Scripted Clocks

The three scripted clocks ignore the host timestamp and supply their deltas from a constant, a repeating list, or a seeded draw. Under engine.advance a frame therefore costs exactly what the clock says and no real time is involved, so a validator chooses the shape of time a scenario runs under.

Every example below uses the harness and the debug surface from Validating a Game. The harness takes the clock as an option and installs it at construction, and the surface is read off engine.debug.

The three clocks

Clock	Each frame is worth
`ConstantClock(stepMs)`	`stepMs`, every frame
`SequenceClock(stepsMs)`	the next entry, repeating from the start
`JitterClock(minMs, maxMs, seed)`	a draw from the range, as a function of the seed and the frame index

A seeded draw replays exactly, so two engines given the same seed take the same deltas in the same order and a failure found under jitter can be run again.

The deltas each clock delivers

engine.frame().lastDeltaMs reports the step the current frame was worth, so a check reads the clock’s own output back off the engine. The three assertions below state the whole of what each scripted clock guarantees: one step held, a list repeated, and a draw kept inside its range.

import {
  ConstantClock,
  JitterClock,
  SequenceClock,
  type Clock,
} from "@clockwyrks/structured-3d";
import { expect, it } from "vitest";
import { createHarness } from "./harness";

const PATTERN = [4, 4, 4, 4, 33, 16];

async function deltas(clock: Clock, frames: number): Promise<number[]> {
  const harness = await createHarness({ clock });
  const steps: number[] = [];
  for (let i = 0; i < frames; i += 1) {
    await harness.engine.advance(1);
    steps.push(harness.engine.frame().lastDeltaMs);
  }
  harness.dispose();
  return steps;
}

it("holds one step under a constant clock", async () => {
  const stepMs = 1000 / 240;

  expect(await deltas(new ConstantClock(stepMs), 4)).toEqual([
    stepMs,
    stepMs,
    stepMs,
    stepMs,
  ]);
});

it("repeats the sequence in order", async () => {
  const steps = await deltas(new SequenceClock(PATTERN), PATTERN.length * 2);

  expect(steps).toEqual([...PATTERN, ...PATTERN]);
});

it("draws every jittered delta from its range", async () => {
  const steps = await deltas(new JitterClock(4, 40, 20260819), 200);

  expect(Math.min(...steps)).toBeGreaterThanOrEqual(4);
  expect(Math.max(...steps)).toBeLessThanOrEqual(40);
  expect(new Set(steps).size).toBeGreaterThan(1);
});

One cycle of the pattern is 65 milliseconds over six frames, which is how a step six times the usual size is put in front of a build at a known frame, every run. The jitter check states the bounds the clock guarantees and the distinct count states that the run saw more than one step size.

Advancing by duration

A frame count fixes a duration only under a constant step. Where the step varies, a validator advances one frame at a time until the simulated clock reaches the target.

import type { Engine } from "@clockwyrks/structured-3d";

export async function advanceMs(engine: Engine, ms: number): Promise<void> {
  const target = engine.frame().timeMs + ms;
  while (engine.frame().timeMs < target) {
    await engine.advance(1);
  }
}

The loop overshoots the target by at most one step, which is the figure a check allows for when it asserts against elapsed simulated time.

Replacing the clock mid-scenario

engine.setClock swaps the clock in place, and the next frame takes its delta from the new one. The frame counter and the accumulated simulated time carry over, so posing a scenario under an exact step and then running it under jitter is one scenario rather than two.

const harness = await createHarness({ clock: new ConstantClock(1000 / 120) });
const { engine } = harness;
engine.debug.placeRunner({ x: -3, y: 0, z: 0 });
await engine.advance(60);

engine.setClock(new JitterClock(4, 40, 20260819));
harness.hold("right");
await advanceMs(engine, 2500);

One scenario, three step sizes

The scenario reduces the world to two orbs, puts one five units down a clear lane and the other clear of the runner’s path, places the runner at the near end, and holds the right action for two seconds of simulated time. A correct build collects the orb once, scores it, and finishes eight units along.

import {
  ConstantClock,
  JitterClock,
  SequenceClock,
  type Clock,
} from "@clockwyrks/structured-3d";
import { expect, it } from "vitest";
import { ORB_POINTS, TAGS } from "../src/constants";
import { advanceMs } from "./advance-ms";
import { createHarness } from "./harness";

const START = { x: -3, y: 0, z: 0 };
const TARGET = { x: 2, y: 0, z: 0 };
const PARKED = { x: 6, y: 0, z: -3 };
const TRAVEL_MS = 2000;
const MAX_STEP_MS = 40;

interface Outcome {
  frames: number;
  travelMs: number;
  collects: number;
  collectedAtMs: number;
  score: number;
  remaining: number;
  x: number;
}

