Both players run from a video, and the middle chip finally plays you

Six sessions of asking an image model for a row of running figures, and it
never once drew a gait: three phases repeated with the same leg leading,
"ropeskipping, not running". A frame-by-frame LEFT/RIGHT skeleton spec, a
closing checklist, a loop clause, six figures instead of eight — all lost.
One 5-second image-to-video clip has no such problem, because it moves a
body instead of drawing poses. ~$0.42 a character through the key already in
.env, pose pinned at BOTH ends (pinned only at the front, she rotated into
profile halfway), stride period MEASURED off the frames rather than sampled
by eye — hers repeats in 6 clip frames, his in 23 — and the phases cut and
palette-snapped through _stride.cut. assets/_vidroll.py films and cuts it;
her first frame came from one still roll that took his approved pose as a
picture instead of a paragraph.

Wired: runN probes the sheet through SK() so the women's tour counts HER
frames, the player is no longer clamped to a two-frame pair, runHz 2.4 ->
1.9 ("this looks like roadrunner"), and the bench grew two PLAYER stops —
a cycle only ever seen mid-rally is a cycle nobody has watched.

Also in: the difficulty retune (medium is a real step above the shipped CPU
and is where `smart` first appears; hard stepped back off superhuman; easy
closed its gap) with MEDIUM's dial-for-dial pin to TUNE removed from the
test, and Oliver's own transparency fixes on the male frames.

SPRV 138 — the frames were right on disk and wrong on screen for an hour
because assets are cached for a year unless the ?v= moves. main.js v269.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
wuwiandClaude Opus 5 committed 2026-08-15 17:53:58 +02:00
1 parent 2b7eb04e19
commit 1b7fc2cdcf
23 files changed
+352 -71

No files matched your search

+15 -10
View File
@@ -46,8 +46,9 @@ export const TUNE = {
// with Oliver 2026-08-14: 4.5 on six frames was strobe, 1.7 on two read as jogging,
// and 3.4 with the stride-synced bob is the one he called "way better". (6.8 was
// tried and reverted on sight — the ceiling is real.)
runHz: 2.4, // stride cycles/s for the rolled 6/8-frame run sets. The old hardcoded 4.5 was
// tuned on 2-4 near-duplicate frames; on eight real phases it strobed
runHz: 1.9, // stride cycles/s for the rolled 6/8-frame run sets. The old hardcoded 4.5 was
// tuned on 2-4 near-duplicate frames; on eight real phases it strobed — and so
// did 2.4 on the back-view re-roll (Oliver: "this looks like roadrunner")
faceFlip: 0.13, // s an auto-steered direction must persist before the body mirrors (earnFlip).
// The rolled cycles hold a real racket, and an un-earned flip teleports it
// across the body — the "racket flashing left and right" report
@@ -710,12 +711,16 @@ export const MENU = [
// CPU pre-plans its intercept, and the dials that actually read as skill were untouched. What a
// player FEELS is: how fast the CPU reads a shot (reaction — "well-aimed corners are winners"),
// how long its racket is (reach), the FLOOR of its shot quality (qMin — a raised floor means every
// ball comes back with pace), how often it just misses (flub), and its serve. MEDIUM repeats the
// shipped TUNE values dial for dial (pinned by test), so the middle chip IS the old game.
// ball comes back with pace), how often it just misses (flub), and its serve.
// `smart` is the probability the CPU aims at the OPEN court instead of a random lane (trySwing) —
// the dial that turns "returns everything" into "plays you". HARD got a second pass after
// Oliver's "still too easy": it now reads your shot in 0.22 s, covers court at 1.3x, floors its
// quality at 0.70, and hunts the corner you are not standing in four times out of five.
// the dial that turns "returns everything" into "plays you".
// THE 2026-08-15 RETUNE (Oliver: "medium is too easy and hard is a tad too hard, easy could also be
// a little harder"): the three chips were bunched at the bottom and then jumped. MEDIUM no longer
// repeats the shipped TUNE values — it is a real step ABOVE them, and it is where `smart` first
// appears (0.35), so the middle chip plays the court instead of just returning. HARD stepped back
// on the dials that made it feel superhuman rather than good — speed 1.5 -> 1.38, floor 0.70 ->
// 0.66, smart 0.8 -> 0.7 — and EASY closed most of its gap to the old MEDIUM. The order across all
// three still climbs on every dial (pinned by test); only MEDIUM's old pin to TUNE is gone.
// `recover` scales the between-shots jog (the rally block's 0.5 "slow recovery: wrong-footing wins
// points") — a HARD CPU that ambles back to centre is free points, a recovering one is a wall.
// Speed calibration: the zombie RUNNER covers 2.8 x 1.8 x 1.22 = 6.15 m/s with no reaction gate,
@@ -723,9 +728,9 @@ export const MENU = [
export const DIFFS = [
// EASY's first cut (flub 0.30, quality cap 0.62, reaction 0.62) was "way too easy" once HARD
// taught us which dials bite — it is a mild step below MEDIUM now, not a training dummy
{ id: 'easy', name: 'EASY', flub: 0.18, quality: 0.76, qMin: 0.45, speed: 0.90, reaction: 0.54, reach: 1.52, serve: 0.9, smart: 0, recover: 0.45 },
{ id: 'medium', name: 'MEDIUM', flub: 0.10, quality: 0.86, qMin: 0.45, speed: 1.0, reaction: 0.45, reach: 1.6, serve: 1.0, smart: 0, recover: 0.5 },
{ id: 'hard', name: 'HARD', flub: 0.01, quality: 1.0, qMin: 0.70, speed: 1.50, reaction: 0.22, reach: 1.85, serve: 1.15, smart: 0.8, recover: 0.85 },
{ id: 'easy', name: 'EASY', flub: 0.15, quality: 0.80, qMin: 0.45, speed: 0.95, reaction: 0.50, reach: 1.55, serve: 0.95, smart: 0, recover: 0.48 },
{ id: 'medium', name: 'MEDIUM', flub: 0.06, quality: 0.92, qMin: 0.56, speed: 1.18, reaction: 0.34, reach: 1.70, serve: 1.05, smart: 0.35, recover: 0.66 },
{ id: 'hard', name: 'HARD', flub: 0.02, quality: 1.0, qMin: 0.66, speed: 1.38, reaction: 0.27, reach: 1.80, serve: 1.15, smart: 0.7, recover: 0.80 },
];
// THE TOURNAMENT (Oliver's pick over the endless ladder, 2026-08-13): four rounds up the club's