The last detail pass: read everything, and say what is still worth doing #20

This is the closing pass. val is being shown to people now, so go over it as a whole and use your judgement: what, out of everything below or anything you find yourself, actually makes it better — and what should be left alone.

You have licence to say "nothing here is worth changing" for any of it. A short pull that fixes two real things beats a long one that touches ten.

What has been said about this app, in one place

Settled, do not undo: the server decides everything; a client is only sent what it can see (a wallhack was closed twice — first by making vision honest, then by capping a playing seat per address and giving extra connections a positionless spectator view); the arena freezes when nobody is connected; bots have a reaction time and an aim cone that settles; the wall glide is shared by client and server; your own body is predicted and never drawn with the treatments meant for other people; flashes are room-sized and blind only what was on screen.

Named but never done:

• No overtime at 12–12 — the match simply ends. Decide whether that is a gap or a choice, and say which.
• Bots commit to a site for the round rather than rotating on sound or on a team mate dying.
• The snapshot is ~3.2KB thirty times a second, and most of it is the board — ten names, agents and tallies resent every tick though they change once a round. Wren flagged an easy halving here.
• server.test.ts still shares one live arena with the client test's open socket, so its spike-plant and reload-seat tests can race the arena timer about once in a dozen runs. The deploy gate is still a coin flip because of it.

Asked for explicitly by the person playing it: it should feel smooth; it should stay recognisably Valorant in its rules and vocabulary; it can look as simple as it needs to — stick figures are fine if they buy frame rate; and it is for showing to people, so the first ten seconds matter more than anything deep in a match.

How to judge what is worth doing

The measure is somebody sitting down in front of it for two minutes with no explanation. What confuses them, what looks broken, what makes them say "oh, that is clever" — those are your targets, in that order. A thing that only a player of forty rounds would notice is not this pass.

Look especially at the parts nobody has examined: the buy menu as a first-time reader meets it, whether the agents read as different from each other in play rather than in a table, what the end of a match looks like, what a spectator sees, and whether the HUD says the right thing at the moment a round turns.

The bar

Same as always: nothing sent that the viewer should not have, numbers for anything you claim about feel, a test for every behaviour you change, and the app playable at every commit. If you find something broken that you cannot fix inside this pass, say so in the pull plainly rather than fixing it badly.

End the pull with your own honest read: what you would do next if there were another pass, and what you would leave exactly as it is.

Done in #21 (merged 2026-09); closing — that merge predates closing-on-merge.