Bound the lobby, and pay everyone who never got a game (C&WG)

[htsukamoto5]

Closes #6.

Stacked on #17. Review order is #14 → #16 → #17 → this. It reuses #17's exit-screen shape and #16's Pipeline.flush(), which is why it is small.

Two ways a participant could reach a dead end where they could not submit, and therefore could not be paid.

1. The lobby never ended

jsPsychMultiplayerSync's timeout default is null, so wait() was called unbounded. Someone who arrived and was never matched sat on that screen forever — no timeout, no exit, no completion code, no way to be paid for time they had already given up. They close the tab and open a support ticket.

Now bounded by CONFIG.LOBBY_TIMEOUT_MS.

One detail worth checking in review, verified against the bundle rather than assumed: on timeout the sync trial resolves with timed_out: true — it does not abort.

try { const t = yield l.wait(e.wait_for, s); yield o(), y(t, !1, null) }
catch (t) { … e.on_timeout(t), yield o(), y(a(), !0, t.message) }

So without a guard the timeline walks straight into roleTrial, which waits on its own ready predicate — swapping an unbounded lobby for an unbounded pairing screen one step later, which would have looked fixed while being the same bug. pairingPhase gates that.

2. The spectator was shown the door with no code

The overflow path showed "This game is already full" for four seconds, called disconnect(), and ended with no completion code. Someone who arrived on time and was turned away for reasons entirely outside their control could not submit.

It now routes to the same paid exit as a lobby timeout, because from the participant's side it is the same event: they showed up, and we could not give them a game.

Both routes share one completion code — Prolific configures a single code per exit — and stay separable in the data via no_match_reason (A7). That matters: the lobby-timeout rate and the odd-arrival rate answer different questions, and #10's waiting room is sized off both.

The exit screen deliberately does not set trial_duration. The old spectator screen auto-dismissed after four seconds, which is fine for a screen that says nothing and fatal for one carrying a completion code.

Lobby copy

• A live countdown, because an indefinite-feeling wait is what makes people abandon. The plugin renders message once and never updates it, so the ticker is driven from on_load and cleared in on_finish.
• Plain statement that an unmatched participant is still paid, and how much.

The amount comes from CONFIG.NO_MATCH_PAYMENT_USD and is templated into every screen quoting it. Participant-facing copy drifting from what is actually configured is on the launch checklist, so there is now a test that fails if any screen hardcodes a dollar figure.

The two publish-locked decisions here

Both are suggestions, both are named constants with their reasoning inline, because neither can be changed once the study is published:

• B6 — LOBBY_TIMEOUT_MS, 5 min. Not merely a UX number: it sets the advertised study duration (B5), which must include the expected wait. Refine from observed arrival rate in the pilot before publishing.
• B2 — NO_MATCH_PAYMENT_USD, $1.25 (full task rate for the wait), not Prolific's $0.14/min floor. A floor-rate payment loses to returning the study and taking a short survey, which teaches people to abandon the lobby at exactly the moment we need them to stay. Costs ~$10–15 across a whole run.

Testing

node tests/exits.test.mjs — 22 checks over the real conditional_function bodies, extracted from the file rather than restated in the test.

The property under test is that every reachable end state lands on exactly one terminal screen. Two means the participant sees a contradiction; zero means they cannot be paid. This file has shipped both, so there are explicit regressions for each: a spectator reaching "Game complete!" with the complete code, and a dropout survivor doing the same.

Covers all six end states (director/matcher completing, either partner dropping, lobby timeout, spectator overflow), that pairing is skipped after a timeout, and the config sanity checks above.

tests/dropout.test.mjs 16/16 and tests/pipeline.test.mjs 39/39 still pass.

Not in here

• Ghost pairing. Entries are only removed on an explicit disconnect() — no heartbeat, no beforeunload in any bundle — so a lobby tab closed mid-wait leaves an entry that can satisfy wait_for and produce a match with nobody. That mainly bites on Firebase and is tracked with Pairing: build a rolling waiting room (room bucketing, ghost expiry, odd arrivals) #10.
• The real rolling waiting room (Pairing: build a rolling waiting room (room bucketing, ghost expiry, odd arrivals) #10), still blocked upstream. This bounds the wait; it does not match strangers.
• no_match is still an empty placeholder code until Create the DataPipe experiment and a draft Prolific study — critical path #15, so this path currently shows the configuration-error screen rather than a code, and warns at startup.

🤖 Generated with Claude Code