Pass & Play Chess
The Board I Could Not Find
Two people, one device, one chess board. Every site I found squeezed the board between adverts, and none of them would fill the screen. So I built the one I was looking for.
Pass & Play Chess gives the board the whole screen. No account, no pairing, no matchmaking, and no advert on the board. Open the page and play. It lives at chess-offline.com, and I am the whole team: developer, product owner and growth.
The first commit landed on 5 May 2026 and a playable first draft followed four days later. Between 21 October and 17 November 2026 it recorded 169,000 moves across 1,900 completed games, from 1,700 monthly active users, up 31% on the previous 28 days. Almost all of them arrived from organic search, on no marketing spend.
The Bet
Pass and play is the slice of chess that online chess ignores. chess.com has over 100 million users and lichess is free and open source, but both assume two accounts and a network between them. Two people at a caf茅 table with one phone are served by neither.
There is no network effect here and no viral loop. A single device game cannot invite anyone. The bet was that search demand alone could carry a product with no marketing budget, and that the way to find out was to ship, instrument and read the numbers rather than argue about them.
So I shipped it in May and let it stand on its own. Nine commits went in between June and September. In October I sat down to read what the product had been doing without me.
The Numbers Were Wrong
The first thing the reports told me was that I could not trust them. Google Analytics 4 (GA4) has a slot that is meant to record which website sent a visitor. My code was putting something else in that slot: the reason a game began, such as fresh for a new game or reset for starting over. Those words then showed up in the reports as if they were websites that had sent players, so every report built on that slot was quietly wrong.
I rebuilt the events the site sends in October around one question: does a visitor make a legal move?
| Event | What it means |
|---|---|
game_setup_view | A player saw the setup screen. It is not yet a start. |
game_start | A confirmed start, with the chosen timer settings |
game_first_move | Sent once per game, at the first legal move |
game_end | A game that was played to a proper finish |
Some things are left out on purpose. The code that links a start to its first move is never turned into a report field, because it does not identify a player. Prompts and individual moves are not counted as goals, since that would inflate engagement and measure nothing. A finished game that is later reopened cannot count as a second finish. And GA4 does not see all play: ad blockers and lost connections drop events, so every figure here is a floor, not a total.
The fix is measurable. Of 4,200 starts between 21 October and 17 November 2026, exactly 5 still carry a gameplay reason in that website slot, down from all of them. The step conversion chart in the next section and the "Bing Sends More Players Than Google" chart both rest on this fix, because both depend on the right source being recorded against each start. Only then were the numbers worth reading.
Somebody Was Looking
Every figure below comes out of GA4, over the fixed window of 21 October to 17 November 2026.
The part I was least sure of is setup: two people on one device have to agree on a timer and a side before anything happens. It holds at every step.
Nobody is lost on the board itself, so attention went to what happens before a game starts rather than to the game.
Bing Sends More Players Than Google
82% of sessions arrive from organic search. The split is not the one I expected:
Bing out delivered Google for this niche, and did so only once Bing Webmaster Tools was set up in October. Before that the channel was invisible because it was unmeasured, not because it was absent. The lesson generalises: the cheapest growth available was a search engine I had not bothered to register with.
The Feature Players Turn Down
Games save automatically and offer to resume. The two figures below count different things, so they do not add up to 100%.
The first counts prompts. Of 1,200 resume prompts, shown to 538 people, 257 were accepted (21%). The second counts confirmed starts. Of 4,200 starts, 2,700 (64%) were resets and the other 1,500 (36%) were fresh loads. The 257 accepted resumes may not be among those 4,200 starts, so the two counts are not a split of one total.
Counted per prompt, about four in five resume prompts were not accepted. People arrive, and they want a clean board.
This is the clearest product finding in the dataset and it is an uncomfortable one, because the save system is where a good share of the engineering went. The work made resume correct. The data asks a different question: whether a resume prompt should interrupt the path to a new game at all. That is the next thing I will change, and it is a product change, not a bug fix.
Under The Hood
Two decisions carried most of the weight.
Shared device orientation. Both players read the same screen from opposite sides. Two strategies ship and the player chooses: by default the board stays white oriented and black's pieces rotate 180掳 in CSS below the desktop breakpoint; opt into auto flip and the whole board turns between moves, on a 300ms delay so the move is seen landing first. The two mechanisms would fight, so a wrapper class disables the CSS rotation whenever auto flip is on.
Clocks when a tab is hidden. If a player switches to another tab mid game, the browser slows down the timers a page can run in the background. When the player came back, the clock could jump forward by all the time they were away and end the game on time unfairly. The clock now limits how much time it can add in one step, and checks the real time again when the player returns to the tab.
The board is also fully keyboard playable. Arrow keys move a cursor, enter selects, and a polite live region announces each square, driven from the board container because the underlying library's keyboard sensor pans by a fixed 25 pixels, about a third of a square.
16 Jest suites and 14 Playwright specs cover Chromium, Firefox, WebKit and two mobile profiles. Most are named after a specific defect and assert outcomes rather than calls: Scholar's Mate produces a white win, and confetti does not fire on a draw.
Outcomes
- 169,000 moves and 1,900 completed games in 28 days, from 1,700 monthly active users growing 31% period on period
- 90% of confirmed starts reach a first legal move; half of all games finish
- 82% of sessions from organic search with zero marketing spend, led by Bing ahead of Google after registering with Bing Webmaster Tools
- An analytics rebuild that cut acquisition field pollution from every event to 5 in 4,200, making the activation funnel trustworthy for the first time
- A measured product failure: a save and resume system whose prompts were accepted 21% of the time, with 64% of confirmed starts being resets, now the top of the backlog
