Five Across

I built a live bingo game for a nine-night Mediterranean cruise. It ran at sea, then became the engine for a second event. Getting people to play was a separate problem.


The problem

Five Across started as Gay Cruise Bingo, a phone-first game for a nine-night Mediterranean cruise from Trieste to Barcelona. I had eight days to build it.

A printed card could carry the inside jokes and dares. I wanted shared marks, proofs and standings, so the game could keep running through the trip instead of being a novelty for a day.

Each player got a frozen, randomized 5×5 card from a community-editable prompt pool. They could tap a square when something happened and see who else had already marked it. A fresh themed card unlocked at 8:00 a.m. local time each morning after the first, matched to the day’s port and party. Tutorial cards bookended the trip. The live feed carried proofs, milestones, doubts (“pics or it didn’t happen”), and hearts.

  • 8 daysto launch
  • Ship Wi-Ficonnectivity
  • 10 dayslive operation
  • 16 playersfriend group

The decisions

The deadline made scope choices necessary. Each had a reasonable alternative: server checks for marks, an online-only launch, score incentives for reactions, an unchangeable deal, an automated platform cutover. I chose differently for this launch. One of those choices needed revising while the game was live at sea.

I also sized moderation for a private adult friend group. Before departure, the app had server-authoritative report auto-hide, an admin roster and a Cloud Vision gate deliberately off by default. The repository has no record of those controls being used during the sailing. That silence doesn’t tell me whether they were used.

  1. ScopeValidated

    Let friends mark their own boards

    What I encountered
    I had eight days to build a multiplayer game for an adult friend group. A server-authoritative game would check every claim before it counted, so players couldn't cheat.
    Over
    Cloud Functions checking every mark. That's the standard multiplayer approach, and the safer choice for an audience of strangers.
    What I decided
    I chose honor-system marking, with no server validation. Players write their own boards, and the client sorts the leaderboard. Each event can require a pledge, proof-to-mark, or admin confirmation. The default is a one-tap Cross My Heart.
    Why
    Cheating wasn't in the threat model for this group. I trusted the friends and the feed to challenge a claim if they needed to. Server enforcement would have added work for a problem I wasn't designing this launch around. Leaving it out also let Phase 0 deploy without Cloud Functions at all.
    Cost
    The standings are claims, not independently verified results. A player writes their own board, so the record can't distinguish a real mark from an invented one. Any player could demand proof.
    What it changed
    The frozen production Firestore holds 845 squares. Analytics recorded seven proof-demand events over ten days, all from the same player. Those are events, not seven distinct challenged squares; they can't be subtracted from the square count. The rest stood on the pledge.
  2. ReachMixed

    Assume the connection is already gone

    What I encountered
    Ship Wi-Fi was flaky and expensive, with dead zones at sea. Requiring a connection every time someone tapped a square would make the game unavailable when the group was together without coverage.
    Over
    A plain online web app. It would have been the fastest thing to build in eight days and a reasonable choice for a launch on land.
    What I decided
    I chose an installable PWA with an offline-first Firestore cache. Marks queue on the device and sync when the connection returns. A cold boot doesn't wait for the network.
    Why
    Coverage could disappear without warning. I wanted a player to be able to mark a square wherever they were, without finding a signal first.
    Cost
    I took on the PWA cache and update path. It became one of the two operational problems during the sailing, alongside authentication; the commits skew toward both. The surviving database and analytics records don't let me reconcile the two.
    What it changed
    Over the scoring window, from embarkation to the freeze, Firestore holds 845 squares and analytics logged 783 marks: a gap of 62. The raw analytics total also includes a pre-embarkation day and 30 marks after the freeze on the ceremonial card. Neither belongs in the standings comparison. Offline queueing could explain the gap. Ordinary analytics loss could too. It doesn't establish that players marked without coverage and synced later. I'd assumed the moments worth capturing would happen furthest from a signal. The marking data challenges immediate capture: 288 of 703 main-day marks, 41%, came after the next day's card unlocked, and 32% of all marks landed between 19:00 and 20:59 ship time. Players' debrief accounts describe recalling previous days together at dinner. Those timestamps still don't tell me whether they had a connection while marking. A September 26 PostHog query found fifteen cruise-era exception events containing auth/network-request-failed, across five recorded sessions for one recorded user on July 18–19. Those sessions had no recorded marks. Requests failed. Their cause, and whether offline support helped people participate, remain unmeasured.
  3. IncentivesMixed

    Keep hearts out of the score

    What I encountered
    The feed needed a reaction. Points for hearts, or hearts as a tie-break, would give players a reason to use it. My product principle put connection and shared memories ahead of score.
    Over
    Letting reactions affect rank. That's a reasonable engagement mechanic for a social game, but it gives each reaction a strategic use.
    What I decided
    I kept hearts out of stats, standings and win logic. They produce only the Most-Loved Photo award at the freeze. That award carries no points.
    Why
    A heart should be free to give because it buys nothing. Once recognition affects the standings, it's also a move in the game.
    Cost
    A heart that buys nothing is one nobody has to give. I'd removed the score incentive and left no other lever if the group didn't want to react on its own.
    What it changed
    Across ten days and sixteen players, one person gave all nineteen hearts. The principle held, but hearts failed as a social mechanic with this group. Their weak use didn't affect the standings because they never carried score.
  4. FairnessRevised

