Repository navigation
Mythic Masters: friends walk the open world together, cheap by design, hill climb to 8/10, rounds 21+ (from #4081) #4165
Description
Score: 4.5/10 (target 8, on the owner's scale: agent judges halve their number, per the 2026-11-05 reset)
Carries the climb on from #4081 (round 20, PR #4164: the room keeps the party's contracts for a late joiner, at the furthest sign anyone read, and drops one when its lair's creature is beaten). Round 20 gained nothing: the judge gave 9.0 to main and 9.0 to the change, so this is one round without a gain. Before that came #4063 (round 19, PR #4078: one health bar at a lair), #3862 (rounds 17-18), #3792 and the rounds listed there. Their score comments have the judges' tables. The owner's brief and the five points are in #3862's body.
Next steps, in order:
- One fight, not one health bar. Friends at one lair still each play their own fight. That needs the 2v1 sim in Mythic Masters: co-op PvE, two friends against a wild one (needs a 2v1 sim) #3929 (teams in
sim.tsandsim-rs), then lockstep for N seats (src/net/duel.ts) or one page hosting. Mythic Masters: co-op PvE, two friends against a wild one (needs a 2v1 sim) #3929 sizes it at a day or two: take it as its own piece of work and score it here once it lands. - A full party's frame cost (Mythic Masters walk together: a full party costs about 13 ms a frame (15 rigged walkers), draw far friends cheaper #3704, with the frame-budget climb Mythic Masters: open-world frame budget (95th under 16.7 ms, no hitch, a party of six), hill climb to 8/10, rounds 6+ (from #3709) #3796). The round 20 judge ranks "walk together" as the lowest point: about 20.6 ms a frame at six, and CI runs the journey with
?draw=0, so fps is checked only on a machine with a GPU. - A contract's sign believed only where it is. The room checks that the teller is by the board (
huntNear), not which sign they read, so a page there can claim sign 3, and since Mythic Masters together round 20: the room keeps the party's contracts for a late joiner (Part of #4081) #4164 the room keeps that claim. The clue spots are insrc/world/hunts.ts, which the room can't import, so the fix is a small table of clue spots (likedens.json), checked against the pose. - A share tied to the fight. The room believes a lair share no faster than a whole creature in 3 s (party.ts
believeHurt). A tighter cap would need the line-up's levels, which the room never hears, or the fight hashes the duel uses. - The region's mythic kept by the room for a late joiner, the way the wilds and now the contracts are.
Open: MANUAL RUN #3697 (the owner's walk with a friend), #3858 (journey timing), #3706 (the Region under workerd locally), #3704 (full-party frame time), #3929 (2v1 sim).
Activity
QuinnsCode commentedon Nov 10, 2026
Score: 4.5/10 (target 8)
Round 20, PR #4164. The room now keeps the contracts the party is on (at most 3, each at the furthest sign anyone read) next to the wilds it already keeps (#3859). They go out with
all, so a friend who joins after the teller has left can still follow a contract to its lair. Beating the lair's creature drops the contract from the room.Why this step and not the first two: step 1 (one shared fight) needs the 2v1 sim in #3929. That is the TypeScript sim and its Rust twin, which #3929 sizes at a day or two, too big for one round. Step 2 (a share tied to the fight) needs a cap from the line-up's own damage per second. The room never hears the line-up's levels, so any cap it could set would still be loose.
A fresh judge scored main 9.0 and this change 9.0 raw. The change counts for about +0.05 on "one world", which doesn't make a quarter point. So the score stays at 4.5 + 0 / 2 = 4.5. This is the first round with no gain. The judge also flagged a stale contract after its lair is won; that fix came after the judging, so the score doesn't count it.
?draw=0skips the fps check.bun scripts/cost.ts 6 60), before and now. Target: under $0.005.Next: one shared fight (the 2v1 sim, #3929), then a full party's frame cost (#3704 / #3796), which the judge ranks as the lowest point.