Queueglass

SIMULATED · LOCAL · SEEDED · REPLAYABLE

A small, deterministic queue simulator. Send work through four steps, change the traffic, and see where it gets stuck. All data is generated locally by an educational model.

Truth boundary

Every displayed value is generated in the browser from a seed and one of three toy scenarios. The repository contains:

• no production telemetry or network client;
• no identity, account, communication, monetary, commercial, or staffing data;
• no AI inference or third-party services;
• no measured throughput, latency, accuracy, uptime, cost, savings, or service-level claims.

Item counts are synthetic. Capacity and compute are arbitrary model units. Latency is expressed only as simulation ticks. Outputs are not benchmarks, forecasts, architecture recommendations, or evidence of real operational performance.

What can be explored

• Steady traffic (nominal) — 1–3 generated arrivals per tick.
• Sudden rush (burst) — extra arrivals at ticks 6–13, then a return to baseline traffic. Queue recovery is not guaranteed.
• Slower checkpoint (policy_degraded) — reduced Check capacity at ticks 7–19, with seeded retry decisions.
• Run, pause, or step — playback controls stay beside the flow. Select a step to inspect its capacity, movement, and waiting work.
• Explore the run — expand the pressure chart, counts, decision trail, ten-tick stepping, and playback speed.
• Replay settings — apply or derive a seed. Shared ?seed=&scenario= links reproduce the run from tick 0.

The interface uses Receive → Sort → Check → Finish for the model's intake, classify, policy, and dispatch stages. Items may traverse all four steps in one tick; queues show work remaining after processing.

The conservation invariant is explicit:

synthetic arrivals = completed items + items still in the model

Run locally

Requires Node.js 24 or a compatible Node release supported by Next.js 16.

npm ci
npm run dev

Open http://localhost:3000/?seed=QUEUEGLASS-7&scenario=burst.

Verify

npm run verify

The verification gate scans for removed claim language and likely secrets, lints, type-checks, runs six deterministic/invariant tests, and creates a production build.

For the real-browser smoke, start the built app and run:

npm run build
npm run start -- -p 4173
# in another terminal
npm run test:browser

The browser smoke checks seed/scenario replay, every intermediate tick in batched steps, the rolling 80-tick history, reset, playback, keyboard focus, stage inspection, fullscreen/Escape, mobile targets, and browser errors. It refreshes the real capture in docs/.

Automation hooks

• window.render_game_to_text() returns concise JSON containing the truth boundary, coordinate system, seed, scenario, tick, metrics, stages, current synthetic work, recent model decisions, and up to 80 history points. History records { tick, depth, arrived, completed }; arrivals and completions are per-tick deltas.
• window.advanceTime(ms) advances fixed 600 ms simulation ticks without using wall-clock values in model state.

Architecture

src/lib/simulation.js          pure seeded state transitions
src/components/SimulatorLab   controls, tick history, inspection, state hooks
src/components/Topology       accessible responsive stage flow
tests/simulation.test.js       replay and conservation invariants
scripts/browser-smoke.mjs      asserted Chromium interaction/capture
scripts/*-scan.mjs             release truth and secret gates

Provenance

This public-facing lab is a clean-room reframing of a contained local interface prototype. The staged revival removed its original theatrical data generator, timer-driven metrics, unsupported operational language, named external connectors, and identity-like examples. No adjacent company documents, dashboards, notes, screenshots, workflows, or datasets were inspected or copied into this project.

Limitations

• The queue model is intentionally small and single-process.
• Capacities, priorities, retries, and compute costs are pedagogical parameters.
• Scenario behavior is deterministic, not statistically calibrated.
• Browser automation currently targets Chromium.
• The project must not be presented as evidence about any company, product, team, deployment, or operating environment.

MIT licensed. Contributions that preserve the truth boundary are welcome; see CONTRIBUTING.md.