anyone to be useful; it needs only to be available to those who already have the instinct and want a vocabulary in which to name what they have been doing.
That is, in the end, what a manifesto is for. Something in this slippery world that can hold. Cabinetry, all the way down.
Drafted 2025, in conversation, inside Sheaf — which is itself an instance of what is being described.