games.productartistry.com

What is Product Artistry Games?

Product Artistry Games is a small collection of interactive browser games and playgrounds about how digital products actually work, made by Rohan of the Product Artistry newsletter. Rather than teaching product thinking through long essays, it turns those ideas into short, playable exercises.

What's there

• Dark Patterns Detective — eight interactive scenarios, roughly five minutes total, built to train your eye for manipulative design in checkouts, sign-ups and cancellation flows.
• LLM Evals 101 — listed as coming soon: an exercise where you pick a model, write tests and tune guardrails until an AI assistant is good enough for real users.
• More pieces are described as in progress, released alongside the newsletter.

Who it's for

Product managers, designers and founders who want a quick, hands-on way to sharpen judgment rather than read theory. The format suits a coffee-break session or a team warm-up: one scenario, then a discussion about what the interface was doing and why.

How to use it

Start with Dark Patterns Detective if your work touches sign-up, pricing or cancellation flows — it's the concrete, finished piece. Compare what you spot in the game against a flow your own team owns, and note where the same patterns appear. Then check Product Artistry for the essays the games are based on, since the site says most of them begin as newsletter posts.

How can I use Dark Patterns Detective to spot manipulative design in checkouts and cancellation flows?

Dark Patterns Detective is an interactive training tool rather than a checklist or audit report. It puts you through eight short scenarios — the whole set takes about five minutes — set in the places where manipulation is most common: checkouts, sign-ups and cancellation flows. The point is pattern recognition: you look at a simulated screen, decide what is wrong, and learn to name the trick.

How to get value from it

• Play it once straight through to see which scenarios you miss. Most people catch fake urgency and hidden costs quickly, but stumble on confirm-shaming or pre-selected add-ons, because those feel like normal defaults.
• Play it again with a real product open in another tab. Pause at each scenario and ask whether the screen you are looking at does the same thing. This turns a five-minute game into a rough audit of your own checkout.
• Use the scenario names as a shared vocabulary. "This is a roach motel" or "this is a sneak into basket" is faster in a design review than describing the flow from scratch.

What to look for in cancellation flows specifically

The scenarios train you to notice asymmetry between the two directions of a journey. Signing up is one click; cancelling takes a phone call, a chat, or a hidden link. Watch for:

A practical next step

Run the game, then walk your own cancellation flow on a phone, as a logged-in user, without internal shortcuts. Time it and count the clicks. If cancelling takes more steps than subscribing, you have found the same asymmetry the game is teaching you to see.

For a broader product-design perspective alongside the game, Rohan's writing lives at Product Artistry, and the games themselves are at Product Artistry Games.

What should I consider when choosing a model and writing tests for an AI assistant in LLM Evals 101?

LLM Evals 101 is a planned Product Artistry playground, so its specific mechanics aren't documented yet. The page describes the premise: you're shipping an AI assistant, and you pick a model, write the tests, and tune the guardrails until it's good enough for real users. Treat the following as general advice for that kind of exercise rather than confirmed features.

Choosing a model

• Start from the failure you can't tolerate. A support assistant that invents refund policies has a different bar than a brainstorming tool that occasionally wanders. Define the unacceptable output first, then ask which models can be kept away from it.
• Separate capability from controllability. The strongest model on your task may also be the hardest to constrain, slowest, or most expensive per interaction. For an assistant, "good enough plus predictable" usually beats "best plus erratic."
• Test at least two tiers. A smaller, cheaper model often handles the common cases, with a larger one reserved for hard inputs. Comparing them on your own test set tells you whether the split is worth the complexity.
• Match the model to the interface. If the assistant must respond in a fixed format, call tools, or stay inside a strict tone, check those behaviours early rather than after you've built the rest.

Writing tests

• Write tests before tuning prompts. Otherwise you're grading your own homework: prompts get adjusted until a few examples look right, and you learn nothing about the cases you didn't try.
• Build a small set of realistic inputs, drawn from actual user phrasing, including typos, ambiguity, and hostile or off-topic requests.
• Score behaviour, not wording. "Did it refuse appropriately?" and "Did it use only the supplied policy?" are stable checks; exact string matching will fail on harmless rewording.
• Include regression cases. Every time you find a bad output, add it. That's how a test set becomes useful instead of decorative.
• Decide pass criteria in advance. If you don't know what counts as acceptable before you look at results, you'll rationalise whatever you get.

Guardrails and trade-offs

Guardrails usually cost something: stricter filters refuse more legitimate requests, and heavier checks add latency and complexity. Track both directions — false refusals as well as unsafe passes. A guardrail that blocks a third of valid questions is a product problem, not a safety win.

