Visual Grammar

1. The two modes
2. Why the source looks like this
3. Shared device library
4. Typography
5. Scale
6. Colour
7. Texture and imagery
8. Composition
9. Invented telemetry

The two modes

Both modes share the device library, the micro-typography rules, and the legal guardrail. Applied is the more common request, and it’s where the style actually lives commercially.

A single piece often nests them: an applied poster containing a simulated spec block, or a garment whose main graphic is art-directed while a corner detail simulates a data plate. That’s the strongest use — the simulated fragment gives the composed piece its credibility.

Why the source looks like this

Every signature traces to a physical or legal constraint on the real artifacts. This matters even in applied mode, because knowing the cause is what lets you break a rule intelligently.

No gradients. Every dominant marking process is binary at the mark level. A thermal printhead element either fires or doesn’t. Laser marking is a physical surface change with no tonal axis. Moulded relief has no ink at all. There is nowhere for a gradient to live.

No halftones — in the source. Flexo has a hard 2% minimum dot; screen printing loses 3–5% highlights entirely and suffers 8–35% dot gain. In pad printing the shallow etch required for fine text is chosen precisely so fine lines and halftones don’t plug. (Applied work reintroduces halftone deliberately as texture — see below. That’s a knowing violation, not an accident.)

Hairline minimums, and an asymmetry. Flexo: positive rule 0.13 mm, reverse rule 0.38 mm. Positive copy 6 pt, reverse copy 8 pt. Ink spreads into the counter, so knocked-out elements need roughly 3× the rule weight and +2 pt of type. This is the most transferable single detail in the idiom — authentic artwork has visibly chunkier strokes and larger type inside black bars, and reproducing it sells a piece instantly.

Geometric simplification. Sharp corners, hairline serifs, and delicate joints are unmanufacturable: moulding guidance calls bold sans-serif mandatory and rejects serifs because thin serifs are “too small to mill.” The primitives — circle, square, triangle, trapezoid — survive a 12 dots/mm printhead and 20 laundry cycles. That’s why they’re the primitives.

Dense information packing. The regulatory payload is fixed and additive; the plate is not. When the FCC rule concedes that if 4-pt type won’t fit the statement moves to the manual, that’s a regulator admitting the plate ran out of room. The density is the trace of a fixed payload meeting a fixed area — not a stylistic choice.

Rules and boxes instead of colour. Colour costs a pass — one per pad cycle, one per screen, none at all for laser. A hairline rule is the cheapest mark available. So the grouping job colour normally does gets handed to rules, boxed compartments, and reversed bars. That single economic fact explains most of the layout, and it survives into applied work as a habit.

Shared device library

The vocabulary both modes draw on:

Structural

• Hairline rules — positive 0.25–0.5 pt; reversed rules ~3× heavier
• Boxes, cartouches, and compartments — a box of text reads as a mark
• Reversed banner bars — knocked-out type at the heavier weight and +2 pt
• Corner brackets and registration squares, often accent-coloured
• Crosshairs, crop marks, tick scales and graduated rulers
• Leader lines with callouts — elbowed, never curved

Typographic

• LABEL: value pairs — the fundamental unit
• Bracketed and boxed tags — [01], [ NAME ], ( STATUS ), black-box and outline-box labels
• Slash-separated category runs — AI / ROBOTICS / DEFENSE / INDUSTRY, SF NYC LDN
• Chevron runs — >>>>>>>
• Vertical rotated micro-labels running up an edge
• Document metadata — ISSUE 01, DISTRIBUTED BY:, DOCUMENT TYPE:, 01/03, revision and date codes
• Oversized ® as a graphic element

Marks and emblems

• Invented circular seals — ring of type around a monogram or star
• Equipment symbols — AC/DC, Class II double square, hazard triangle, crossed-out bin (tier (c)/(d) only, see marks-and-legal.md)
• Barcodes and 2D codes — real ones only
• Hazard stripe blocks — diagonal barricade striping

Punctuation is structure: : after a label, / between fields within one spec (100-140V / 9.5-6A / 50-60Hz), ; between spec groups.

Typography

Simulation mode: condensed grotesque plus mono, at compliance scale, and nothing else. The real constraint is fitting fixed-length strings into a small rectangle.

Applied mode runs four type registers, not two — wordmark, display headline, techno/squared, and mono micro-layer. Using fewer is the most common reason generated work looks flat. references/identity-system.md specifies all four; read it before setting any type in applied mode.

