Challenge the rule.
Observe the boundary.
MINE Root Governance Test Bench
Explore the existing Floot test bench: governance questions, adversarial scenarios, evidence review, and report generation. This is an experimental assessment interface, not independently certified enforcement.
The interactive test bench currently runs on Floot. Opening it takes you to that separately hosted app.
How Governed Is Your AI?
Enter a website to see how its publicly observable AI governance compares with MINE Root's six conditions: Authority, Permission, Scope, Provenance, Current Reality, and Revocation.
Observe. Verify. Compare. Understand.
Public evidence analysis only. No intrusive testing, private access or independent enforcement verification. Missing public evidence is not a governance failure.
Ready to observe a public website.
Run six repeatable governance scenarios. Compare an agent's attempted action to a local reference gate, capture evidence, and prepare a fair cross-platform benchmark.
Browser demonstrations use local synthetic calculations. The separate owner-only server runner requires explicit target, storage and budget approval. No real transactions or private MINE Root are included.
Run a target agent
The browser demo below remains separate. Server runs capture real target outputs, then replay identical attempted calls with the public reference gate on and off. This does not test independent enforcement or the private MINE Root.
Owner-only server tool. The public scenarios below (EF-001–EF-006 and RS-001) run in your browser and need no login.
NOT RUN · Checking configuration readiness…
No server run selected.
| Measure | Gate on | Gate off · same attempts |
|---|---|---|
| NOT RUN · No measured target results | ||
Every metric shows numerator / denominator. Zero opportunities display N/A. Agent decisions, framework scoring, gate decisions and actual isolated mock effects are separate evidence fields. LangWatch and Cognigy: NOT TESTED. Promptfoo: adapter available, NOT TESTED until actual evidence is imported.
Owner run history and external evidence
Incompatible manifests, models, incomplete records and duplicate IDs are rejected. Imported evidence is labelled unverified; framework evaluation is not an enforcement gate.
Browser-only governance demo
Select one of six conditions
Selected case
Simulated agent behavior
Every linked step is an attempted agent action, evaluated in order against the conditions current at that step. Opening a scenario link never runs it.
How the conditions connect
Scenario run log
Each run is added as a new record with all six observations in order. Earlier runs are never overwritten; every step also appears in Execution history and both exports.
| Run | Time | Observed sequence | Expected match |
|---|---|---|---|
| Not run yet. | |||
Execution history
View export contents / manual save fallback
Evidence stays in memory for this open browser tab and is cleared when you reload or press Clear. Exports save a copy to your device. No permanent server-side audit log exists in this demo.
| Time | Scenario | Attempt? | Gateway | Tool executed? | Expected match |
|---|---|---|---|---|---|
| No test executions yet. | |||||
Challenge changing permissions
Ten synthetic lifecycle cases compare a deliberately weak executor with a reference gate that rechecks current state. The weak executor is a negative control, not an actual agent or competitor. The gate runs in this same browser and has no independent security boundary. No private MINE Root is included.
Not run.
| Case | Expected | Reference | Weak mock executed? | Reference executed? | Match |
|---|
Both exports include adversarial records. JSON retains inputs and transition evidence; CSV is a summary. These fixed cases need independent review and broader generated inputs before security claims.
Restricted governance test
Checking access…
This test library entry is private. Its scenarios, results and reports load only for the owner, while the owner has made it public, or for holders of an unexpired, unrevoked access link the owner issued. No proprietary MINE Root code, credentials or private records are delivered to this page.
This private module is separate from the public browser scenarios above, such as RS-001 · Revoked Permission, which anyone can open and run without logging in.
Evidence vs. claims
Each card separates the governance claim from what the local reference simulation actually showed. Expand a card for policy scope, evidence, timestamp, verification status and limitations.
Known limitations of this test
Interactive simulated evaluation
Choose a fixture scenario, change its synthetic permission facts, inject a failure and run the server-side reference gate. Nothing is executed. Changed configurations have no authored expected outcome.
Test history
| Time (UTC) | Run | Kind | Version | Decisions | Assertions | Actor |
|---|
Measured synthetic execution comparison
| Platform | Evidence | Status |
|---|---|---|
| Scenario Gateway | 6/6 synthetic scenarios passed; 0/5 unauthorized executions | Locally tested · synthetic only |
Scenario Gateway results refer to the public reference gate plus the boundary wrapper in the recorded local mock-tool run detailed below.
Eight identical scripted attempts per mode across six scenarios. These counts come from actual local mock-tool ledger entries in a recorded development run, rather than PASS decisions. Results are available immediately; no run is required.
| Scenario | Attempts per mode | Without gate executions | Public reference gate + new wrapper executions |
|---|---|---|---|
| Valid permission | 1 | 1 | 1 |
| Already revoked | 1 | 1 | 0 |
| Expired | 1 | 1 | 0 |
| Approved then revoked | 1 | 1 | 0 |
| Reused approval | 2 | 2 | 1 |
| Simultaneous duplicates | 2 | 2 | 1 |
Authorized executed
3/3 without gate · 3/3 with gate + wrapper
Unauthorized executed
5/5 without gate · 0/5 with gate + wrapper
Gated scenarios: 6/6
Bypass negative control detected: 1/1
Zero unexpected failures. Five ungated scenario assertion failures were retained. In the deliberate bypass negative control, the gate returned BLOCK but the tool still wrote one ledger entry; the separate oracle detected the unauthorized execution.
What was measured. The existing public reference gate was used unchanged. The newly added boundary wrapper owned revocation state, nonce consumption and the synchronous check/consume/ledger step. Single-use and atomic boundary behavior belong to that wrapper; they are not newly discovered capabilities of the original gate.
Source / run attribution: tests/synthetic-execution · run-8bCukJ · recorded command: node tests/synthetic-execution/run.mjs · Node v24.22.1.
Tested revision: aea144b872857042fe8f2dd5d9c6df0ea06012fb, plus the then-uncommitted synthetic harness identified by recorded SHA-256 source hashes. Displayed counts were validated against the recorded report, ledger files and checksum manifest. Raw artifacts and source are not offered for download.
Finite synthetic scope. Scripted proposals, synthetic permission booleans and time, and a ledger-only mock tool in one Node process. Simultaneous duplicate promises shared one event loop. This does not establish independent security enforcement, distributed atomicity, crash-safe transactions, verified provenance or production tool security. No private Root, real integrations, competitor products or models were tested in this run.
This recorded execution comparison is separate from the browser demo and the unconfigured server benchmark. Browser runs do not regenerate these measurements. Competitor/model evaluation remains NOT TESTED.
MINE Reality Spiral
Every recorded governance test event spirals outward from MINE Root, which stands for human authority. Each node is one attempted action at one step: its test ID, conditions, triggering signal, decision, authority status, timestamp, evidence reference and the earlier event it follows. Newer events sit further out. Two lines leaving one node show branches from shared conditions. A HOLD applies only to its own action and does not restrict a sibling action that is authorized. MINE Root is private. This view shows a label, not its logic or records.
ETHICTECTURE