One practical way to keep this honest is to run the same test set against each candidate model and guardrail configuration, then compare the failure lists side by side. The differences are usually more informative than the pass rates.

If you want to see how the author approaches product problems before the playground ships, his newsletter essay on de-risking delivery is the closest existing material on the site, and the page says new playgrounds go out with the newsletter. You can also read more of his writing at Product Artistry.

How can I get notified when new games or playgrounds are released?

Sign up for the Product Artistry newsletter. The games page says new playgrounds go out with the newsletter, and the one unreleased item shown — an LLM Evals 101 playground — has a "Get notified" action rather than a live link. So the newsletter is the intended notification channel, not a separate release feed or changelog.

What to expect from that channel

The same page notes that most games start life as an essay in the newsletter, so subscribing gets you the thinking behind each interactive piece, not just a launch ping. If you only want the games and nothing else, expect some product-writing posts in between.

If you'd rather not subscribe

Check the page periodically. The layout lists each game with a short description and a status label ("Play now," "Coming soon"), so a quick scan tells you what's new without opening anything. To widen your net beyond this one site, design and product communities often surface these releases: Hacker News front-page threads, Product Hunt launches, and Department of Product's newsletter. That's general advice about where small interactive product tools tend to get shared, not a claim about this site's own channels.

A practical step: subscribe, then keep the confirmation email — that's your route back in if you later want to unsubscribe once the playgrounds you were waiting for have shipped.

How do the games relate to the Product Artistry newsletter essays?

The games are built as interactive companions to the newsletter essays, not as a separate product line. The site describes them as “interactive pieces that make product ideas easier to see,” and notes that most start life as an essay. So the typical relationship is: an essay introduces a product argument, and a game lets you practice spotting or applying it.

What that means in practice

• Essay first, game second. The written piece carries the reasoning and the examples; the game turns that reasoning into a short, repeatable exercise.
• Small and focused. Dark Patterns Detective is eight scenarios lasting about five minutes, which matches the newsletter’s style of one clear idea per issue rather than a broad course.
• Practice over reading. Instead of just reading that cancellation flows can be manipulative, you work through checkouts, sign-ups and cancellation flows yourself and build the recognition habit.
• A funnel in both directions. The games page points readers to the newsletter for new releases, and the newsletter surfaces the latest essay, so each format feeds the other.

Choosing which to use

A practical example: a product manager preparing a checkout redesign could read the relevant essay for the reasoning, play Dark Patterns Detective to sharpen their eye for manipulative patterns, then bring the scenarios into a team review as discussion prompts.

The main trade-off is depth versus speed. Essays can carry nuance and caveats; a five-minute game cannot. Treat the game as a way to make the essay’s point stick, not as a replacement for it. If you only have time for one, read the essay when you need to decide something, and play the game when you need to notice something.

What is the best way to use these interactive scenarios in a product or design training session?

Use the scenarios as a diagnostic warm-up, not as a lecture. Run people through a few cases first, collect their calls, then name the patterns they missed. The site describes its Dark Patterns Detective as eight interactive scenarios, about five minutes, covering checkouts, sign-ups and cancellation flows, so it fits inside a short workshop slot rather than a half-day course.

A workable session shape

1. Frame it as a real task. Tell participants they are reviewing a flow for a company they would actually advise, and that they should note why each screen works on them.
2. Play individually, silently. Roughly five minutes of solo play produces more honest answers than a group clicking through together, where the loudest voice dominates.
3. Debrief by pattern, not by score. Ask which scenarios split the room. Disagreement is the teaching material: one person reads a pre-ticked box as convenience, another as consent theatre.
4. Move to your own product. Have teams audit one real flow — a checkout, a trial sign-up, a cancellation path — and label each moment as helpful, neutral or manipulative.
5. Agree a rule. End with one concrete commitment, such as "no countdown timers on pricing pages" or "cancellation reachable in the same number of clicks as sign-up."

Who it suits

Practical cautions

The five-minute format rewards speed, so people may click through without reading carefully. Give a written prompt — "note one sentence per scenario on what it tried to make you do" — to slow them down.

Dark patterns also sit close to legitimate persuasion. If the debrief only produces a list of banned techniques, teams will treat it as a compliance exercise. Better to ask what the honest version of each flow would look like, and whether it would still convert.

For a longer programme, pair the scenarios with a written piece on the same theme; the site notes that most of its pieces begin as newsletter essays, so the interactive work and the reading reinforce each other. If you want more practice material, Product Artistry Games is the source, and its newsletter is where new ones appear.