MixSafe — design & distribution

Party hard.
Mix safe.

I built MixSafe — a harm-reduction tool that rates any combination of 41 substances in about four seconds — and then I built everything that got people to it.

Every asset on this page is mine: the logo system, the campus flyer series, twelve rounds of stickers, and the social kit. No agency, no template, no stock. The engineering problem didn't stop at the site — a tool nobody finds helps nobody, so distribution got designed with the same care as the checker.

221Artboards

23Campus flyers

12Sticker rounds

300 dpiPress-ready

See the full selected work — web, stickers, print, social →

01 / Identity

The information pill

A capsule split clean across its seam: two halves that meet, and a visible gap where the answer goes. It had to work at 24px as a favicon and at four inches on a poster, in one colour, on both grounds. The study below is the argument for the mark, not just the mark.

Key visualTwo paths, one intersection — the whole premise of the tool in one image.

The markConcept board with the construction and the reasoning behind it.

The seamWhere the gap sits, how wide it runs, and why it survives being shrunk.

The wordmarkThe mark set into the name, so the logotype carries it without explanation.

02 / Campus flyers

Twenty-three sheets, one argument each

Every poster makes exactly one claim and backs it with a number from the tool's own data. No sheet asks anyone to stop doing anything — the campaign works because it answers a question people are already asking, rather than delivering a lecture they've heard before. Letter size, designed to hold up taped at eye level in a corridor.

Poster reading Party hard. Mix safe. on a dark ground with a QR code — **Party hard. Mix safe.**The flagship sheet; the line the whole campaign hangs on.

High-contrast red poster reading Party hard, Mix safe — **Red cut**Same line at maximum contrast — built to read from down the corridor.

Poster showing all 820 rated substance pairs as a grid of coloured cells — **820 combos. One search.**Every rated pair as one colour cell — the whole dataset as the artwork.

Poster with nested squares sized to the counts 820, 421, 218 and 139 — **Drawn to scale**820 / 421 / 218 / 139 as nested areas, so the proportion is the argument.

Poster with a ranked bar chart of the eight substances in the most dangerous pairs — **Four of the six worst are in a pharmacy**Ranked bars, split by verdict — the finding that surprises people most.

Poster with stacked bars showing how the 820 pairs split across six verdicts — **Six answers, none “safe”**Bar height is the share of pairs. The best any combination gets is *low risk*.

Light poster reading Before you mix, check, with the four verdict levels listed — **Before you mix, check.**The plainest sheet in the set — the one a health centre can put its name on.

Poster breaking down the sources behind the 820 ratings — **Where the data comes from**421 / 244 / 155 by source. A rating without a provenance isn't a rating.

Green poster reading Nothing happened, about the redose trap — **Nothing happened.**The redose trap — the failure mode where the quiet result is the danger.

Poster contrasting physical harm and headspace harm under the same caution label — **Same warning. Two kinds of harm.**One severity word can't tell you *which* — so the sheet labels both axes.

Dated poster for Halloween weekend reading Someone handed you something — **Someone handed you something.**Dated for one weekend — the campaign's only sheet with an expiry.

Corkboard poster with tear-off strips along the bottom edge — **Tear-offs**The only sheet that leaves the wall with you — one strip, one QR.

03 / Stickers

Twelve rounds to get to five

A sticker has to land in the two seconds someone spends at a sink. Twelve rounds went by before a set earned a print run — most of what got cut was cut for being clever instead of clear. These ship as trim and bleed files at 300 dpi, with the die radius flattened so a cut landing 0.02in off still takes artwork rather than white.

Black 4x1 inch sticker reading Your medicine cabinet is on the list — **Cabinet**4 × 1in strip. The whole pitch in seven words.

Sticker reading 36 of 40 pairs with the allergy pill carry a warning — **Antihistamine**The most over-the-counter drug in the set, and the most warned about.

Sticker comparing acetaminophen and diphenhydramine warning counts — **Same aisle**Two drugs from one shelf, 4 warnings against 36.

Cream sticker reading LSD has one dangerous pair on the whole list — **One pair**Runs against the expectation — which is exactly why it gets read.

Dark sticker reading You can ask without telling anyone — **Without telling**Aimed at the real barrier: a phone that other people can see.

Sticker with alcohol, benzos and opioids at the corners of a red triangle — **Triangle**Three nodes, three edges, every edge dangerous. No text needed.

Sticker reading Tramadol has more dangerous pairs than fentanyl — **Tramadol**22 of 40 against fentanyl's 12 — the statistic that stops people.

Sticker showing warning counts for melatonin, caffeine and alcohol — **Zero**Melatonin 4, caffeine 13, alcohol 27. Nothing on the list is inert.

Circular sticker plotting all 820 pairs as coloured dots by verdict — **Field**All 820 pairs, one dot each, sorted by verdict — the dataset as an object.

Cream sticker reading Nobody plans to mix — **Nobody plans**“It is a drink on top of a prescription.” The most common case, named.

Sticker reading It's probably fine is not a source — **Not a source**Overheard at a party, promoted to a claim. The campaign's thesis.

Wide sticker with the URL over a bar of the four verdict colours — **Scale**The URL sitting on its own data — 139 / 79 / 203 / 399, to width.

04 / The system underneath

Designed to be measured

The part that makes this engineering rather than decoration: every sheet is instrumented, every figure is traceable, and the print files are built to survive a vendor rather than to look right on a screen.

Attribution

Each sheet carries its own QR code resolving to mixsafe.org/?f=<code> — two letters, because every extra character makes the code denser and a dense QR fails at sticker size. Before that, all twenty-three posters shared one code and were a single undifferentiated blob in the analytics. A flyer code identifies a poster, never a person, so the “no tracking” line on every sheet stays literally true.

One number, one source

Every figure printed here is pulled from the tool's own database, not estimated: 41 substances, 820 rated pairs, and the 139 / 79 / 203 / 399 verdict split. An earlier run of sheets quoted 35 and 595 — printing those would have meant paying to publish a stale claim, so the whole set was regenerated against the live data first.

Press-ready output

Boards are authored at trim and rendered to both trim and bleed at 300 dpi, with a pHYs chunk written so a printer reading DPI from the file lands on 300 instead of 72. Bleed is edge replication rather than a guessed fill colour, which is the only thing that works against a gradient ground.

Rendered from source

No asset here was exported by hand. Every board is HTML, laid out on a canvas and screenshotted headless by a build script, so a corrected figure means editing one value and re-running — not reopening twenty-three files and re-exporting each one.