Two things to fix immediately if output feels wrong:

• The display face is warm, and the headline is sentence case — not condensed, not all-caps. Caps belong to the techno and micro registers only. A soft, heavy, tightly-tracked sentence-case headline against dense mono micro-type is the actual signature of the style.
• The squared “techno” face is usually missing. That wide, stencil-ish industrial face used for years, versions, and one-word shouts (2026, V 1.2, 24/7) does a large share of the industrial signalling by itself.

Under a strict CSP nothing loads from Google Fonts — embed as base64 or use a system stack:

Micro-typography (both modes)

Figures. font-variant-numeric: tabular-nums lining-nums slashed-zero on every serial, lot, part, and rating number. Two independent published sources require 0/O and 1/l/I disambiguation for exactly this context.

Tracking. +5–12% on caps and small caps (letter-spacing: 0.05em–0.12em). The guard-rail: if the gaps could fit another letter, you’ve gone too far. Letterspacing lowercase is permitted under 9 pt, which licenses opening tracking as type shrinks.

For a more engineered feel, ISO 3098 spacing is much looser than typographic tracking — character spacing = 2 × stroke width — which is why authentic technical lettering reads wide and slightly under-kerned.

Caps are right for labels, codes, headers, and signal words — most of this idiom. Avoid for running text, and avoid synthetic small caps (most OFL condensed grotesques have none, and scaled caps read too tall and too light). Prefer letterspaced full caps at reduced size.

Don’t justify. Flush left, ragged right — which is what real plates do anyway.

Scale

Simulation: the 4–9 pt band, throughout. Five independent regulators converge there — FCC 4 pt, MIL-STD-130N 5.76 pt, UL 969’s 1.60 mm test ceiling, FDA 6 pt, EU pharma 9 pt. On screen that’s ≈5–12 px at 1×. Below ~4 pt it becomes noise; above ~12 px it stops reading as compliance labelling.

Applied: extreme scale contrast is the engine. Enormous display type against tiny mono, with almost nothing in between. The two-sizes-and-no-intermediate rule survives from the source — but the two sizes become huge and tiny rather than ratings and boilerplate. An intermediate size still reads as designed-by-committee and still kills it.

ISO 3098 lettering heights give a ready-made scale on a √2 progression: 1.8 / 2.5 / 3.5 / 5 / 7 / 10 / 14 / 20 mm, stroke = h/10.

Colour

Simulation: monochrome. Black on white, silver, or grey; or white on black. Single-colour printing isn’t a preference, it’s what one pass gives you.

Applied: the palette is open, and this is where my first draft was most wrong. Observed in practice: safety orange, mint green, cream, deep blue, yellow, and black used as full grounds — the accent is frequently the entire field, not a sparing highlight. Saturated primaries (red/yellow/blue) appear as flat fills, often with line-screen texture through them.

What stays constant:

• Flat fills only. No gradients as ground. Colour is either fully on or fully off.
• One ground + ink + at most one or two accents. The discipline is in the count, not the saturation.
• Reversed panels need the weight bump — ~3× rule weight, +2 pt type, whenever you knock type out of a dark field.
• Dark grounds are fully legitimate in applied work, and common. The “light background, print-friendly” rule belongs to the source artifact, not the style.

Texture and imagery

Absent from the source, central to the practice:

• Line-screen fills — horizontal rules struck through solid letterforms and shapes. Probably the single most recognisable applied-mode move.
• Halftone and dot-matrix imagery — engravings, blobs, flags, portraits, gradients built from visible dots. A knowing violation of the source’s no-halftone constraint.
• Film grain / noise overlay at low opacity across the whole composition.
• Graph-paper and dot grids as background fields.
• Hazard striping — diagonal barricade blocks.
• Redaction bars — solid black rectangles struck over body text, as though declassified. Cheap, and it instantly implies an institution with secrets. Use once or twice, over plausible words.
• Checkerboard blocks — as a band, a corner fill, or a frame around a headline.

These are what keep applied work from looking like a spreadsheet. Use two or three per composition, not all of them.

