verifiedBSD-3-Clause

Scientific Python simulation

A sealed numerical-analysis environment for declared models, parameters, seeds, sensitivity checks, and reproducible outputs.

Jump to experiment form Saved experiments

What it is for

Mechanistic model comparisonParameter sweepsSimulation-backed research

Inputs

Declared model template
Parameters with units
Random seed
Acceptance checks

Outputs

Run manifest
Numerical result
Sensitivity report
Reproducible figure

Scientific limitation

A simulation tests stated assumptions; it does not validate the assumptions or replace physical experiments.

Upstream provenance

https://scientific-python.org/

Request a managed run

compute ready

Declare a Python, R or Julia program, inputs, hypotheses, criteria and budget. Cathedral hashes the protocol before execution. Programs run only in the isolated sandbox, without application secrets or outbound network.

First seal the protocol. Then inspect related attempts and explicitly approve the saved run. Registration alone does not start compute or a paid model.

Experiment protocol and inputsResearch questionResearch domain

Write one statement per line. Declare the test and its failure conditions before registration. The numerical example is replaceable; it is not a malaria experiment or a scientific discovery.

HypothesesPredictionsSuccess criteriaFailure criteriaLimitationsExperiment modeProgram languageProgram entrypoint

Code is never run in this editor. Use the selected language and packages already present in the qualified runtime. Outbound networking and package installation during the run are blocked. Write result files under outputs/.

Attach code or data files

Up to 1 MiB per file and 4 MiB total, at most 32 files. Use relative filenames with letters, numbers, dots, dashes, underscores or folders. Uploads stay in this browser until you approve registration. Large datasets are not silently truncated or downloaded.

Inspect or edit a file

Relative file pathFile contents

Edited UTF-8 text: the server computes its exact byte hash at registration.

Dataset identifiers (one per line)Source DOIs (one per line)

An identifier does not attach its data or prove access rights. Include the actual required files and confirm permission. Uploading a file does not qualify its units, cohort or methodology.

Random seedWall time ceiling (seconds)Run spending ceiling (USD)Model token ceiling

The per-run ceiling is not a price quote. The shared daily admission cap still applies. No computation starts until you explicitly approve the saved inputs.

Private datasets

For files larger than inline attachments, stage immutable private data directly to storage. Launch limits: 512 MiB per owner, shared 1 GiB storage and 2 GiB monthly transfer reservation. Every verification and compute transfer consumes that allowance. Never upload identifiable patient data without approved permissions and a suitable data agreement.

Private dataset fileDataset path in the experimentPermission or licence basisI have permission to upload and privately process these data. Staging transfers bytes now, separately from approval to execute a run.

Advanced JSON

Advanced run input

{
  "question": "How accurately does forward Euler approximate a declared exponential-decay model?",
  "domain": "numerical-methods",
  "protocol": {
    "hypotheses": [
      "Forward Euler with the supplied step size approximates the analytic exponential-decay solution."
    ],
    "predictions": [
      "For initial=1, rate=0.2, seconds=10 and dt=0.01, the absolute final error is below 0.002."
    ],
    "successCriteria": [
      "Zero exit, inspectable trajectory/result/figure and final error below the registered threshold."
    ],
    "failureCriteria": [
      "Invalid parameters, nonzero exit, missing artifacts or absolute final error at least 0.002."
    ],
    "limitations": [
      "A numerical-method check only; no real-world decay mechanism is validated.",
      "The declared rate, units and model assumptions require separate experimental evidence.",
      "Matching a clean rerun is not independent scientific replication."
    ],
    "mode": "computational",
    "seed": 42,
    "datasetIds": [],
    "sourceDois": [],
    "program": {
      "language": "python",
      "entrypoint": "analysis.py",
      "files": [
        {
          "path": "parameters.json",
          "text": "{\"initial\":1,\"rate\":0.2,\"seconds\":10,\"dt\":0.01}"
        },
        {
          "path": "analysis.py",
          "text": "import csv, json, math\nfrom pathlib import Path\nimport matplotlib\nmatplotlib.use(\"Agg\")\nimport matplotlib.pyplot as plt\np = json.loads(Path(\"parameters.json\").read_text())\ninitial, rate, seconds, dt = [float(p[k]) for k in (\"initial\",\"rate\",\"seconds\",\"dt\")]\nassert all(math.isfinite(x) for x in (initial, rate, seconds, dt))\nassert 0 < dt <= seconds <= 1000 and 0 <= rate * dt < 1\nsteps = round(seconds / dt)\nassert 1 <= steps <= 100000 and abs(steps * dt - seconds) < 1e-8\nvalue = initial\nrows = [(0.0, value, initial)]\nfor i in range(1, steps + 1):\n    value *= 1 - rate * dt\n    t = i * dt\n    rows.append((t, value, initial * math.exp(-rate * t)))\nerror = abs(value - rows[-1][2])\nwith Path(\"outputs/trajectory.csv\").open(\"w\", newline=\"\") as f:\n    writer = csv.writer(f); writer.writerow((\"time\", \"euler\", \"analytic\")); writer.writerows(rows)\nPath(\"outputs/result.json\").write_text(json.dumps({\"parameters\":p,\"finalEuler\":value,\"finalAnalytic\":rows[-1][2],\"absoluteError\":error,\"criterionPassed\":error < 0.002,\"scope\":\"Numerical check, not validation of a physical system.\"}, sort_keys=True))\nplt.plot([r[0] for r in rows], [r[1] for r in rows], label=\"Euler\")\nplt.plot([r[0] for r in rows], [r[2] for r in rows], label=\"Analytic\")\nplt.xlabel(\"Declared time units\"); plt.ylabel(\"Declared quantity\"); plt.legend()\nplt.savefig(\"outputs/decay.png\"); plt.close()\nassert error < 0.002, \"Registered numerical criterion failed\"\n"
        }
      ],
      "expectedOutputs": []
    }
  },
  "budget": {
    "maxTokens": 64000,
    "maxCostUsd": 2,
    "maxWallSeconds": 180
  }
}

Expected: Declared model template; Parameters with units; Random seed; Acceptance checks

I approve these inputs for an isolated run and understand that outputs remain reviewable research artifacts, not automatic scientific truth.

Execution boundary

ready

The immutable Python/R/Julia runtime is available. Protocols are pre-registered, owner-approved and independently critiqued.

Every execution uses sealed code, an immutable qualified image, declared resource limits, blocked outbound networking, and an auditable event stream.