    Deal each card once

    What I encountered
    I wanted luck to be equal across the standings. If a player could keep trading cards, they could shop for a better board.
    Over
    Mulligans or rerolls whenever a player disliked their card. That would be a forgiving choice for a casual game.
    What I decided
    I started with frozen cards from a seeded shuffle, with no rerolls.
    Why
    A fixed deal made the fairness rule clear: nobody could negotiate their board or the score it produced.
    Cost
    A player whose card didn't fit the trip was stuck with it, with no way to tell whether the problem was the card or the trip. Two days into main-pool play, no main card had produced a bingo. The two cards after the embark tutorial would finish the cruise with one bingo apiece, both claimed on the last full day. I had to revisit the rule while the event was running.
    What it changed
    I changed the rule from a final deal to a deal that becomes final when the player acts on it. A card with zero marks can be traded for a fresh one, three times per player. Once a player has marked it, they can't trade it. That made the first deal negotiable for anyone who used the feature. The limit prevents shopping for a better board mid-play; it doesn't preserve the original rule intact. The repository calls it a reshuffle and explicitly avoids mulligan. It shipped on the itinerary's sole sea day, alongside easy embark-pool squares blended into main cards. Three of sixteen players used seven reshuffles, all during the sailing.
  5. PlatformPending

    Hold the cutover

    What I encountered
    Five Across is becoming a platform: one codebase, a wildcard event router for the edge, and centralized authentication with a single-use handoff. The next step is unrelated groups sharing a backend. Tenant isolation for that hasn't shipped.
    Over
    Attach the routes now. The worker is deployable and tested, and the route config is written. As of 2026-10-06, cutover would mean uncommenting two blocks, one per zone.
    What I decided
    I held the cutover. Separate events run on separate Firebase projects until the isolation workstream lands. Attaching the routes remains a deliberate human step.
    Why
    The repository's specs distinguish path-scoped rules from tenant isolation. Uncommenting a route block looks like a deployment step, but it makes a decision about other people's data. I wasn't treating it as routine.
    Cost
    Each new event for an unrelated group needs a Firebase project provisioned by hand. The platform code is built and waiting. Until the isolation work lands, I pay that setup cost again for each event.
    Validation boundary
    This decision isn't validated yet. The router carries no production traffic. The auth handoff was provisioned by 2026-09-10 and was live in the client as of 2026-10-06, but no sign-in through it has been demonstrated: registered hosts still sign in directly. There is no behavioral evidence either way. The remaining steps are the tenant-isolation rules workstream and the deliberate act of attaching the routes.

What happened

The production Firestore preserved the standings, frozen at 08:00 on disembarkation morning. The app’s final share card survived too. Those records let me look back at ten days of live play.

Firestore’s day buckets credit the embark tutorial with 27 bingos from seven players. None landed on embark day itself; the easy card kept paying out for days.

The player rows show ten people with an event-wide bingo, and PostHog independently counts ten. The scoring tests explicitly include embark bingos in the event total. Seven people with an embark bingo fits inside ten across the event.

My read is that the group adopted the scoreboard. Of 499 sessions across the ten days, 384, or 77%, contained no mark. Checking standings and reading the feed would explain those sessions, but the event log doesn’t measure what people did instead.

The social mechanics had little use. One player gave all nineteen hearts. One player made all seven proof demands. Three of sixteen players used seven reshuffles. The possible ceiling of 48 is arithmetic, sixteen players times a cap of three; it isn’t a recorded total.

Marking reached most of the group. Fourteen of sixteen players marked at least one square in the frozen Firestore. PostHog sees fewer marks and twelve markers, nine of whom marked on five or more days. The frozen event totals 845 squares and 61 bingos, with zero blackouts.

Zacaria Arab won with 16 bingos across 124 squares. Logan Murdock marked more squares, 129 to 124, and finished second. The leaderboard sorts bingos first. The rule, not volume, crowned the champion.

The app's own final-standings share card for Day 10, Barcelona, with champion Zacaria Arab at 16 bingos and 124 squares, Turntilla's cruise-wide First to BINGO, and the ten daily honors spread from Trieste to Barcelona.
Figure 1: The app's own final-standings share card for Day 10, Barcelona, with champion Zacaria Arab at 16 bingos and 124 squares, Turntilla's cruise-wide First to BINGO, and the ten daily honors spread from Trieste to Barcelona.
  1. Expected

    I expected a fresh themed card each day to keep the group marking throughout the cruise. Marking was the core loop.

    Observed

    384 of 499 sessions, 77%, contained no mark. The two cards after the embark tutorial produced one bingo apiece, both claimed on the last full day. The event log records the absence of marks, not why people opened the app.

    Response

    On the morning the ship docked, I put the work into a final-standings share card, committed at 10:47. Three more features landed by evening, so it wasn't the day's last feature. It was the one built to keep the record rather than extend play.