The emissive variant

A significant strand of this work is screen-lit rather than printed — and my earlier framing wrongly ruled it out. When the subject is software, AI, or anything with a display, the idiom shifts to a CRT/LED register:

• Glow and bloom around saturated marks and type on near-black
• Scanlines through letterforms and shapes
• LED-matrix and shutter grids overlaid on imagery — vertical or crossed bar arrays that break the image into cells
• Seven-segment digits for clocks, counters, and countdowns
• Phosphor colour — saturated red, amber, yellow, blue against near-black
• Status furniture rendered as though on a monitor: ▪ LIVE in an outlined box, WAITING…, a red-ruled header bar

This shares the emissive quality with dreambase-industrial-schematics, and the two genuinely do overlap here. The distinction that still holds: schematics is cinematic and dimensional — perspective, depth cues, rendered machinery. Micrographics stays flat and frontal even when it glows, and it keeps the LABEL: value document grammar. If the piece has a camera angle, it’s schematics; if it has a form field, it’s micrographics.

Print and emissive are a surface choice available within applied mode — not two different styles. A single identity often ships both: a matte orange container and a glowing red terminal.

Composition

Simulation mode — assemble by constraint. Real plates are assembled by regulation, not taste, and no authority can harmonise marks from a dozen standards bodies. Each mark’s licence forbids altering its proportions, so the only free variables are position and scale of each intact lockup. So, deliberately:

1. No shared baseline or x-height across marks. Optical centring, with cap heights differing up to 3×.
2. Let each mark bring its own typography — no type system across the mark band. The strongest authenticity signal available.
3. Irregular gaps, varying 2–3× within one band.
4. Derive spacing from the shapes. The care-symbol system’s gaps are 10, 22, 20, 17 — computed per pair. Evenly-spaced symbols are the number-one giveaway.
5. Satellite text doesn’t align to the text grid.
6. One or two marks anchor at large size; the rest pack around them.
7. Variable data is a separate layer — a serial sticker applied over the artwork, slightly rotated, overlapping the border.

Applied mode — compose deliberately. Centred, balanced, art-directed. Alignment is a virtue again. The idiom’s density and micro-detail carry the technical signal; you don’t need misalignment too, and faking it here just reads as sloppy.

Density is enforceable, not aspirational

The most common failure in generated work is a composition that is correct but empty — right devices, right palette, and a dead void through the middle. The source artifacts have no voids because a fixed payload met a fixed area, and applied work has to earn the same quality deliberately.

Hard rules:

• No empty region larger than ~15% of the canvas. Measure it. A blank band across the middle is the classic tell.
• If content doesn’t fill the canvas, shrink the canvas — don’t distribute the content to fit. Spacing elements apart to fill space is exactly backwards.
• Never use a flex spacer to push a footer down. That is the mechanism that produces the void. Stack blocks in reading order and let the canvas end where the content ends.
• Every quadrant carries something — a micro-label, a rule, a tag, a texture field, an illustration. The corners especially: this idiom anchors all four.
• Whitespace appears between dense blocks, not inside them. Tight groups separated by clear gutters — not evenly-distributed elements floating in air.
• Two or three texture moves minimum in applied mode, and one of them should occupy real area (a halftone image, a line-screen field, a dot grid), not just a hairline.

A useful check: squint at the piece. You should see three or four distinct masses. If you see scattered marks with air between them, it isn’t finished.

The strongest work nests the modes: a composed piece containing one simulated fragment — a spec block in a corner, a data plate on a sleeve, a credential in a hero shot. The simulated fragment lends credibility; the composition makes it a design.

Invented telemetry

The heart of applied micrographics, and something the source has no equivalent for: LABEL: value applied to narrative rather than measurement.

Observed in practice: STATUS: DEPLOYED IN SECTOR, FIELD OPERATION: ACTIVE, OBSERVATION LAYER: ACTIVE, CLEARANCE: LEVEL 3, DIVISION: CORE PRODUCTION, ROLE: PRIMARY OUTPUT, LOCATION: DOMESTIC, UNIT ID: HM-FU-01, plus live clocks, temperatures, SCANNING THE GLOBE, WAITING…, ● LIVE.

This is world-building, and inventing it is the point — not a violation of the no-fake-data rule. The distinction that matters:

The rule underneath: fiction is fine when it reads as fiction; fakery is not fine when it reads as fact. A CLEARANCE: LEVEL 3 on a poster invents a world. A wattage rating on a product label that nobody measured is a lie. A barcode that doesn’t scan is just bad craft.

The vocabulary works because it implies an institution — an org chart, a division, a supply chain. Invent the institution first, then the telemetry writes itself and stays consistent.