0.4.0rc1 — Predict, run, explain

Predict. Run. Explain.

How long does 100 kWh support a 1 MW load after the utility fails? In the charging-disabled reference case, 0.90 battery efficiency and 0.95 distribution efficiency give 307.8 seconds of ride-through. Halving reserve gives 153.9 seconds. This release makes that question the first browser experience, then lets you follow the full energy ledger.

Try the five-minute guide · Twelve interactive lessons · Research evidence

What changed

• A focused predict → run → explain guide, with a before/outage/recovery diagram and a large, unit-labelled result.
• Separate learning, advanced and evidence views. Eighteen presets, four facility profiles and all twelve color-themed interactive lessons remain available.
• One-parameter comparisons for battery reserve, generator delay and surviving-path capacity; complete result exports.
• Three versioned canonical results with input hashes, source-commit links, Markdown/HTML reports and independently reconstructed interval ledgers.
• A real captioned one-minute demonstration, annotated screenshots, measured benchmark records, contributor extension guidance and focused keyboard/mobile/contrast/reduced-motion checks.

The numerical continuity and planning algorithms are unchanged. The software version and deterministic run identifiers advance to 0.4.0rc1; record the version and full input hash when comparing runs.

Exact artifact and checks

Source: f83c0a3277b01f275a5961f38af284594453c683. Successful CI and Pages publication. Four Windows/Linux × Python 3.12/3.14 jobs, 125 full-suite tests (two Windows privilege skips), installed-wheel verification, eight API-dashboard journeys, eight JavaScript/native engine tests and 30 static browser tests passed. Optional browser-Python verification and all 24 lesson starting/challenge results are covered.

Download the source, wheel, three-case evidence packet, environment-specific CI evidence and exact tested browser bundle below. Start with REPRODUCIBILITY.md. SHA256SUMS.txt hashes every asset; BROWSER-SHA256SUMS.txt hashes every file in the browser archive. The browser archive has a datacenter-twin-lab/ directory: serve its parent on loopback and open that path. The first result uses JavaScript; optional Python verification downloads its runtime on demand.

git clone https://github.com/mohammadrezwankhan/datacenter-twin-lab.git
cd datacenter-twin-lab
git checkout v0.4.0rc1
python scripts/build_evidence.py --output outputs/reproduction --revision f83c0a3277b01f275a5961f38af284594453c683 --repository mohammadrezwankhan/datacenter-twin-lab
python scripts/build_evidence.py --verify outputs/reproduction

Why a release candidate

This is a reproducible synthetic continuity laboratory for datacenter power systems, for learning and research. It is not a production reliability model or a calibrated facility twin. It does not model AC switching transients, protection, cooling, GPU workload performance, safety certification or physical control. Replay timing is an animation, not equipment response.

Independent external reproduction has not yet been obtained. Maintainer tests, ledger calculations and CI are not an independent endorsement. Use the existing review discussion and the attached blank reviewer template to report a reproducible mismatch or scoped agreement. Stable promotion and a possible archive DOI follow the documented gates.

Source and original synthetic fixtures are Apache-2.0, with applicable bundled third-party notices. Historical alpha releases remain unchanged. Changelog · Evidence method · Citation.

0.3.0a0 — Browser Python and reproducible reports

Datacenter Twin Lab 0.3.0a0 is a synthetic, local-first power-continuity what-if simulator.

Open the zero-install demo, change battery energy or demand, run the original Python engine on your device, and replay failure through recovery. The static site has no shared simulation API and sends no scenario telemetry. The first visit downloads about 14 MB of runtime and application assets.

This alpha adds:

• Bounded sensitivity sweeps for initial battery energy, IT demand, generator delay and distribution efficiency.
• Reproducible Markdown/HTML reports with scenario assumptions, event timelines, input hashes, warning counts and explicit unknown costs.
• Five scenario guides, independent hand calculations and a reproducibility capsule.
• Community issue forms, contribution/support/conduct guidance, a roadmap and an external-review worksheet.

The Python wheel includes the local dashboard, reports and battery-reserve comparison. Use the Python-only quickstart for the pinned asset and SHA-256. The core still requires only Python 3.12+ and the standard library; API dependencies are optional. The normal wheel does not include Pyodide.

Validation: full Python 3.12/3.14 matrix on Windows/Linux, installed-wheel checks outside the checkout, seven loopback dashboard journeys, and four browser-Python journeys. All five browser presets produce the same full result as native Python. Tests include independent energy calculations, malformed inputs, protected exports, report escaping and unknown costs. Windows Application Control blocked one local generated console launcher; required clean-run launcher checks passed in CI without changing local OS policy.

Assets include a source archive, a wheel, and a release manifest recording their hashes and source revision. The source archive comes from the clean public repository. The private planning/history repository remains private, and v0.2.0a0 assets and tag are unchanged.

These results demonstrate software behavior for synthetic fixtures. Independent external technical review has not yet been obtained. They do not establish facility calibration, cooling/GPU/workload performance, AC transients, protection coordination, uptime/Tier certification, compliance, or secure shared deployment. No physical controls are included.

Datacenter Twin Lab 0.2.0a0 — synthetic continuity alpha

Try the released dashboard with Python only: the verified installation guide uses an isolated environment and the hash-pinned wheel below. It includes Windows and macOS/Linux commands, expected synthetic results, and source-checkout caveats. Git and Node are not required for this installed-wheel route. The original tagged files and downloadable assets are unchanged.

Datacenter Twin Lab is a synthetic electrical-continuity research alpha with a deterministic Python engine, a loopback API, and a local React dashboard.

Replay utility outages, generator startup/failure, finite battery depletion, surviving-path capacity, shared failure and recovery. The worked tutorial derives the 665 kW surviving-path limit and 307.8 s battery ride-through independently, then reproduces both with the CLI.

Validation: 89 application tests executed from the source and extracted archive on Windows/Python 3.12.14; 88 passed and one symlink-privilege test skipped. Locked frontend install, formatting, TypeScript/Vite build, and isolated offline wheel install with module/console, catalogue, continuity and bundled-dashboard checks passed. The release commit's CI verifies Python 3.12 and 3.14 on Linux and Windows, including browser journeys on Linux. Anonymous source access, cloning, manifest hashes and the two CLI examples were verified.

The source archive contains 66 files including a per-file SHA-256 manifest. The wheel includes the compiled local dashboard and dependency notices. Python 3.12+ is required. The core has no third-party dependencies; see the quickstart for optional API installation and dashboard use. This GitHub release does not publish a package to PyPI.

Asset	SHA-256
public-release-2026-09-09.zip	b2cf9f567f1fb89c12698acdd514228a564df6e979433ad6f1a3bf1c025a434b
datacenter_twin_lab-0.2.0a0-py3-none-any.whl	0a6c56ce5fe5fd25d9886d35112eea56408450c8b7181e62b698b3c58a901261

These results establish software behavior for explicit synthetic fixtures. This alpha does not establish facility calibration, AC transients, protection coordination, cooling or workload/GPU performance, allocated cloud capacity, uptime/Tier certification, compliance, or a secure shared deployment. Costs preserve fictional or dated-reference assumptions. Use the documented loopback interface; no physical controls are included.

Reproduce an example and report a focused mismatch with the input, revision, expected value and observed result. Original code, documentation and synthetic inputs use Apache-2.0; retain third-party notices.