  2. Expected

    I expected the game's two independent records to agree.

    Observed

    Firestore holds 845 squares; analytics logged 783 marks over the same window. The bingo counts agree: Firestore's player rows show ten players with an event-wide bingo, seven of them with an embark bingo. PostHog independently counts ten. Embark bingos count toward the event total.

    Response

    I kept both square counts on the page. Each stays attributed to the system that recorded it, and I compare only matching windows. The records don't let me reconcile the disagreement.

  3. Expected

    I treated content as a launch risk worth preparing for. Server-side report auto-hide, an admin roster and a flag-gated Vision check were finished six days before embarkation.

    Observed

    The operational problems were authentication and the PWA update path. The sailing's commits skew toward both. The repository keeps no record of the moderation stack being used during the sailing.

    Response

    After the cruise, the platform work went into authentication: a centralized auth origin with a single-use handoff, implemented, provisioned by 2026-09-10, and live in the client as of 2026-10-06.

Running it at sea

Ninety-six changes landed on main between embarkation and disembarkation. That’s the first-parent commit count between the Rome midnights that bound the sailing.

Those commits don’t prove that every change reached a phone at sea. No deploy log survives in the repository. Firebase Hosting’s release history records 41 deploys between the same midnights, and an incident note records one mid-sailing rules deploy and two days of stale-shell fallout.

The reshuffle gives me evidence that a feature drop reached players. It didn’t exist until the sea day, and all seven uses were recorded during the sailing. That build reached people while the game was live. I don’t have a breakdown of the ninety-six changes into fixes and features, so calling most of them firefighting would be a guess. The log does show a skew toward authentication and the PWA update path.

Agents expanded my build, test and review capacity against the eight-day deadline, then while I ran the game at sea. The specs pin the choreographed finale’s exact instants. ADRs record the decisions that were hard to reverse: on-device share rendering, offline persistence, and a second Firebase project as the interim tenant boundary. Tests turn acceptance criteria into checks, including contrast computed from the shipped CSS across every theme and a launch checklist that pins embarkation day. The mid-cruise drops went through reviewed pull requests.

I owned the product calls: the trust model, the sea-day changes and the held cutover. Those were the part I couldn’t delegate.

From one event to a platform

The engine did get reused. Gay Cruise Bingo remains the original and default live edition, on its own domain and Firebase project. Vacay Bingo is the travel edition. Five Across is the platform under both.

The comment on Gay Cruise Bingo’s brand record is explicit: “It is an ENDORSEMENT, not a rename: the wordmark, the cruise vocabulary, the adult posture, gaycruisebingo.com and the legacy Firebase project are all unchanged—GCB is simply one Edition of Five Across now, on the same engine as Vacay…”

A different host ran the first non-cruise event on the same engine: a Sonoma Coast weekend in August 2026. Unlock-copy fixes landed on opening day. The software was reused, but participation stalled. Four people marked 27 squares, nobody got a bingo, and the guests’ last marks came on Saturday morning.

On the cruise, the software had prompts written for that sailing, a group ritual, and a host promoting and tuning the game live. Moving the engine to another occasion didn’t reproduce that combination. Two events don’t isolate which contribution mattered most.

The themes generalized too. Fifteen shipped for the cruise, thirteen party themes and two tutorials, out of twenty-two across the platform. Every one is held to WCAG AA contrast by tests that compute it from the CSS itself.

The cutover is still held. The wildcard event router has written routes that haven’t been attached; attaching them is a human step. Centralized authentication with a single-use handoff was live in the client as of 2026-10-06, but no sign-in through it has been demonstrated because registered hosts still sign in directly. Until tenant isolation ships, separate events run on separate Firebase projects and unrelated groups don’t share a backend.

What a friend-group launch proves

Sixteen friends can’t validate retention, acquisition, pricing or how strangers treat each other in a game. Those questions need an audience this launch didn’t have.

What it could test, it tested harshly. The launch did test the social mechanics on the people I’d designed them for. Hearts and doubts failed with this group, and I have counts for that. The honor model let the standings run: 845 squares stood. It didn’t make those marks independently verifiable.

Offline-first’s effect remains unresolved. Over the scoring window, Firestore holds 62 more marks than analytics logged. Offline queueing and ordinary analytics loss could both explain the gap. PostHog captured fifteen exception events containing auth/network-request-failed. They establish request failures, not successful offline marking or an effect on participation.

I also got evidence that one person and a fleet of agents could operate a live product at sea. Ninety-six changes landed, and a feature drop reached phones during the sailing.

My friends were a hard audience to delight with the mechanics and a forgiving audience for everything else. I need to keep those two tests separate.

How this was written

At my direction, Claude wrote the first draft from the game's code, analytics, and event records, while I determined the story, its framing, and the visual evidence. I confirmed that the finale's freeze hadn't moved during the cruise, which let Claude correct its timing, and when Codex rewrote the piece in my voice, I requested three precision fixes and had two earlier lines restored because they made the point better. I approved the result and reviewed Claude's later card-summary revision before it went live. More about how I write