FORGE
public demo / deterministic adapters

HOSTED DEMO TRAINING DATA RELEASE PIPELINE

Training data,
from source to release.

Forge processes two source documents through governance, deduplication, contamination checks and source-isolated splitting. Run the workflow, inspect its decisions and download the release record.

Start a test run
PIPELINE READY
  1. 01
    Source controlsIngest / rights / privacy
    2 stages
  2. 02
    CurationGenerate / validate / deduplicate
    4 stages
  3. 03
    Evaluation controlsJudge / contamination / selection
    4 stages
  4. 04
    ReleaseSource split / dataset profile
    2 stages
12 stages7 release gates
EXECUTION12-stage workflow
RELEASE GATES7 enforced checks
SPLIT POLICYZero source overlap
OUTPUTDownloadable evidence
01 RUN FORGE

HOSTED WORKFLOW

Run Forge
in the browser.

Choose a prepared scenario or provide two text documents. Public runs use the same Python workflow as the CLI and test suite, with deterministic adapters and bounded inputs.

A

SOURCE DOCUMENTS

Choose an input

B

PIPELINE STAGES

12-stage execution

0/ 12
    03 LIMITATIONS

    SCOPE OF THIS DEMONSTRATION

    What this run
    establishes.

    The hosted run verifies pipeline execution and artifact integrity. It does not evaluate model quality or establish production safety.

    VERIFIED IN THIS RUN
    • All 12 pipeline stages completed
    • Release artifact hashes verified
    • Zero overlap between train and test sources
    • Declared governance and contamination checks completed
    NOT EVALUATED
    • Production model quality
    • Safety for production use
    • Utility for a specific task
    • Complete detection of sensitive data