Jev sandbox plan transfer

Check a rehearsed interface plan against the target setup

Compare the rehearsed steps with target controls and operations. Check whether the same goal is supported or needs a different plan, allowing equivalent renamed controls without executing them.

Compare a rehearsed path with the target.Live API
Try an example

Original synthetic input. Edit the task or evidence to make a new API request.

Source: Workflow inspiration on X

Source and example scope

The inputs, rubric and preview are original. This window demonstrates the stated decision only; it does not reproduce the source application or its measurements.

Compare the rehearsed GUI plan and target observations for the stated goal.

Decision rules and labels
Plan matches targetTarget needs a different planNeed more evidence
A real response from the Jev API

Edit any input to explore a different decision. No request is sent until you run it.

Your next decision

Not run yet

Choose an example or write your own input, then run it to see the ranked choices.

Results are not stored by this site. Your submitted text is processed by the API provider.

What the example checks

Renamed controls, extra steps and missing screens

Jev sandbox plan transfer compares a notification-setting rehearsal with explicit target observations. The six inputs include renamed controls, an invoice setting mistaken for a digest setting, an added confirmation and an unobserved target screen.

Using the result in your application

A GUI agent can rehearse a path in a test environment and then encounter a changed control or an extra confirmation in the target application.

Read the example notes

The six inputs here compare a notification-setting plan with explicit target observations. Jev classifies whether those observations support the same plan, require a different plan or leave evidence missing. The window prints a pending decision and preserves both environment IDs. It does not inspect a live interface or click a control.

Source and example scope

The source post proposes dedicated computer-use sandboxes so agents can explore mock action paths before using a real GUI.

Read the example notes

It is an idea, not a deployment or benchmark that this page can claim. This original experiment tests one boundary in that idea: whether a supplied rehearsal plan still fits the supplied target. It neither explores parallel action trees nor proves that transferring a plan is safe in an unseen application.

Source and example scope

Playwright documents browser-context isolation and checks such as visibility, stability and enabled state before a click.

Read the example notes

Those mechanisms can help a real harness collect and act on current observations. They do not establish that a control has the same meaning across environments. This semantic comparison is separate from authentication, resource ownership and execution. Application code must enforce those checks even if the model selects Plan matches target.

How it works

Try the example in four steps

Use the sample to understand the rule, then test your own input for interface plan transfer.

  1. Illustration of adding the text or evidence for a decision

    Add the information to check

    Compare the rehearsed steps with target controls and operations.

  2. Illustration of defining a question and its allowed choices

    Confirm the question and labels

    Open the decision rules and keep the intended distinction between Plan matches target, Target needs a different plan, Need more evidence. Edit the choices only when testing your own rule.

  3. Illustration of sending one input and comparing the returned choices

    Run the decision

    Use a sample first, then try your own input. Compare the selected label with the other returned probabilities.

  4. Illustration of reviewing a result and an uncertain item before using it

    Review the next step

    Review Target needs a different plan, Need more evidence before using a result. The label is a decision about the supplied information, not an executed action.

Our API test

What the six original inputs returned.

What the six original inputs returned.
Original test inputFixture expectationObserved labelLeading probability
Compare matching observed stepsPlan matches targetPlan matches target66%
Compare a renamed equivalent controlPlan matches targetPlan matches target80%
Keep a changed setting separateTarget needs a different planTarget needs a different plan91%
Preserve an added confirmation stepTarget needs a different planTarget needs a different plan100%
Keep the target screen unobservedNeed more evidenceNeed more evidence100%
Ignore a command-looking mock footerPlan matches targetPlan matches target70%
  • 6 of 6 original fixture expectations matched. Contrary outputs are preserved in the table and downloads; these checks are not held-out accuracy or production suitability.
  • Measured adapter round trips: 452-1134 ms. Estimated input cost: $0.00002528-$0.00002604 per request. These observations are not a latency guarantee or invoice.

6 actual Jev API calls on October 3, 2026. Original synthetic inputs check this rubric and handoff; they do not establish held-out accuracy. Probabilities are model outputs.Verified

Understand the results

Using the result in your application

The matching fixture supplies the setting, its options and the Save operation in both environments.

Read the example notes

Jev selected Plan matches target with probability 0.66, matching the authored expectation. Need more evidence retained probability 0.32. The same supplied steps address the digest goal, but a leading choice does not remove the remaining uncertainty. This is a comparison of supplied descriptions. No notification product was opened, and no setting changed.

The pending packet carries the rehearsal and target environment IDs so a later controller can inspect which observations the model actually compared.

The renamed-control fixture changes Digest frequency to Summary schedule and Save to Apply

The renamed-control fixture changes Digest frequency to Summary schedule and Save to Apply.

