SynthWorld documentation

Current-main documentation. These pages describe the repository’s current main branch and may include changes not yet released. For a released contract, use the matching signed Git tag, release notes, and packaged artifacts.

SynthWorld generates deterministic, safely fictional identity worlds and separates product-facing observations from evaluator truth so identity, privacy, access, and agent systems can be tested reproducibly.

Mental model

explicit seed + config + schema version + event schedule
                         |
                         v
             deterministic identity world
                    /            \
                   v              v
          public product input   evaluator truth
                   |              |
                   v              |
             system under test    |
                   |              |
                   +---- prediction/trace
                                  |
                                  v
                       independent metrics

Choose your goal

Goal	Start here
Install and create a first world	Getting Started
Build connected identity fixtures	Identity worlds
Evaluate matching or privacy behavior	Identity resolution or privacy exposure
Test agent delegation and audit evidence	Agent authority
Compare RBAC, ABAC, and ReBAC on one agentic world	Enterprise agentic identity experiment
Compile enterprise identity and access truth	Enterprise access
Build and score an enterprise authorization experiment	Enterprise authorization
Add identity and authorization checks to CI	Development pipelines
Plan an enterprise identity and authorization journey	Enterprise Identity Planning
Connect a system to an evaluator	Evaluating a system
Understand benchmark contracts and boundaries	Concepts

Five-minute quickstart

pip install idcognito-synthworld
synthworld generate --seed 20260719 --persona-count 10 --output world.json

world.json is fictional and repeatable for the same explicit inputs. Continue with Getting Started before using a benchmark scorer.

Capability and benchmark status

Capability maturity and benchmark publication are different axes. The documentation build consumes drift-checked resolved capability and benchmark registries and emits a public allowlisted catalogue:

A benchmark being packaged or published does not imply that every related capability is mature, and capability maturity does not authorize external publication.

Featured benchmarks

Current focus

The project is preserving frozen benchmark bytes while improving user journeys, publication controls, generated enterprise/agent depth, and evidence-binding contracts. Candidate benchmark artifacts remain candidates until their explicit publication gates are satisfied.

See the roadmap view for Now/Next/Later framing and the generated registries for current maturity/publication state.

Navigate

Destination	Link
Home	This page
Getting Started	Install and first evaluation
Guides	Journey guides
Benchmarks	Benchmark reference
Experiments	Reproducible experiments
Roadmap	Now/Next/Later
Support	Help and contribution routes
Reference	Technical reference