async function runScenario(clock: Clock): Promise<Outcome> {
  const harness = await createHarness({ clock });
  const { engine } = harness;
  const world = harness.world();

  engine.debug.keepOrbs(2);
  await engine.advance(1);
  engine.debug.placeOrb(0, TARGET);
  engine.debug.placeOrb(1, PARKED);
  engine.debug.placeRunner(START);

  const collected: number[] = [];
  engine.events.on("cue:played", ({ cue, t }) => {
    if (cue === "collect") collected.push(t);
  });

  const startMs = engine.frame().timeMs;
  const startFrames = engine.frame().count;
  harness.hold("right");
  await advanceMs(engine, TRAVEL_MS);
  harness.release("right");

  const snapshot = engine.debug.snapshot();
  const outcome: Outcome = {
    frames: engine.frame().count - startFrames,
    travelMs: engine.frame().timeMs - startMs,
    collects: collected.length,
    collectedAtMs: (collected[0] ?? Number.NaN) - startMs,
    score: snapshot.score,
    remaining: world.byTag(TAGS.orb).length,
    x: snapshot.runner.x,
  };
  harness.dispose();
  return outcome;
}

it("reaches the same outcome under every step size", async () => {
  const clocks: Clock[] = [
    new ConstantClock(1000 / 240),
    new SequenceClock([4, 4, 4, 4, 33, 16]),
    new JitterClock(4, MAX_STEP_MS, 20260819),
  ];

  const outcomes: Outcome[] = [];
  for (const clock of clocks) {
    outcomes.push(await runScenario(clock));
  }

  for (const outcome of outcomes) {
    expect(outcome.collects).toBe(1);
    expect(outcome.score).toBe(ORB_POINTS);
    expect(outcome.remaining).toBe(1);
    expect(outcome.collectedAtMs).toBeGreaterThanOrEqual(1049);
    expect(outcome.collectedAtMs).toBeLessThan(1051 + MAX_STEP_MS);
    expect(outcome.travelMs).toBeGreaterThanOrEqual(TRAVEL_MS);
    expect(outcome.travelMs).toBeLessThan(TRAVEL_MS + MAX_STEP_MS);
    expect(Math.abs(outcome.x - 5)).toBeLessThan(0.2);
  }

  expect(new Set(outcomes.map((outcome) => outcome.frames)).size).toBe(3);
});

it("replays exactly under the same seed", async () => {
  const first = await runScenario(new JitterClock(4, MAX_STEP_MS, 20260819));
  const second = await runScenario(new JitterClock(4, MAX_STEP_MS, 20260819));

  expect(second).toEqual(first);
});

The lane is five units long and the runner walks it at four units per second. The orb is collected once the gap between the two closes to the sum of their radii, which is 0.8 units, so the ideal collection instant is 4.2 units of travel at 1050 milliseconds. Two seconds of travel put the runner eight units along, at x = 5.

The three clocks reach that outcome over roughly 480, 185, and 91 frames. The final assertion states that the three frame counts differ, so the check is known to have run at three step sizes rather than three times at one.

What survives a change in step size

These are the figures a validator asserts on, because a build that integrates against the delta it is given produces them whatever the clock delivered.

Figure	Why it holds
The event count	One overlap begins once, whichever frame finds it
The score and the orbs left	Both follow from the event count
The order of events	The frame order is fixed, so a collection precedes what it caused
Elapsed simulated time	`advanceMs` stops at the target, overshooting by at most one step
A position, as a band	The travel is the same, and the band absorbs one step of it

A position compared for equality across two step sizes is a check of the step size rather than of the build. The band is one step of travel at the widest step the run allows, which is MAX_STEP_MS at the walking speed, or 0.16 units. The check allows 0.2.

What legitimately diverges

Two correct runs of one scenario differ in ways a validator has to leave room for.

The frame count differs, and with it every figure derived from it. Nothing the specification fixes is stated in frames, so a check states a duration and converts it through the clock rather than asserting a count.

The final position differs by up to one step of travel at either end. The last frame overshoots the target duration, and the frame that crossed a threshold did so from a different distance away.

An event’s timestamp differs by up to one step. An overlap is found on the first frame the two shapes are inside each other, and a larger step arrives at that state later and deeper. A check therefore asserts the count exactly and the instant as a band.

The manifold depth differs. A blocking pair is separated by however far this frame’s movement drove it in, so a larger step produces a deeper penetration and a stronger correction. A check reads the normal, which is a direction, rather than the depth, which is a step size.

Accumulated floating-point error differs. A thousand additions of 4.17 milliseconds and a hundred additions of 22 land on different values, so a check of simulated time uses toBeCloseTo rather than toBe.

The outcome of a scenario run twice under the same clock is identical. Each scripted clock is fully determined, so the whole run is, and two JitterClock instances built from one seed produce identical outcome objects.