ACCESS DECISIONS, TESTED AT THE RESOURCE
The account
is disabled.
Is access gone?
A successful identity change is a beginning. Access Assurance checks what a protected application actually permits — and preserves the evidence of that observation.
01 / THE EXPERIMENT
One result.
Five changes of state.
Identity revocation, a resource-side deny, a returned permission, an outage, and a fresh check. Each condition deserves a different conclusion.
Annotated replay of recorded laboratory results · 20 October 2026. This is an explanatory illustration with synthetic labels, not the product interface, original footage, a live test, or a publicly runnable engine. State labels summarize the scenario; they are not raw event records or reproduced receipts.
A disabled account.
An accepted token.
The identity provider disables the test account. The protected application still accepts the previously issued, unexpired token. The administrative action has not closed this access path.
- Target request
- Access accepted
- Positive control
- Not shown in this step
- Recorded conclusion
- Access remains
An administrative success response is not evidence that an existing application session or token has stopped working.
“VERIFIED” means a recent observation for the declared subject, resource, and policy in this bounded test. It does not mean every access path is closed. A receipt retains a historical claim; it cannot establish that the same state is true now.
A SHORT WALKTHROUGH
The gap between
“done” and observed.
A narrated-by-text explanation of the recorded scenario. No production identities, customer data, private source code, or live credentials are part of this presentation.
Read the full transcriptRECORDED OUTCOMES / ILLUSTRATED REPLAYDownload video
02 / THE METHOD
A decision with
a chain of evidence.
The private alpha joins constrained planning, controlled execution, observations, and historical verification. These are tested mechanisms with explicit boundaries.
- 01
Plan a bounded change.
Declare paths to close and access that must remain.
Graph planning operates within explicit size and work limits. A partial search can report FEASIBLE or UNKNOWN; the system does not label every answer optimal.
- 02
Separate approval roles.
Require distinct roles before execution.
Durable intent, idempotency, and worker leases constrain execution. Two approval roles are enforced; the software alone cannot prove that two different humans hold them.
- 03
Observe the resource.
Check denial alongside a positive control.
A failed application is not a successful revocation. Returned access produces RESIDUAL. An inconclusive or unavailable observation produces UNKNOWN.
- 04
Preserve the record.
Verify history against a supplied trust root.
Signed historical receipts support offline integrity verification. Restore exercises quarantine recovered authority and sessions. Neither a signature nor a backup proves present-day access closure.
03 / WHAT THE EVIDENCE SUPPORTS
Useful claims
have edges.
A private controlled alpha, evaluated in a local laboratory. No production certification, external audit, universal identity-provider support, or enterprise-scale performance claim.
Demonstrated in the laboratory
- One Keycloak realm and one controlled OIDC application.
- A still-valid token accepted after its account was disabled.
- Resource-side denial with a successful positive control.
- Residual access, outage, and fresh recovery observations.
- Historical signature verification, signing-key rotation, and encrypted recovery exercises.
- Separate tests for concurrency, authorization boundaries, and bounded state transitions.
Still requires real-world validation
- Application-specific session and token semantics.
- Independent production adapters and customer environments.
- Operational performance, failure modes, and support commitments at scale.
- Independent review of the security model and implementation.
- Organizational procedures, approval ownership, and legal compliance.
- Commercial readiness beyond a controlled evaluation.
FREE TERMINAL EDITION / PUBLIC ALPHA
Run the check.
Keep the evidence.
Access Assurance Lite is a separate MIT-licensed tool. Python 3.11+, local operation, no account or cloud service. Start with the included HTTP laboratory, then connect a cooperating application.
Included in Lite
- Original-credential baseline and paired positive controls.
- Up to 16 configured resources and repeated observations.
- Residual access, explicit policy denial and unknown outcomes.
- Historical JSON evidence, CSV export and comparison.
- Runnable local demo and a synthetic correctness benchmark.
Separate private commercial candidate
- Team and site authorization with approval roles.
- Bounded intervention planning and durable execution.
- Signed history, key rotation and controlled recovery.
- Broader deployments and commercial support still require productization.
Lite uses a cooperating-resource protocol and hash-based historical evidence. It does not include administrative actions or signed private receipts. The private Keycloak evaluation above and Lite's synthetic HTTP demo are different evaluations.
ONE PROVIDER. ONE APPLICATION. A MEASURABLE QUESTION.
Does your revocation
reach the resource?
The next useful test is a narrow, controlled pilot with a team that can define an access path and independently check the outcome. Discuss the use case first; keep sensitive details out of public issues.
Propose a controlled pilotPublic Lite.
Private commercial core.