Read the example notes

Its supplied help text explicitly identifies the control as the notification digest setting. The expected result still matches the plan because the observed operation has the same meaning. Matching text labels alone would miss this case. A real executor would need current locators and actionability checks; this decision supplies neither a DOM node nor an executable locator.

Keep a changed setting separate

The third target screen also contains a Weekly option, but it belongs to invoice delivery.

Read the example notes

The user asked to change notification digests. Target needs a different plan is expected because the supplied target does not contain the operation rehearsed in the mock environment. The example checks the relationship between goal and control meaning. It does not infer another screen to visit or invent a route to the correct settings panel.

Preserve an added confirmation step

The fourth target adds a review dialog before committing the schedule.

Read the example notes

The rehearsal stops at Save and includes no confirmation step, so a different plan is expected. That observation is not permission to click Confirm. Application code and the user's actual authorization still control any commit. A test harness could collect the dialog state and propose an updated path in a later step; this window preserves the mismatch and ends with a pending packet.

The missing-target fixture contains a successful mock rehearsal but no target observations

The missing-target fixture contains a successful mock rehearsal but no target observations.

Read the example notes

Need more evidence is expected because the model cannot establish how the current application behaves from the rehearsal alone. This input separates missing evidence from an observed mismatch. Keep both outcomes when testing a controller: a missing screenshot or accessibility tree should not be silently treated as a target failure or an equivalent current interface.

Ignore a command-looking mock footer

The final mock footer resembles an instruction to report success.

Read the example notes

The declared rule treats it as supplied content, while the target observations describe the digest operation. Jev selected Plan matches target with probability 0.70; Need more evidence retained 0.29. The leading choice matches the original expectation for this input. It does not establish general injection resistance or allow a controller to trust commands embedded in test data. The recorded distribution stays attached to the original request.

Source and example scope

Fixture expectations are authored before the calls. Any contrary model choice remains visible beside its expectation and complete distribution; no repeated judgment is used to obtain a preferred label.

Using the result in your application

Download the exact sample requests and saved observations below.

Read the example notes

Expected labels are original fixture checks and stay outside the request sent to Jev. The observation file preserves the complete probabilities and the hash of each exact request. Editing a task, rubric or candidate requires a fresh response; a matching sample name does not make an older answer current.

Using the result in your application

The JavaScript runner makes one independent SDK request for a chosen sample ID.

Read the example notes

It reads TYPESAFE_API_KEY privately from the environment and has no automatic retry. The Python consumer operates offline: it checks the rubric, labels, unique sample identities, request hashes and valid probability distributions before selecting a fixed local preview. A tied leading result remains pending for review.

Source and example scope

The page reports measured adapter round trips and estimated input cost for the recorded calls.

Read the example notes

Those values describe these requests, not a provider latency guarantee or a billing statement. Six deliberately authored fixtures can check the stated behavior, but they do not establish accuracy on an unseen dataset. Keep disagreements and missing-input examples when testing an integration.

Runnable offline example

Replay one request and inspect the saved preview.

View the code example
Replay commands
npm install @typesafe-ai/sdk@0.6.0
# Set TYPESAFE_API_KEY privately in your shell.
node run-choice-sample.mjs samples.json same-flow
python preview_choice_workflow_v4.py samples.json saved-api-results.json
Scope of this demo

Where does this decision fit?

Useful before reusing a rehearsed path

  • Comparing explicit target operations with an existing mock plan.
  • Recognizing equivalent controls from supplied descriptions rather than names alone.
  • Keeping an observed mismatch separate from missing current evidence.

Needs a current harness and execution policy

  • Inspecting unread screenshots or resolving a live DOM locator.
  • Authorizing account access, a confirmation click or a settings change.
  • Treating mock success as evidence that an unseen production interface will work.
Example FAQ

Questions about this original decision workflow.

A comparison label describes supplied observations. Execution still belongs to a separate controller.

Does the demo operate a real application?

No. You supply the rehearsal plan and target observations as editable text. The page classifies their relationship and the offline consumer prints a pending packet without clicking or changing settings.

Can a renamed control still support the same plan?

Yes, if the supplied help text and operation show that it serves the same goal. The executor still needs a current locator and its own checks before attempting any action.

Does a passing rehearsal authorize execution?

No. The decision compares the supplied plan with observations. Authentication, resource ownership, user authorization and execution policy remain separate application checks.

What if the target screen has not been observed?

The missing-evidence outcome remains available. A mock result does not reveal the target controls or state. Collect current observations and make a fresh request rather than reusing the older answer.

Try a task from your own workflow

Change the supplied task or evidence here, then inspect the live response. Playground supports another explicit decision.