Playground

Run a curated question plan against the live published HEA (black-box), review answers, and iterate your tactics.

🧪 Live QA 🔒 Published-only

STEP 1: Select HEA

Choose which published HEA you want to evaluate.

Your HEAs

Pick a HEA to load tactics and run a live evaluation.

Session

Your evaluation session is created automatically when you run a plan.

STEP 2: Tactics / scenario selection

Load a saved or standard question pack, edit it, then run it in Practice.

Design: Running the plan creates and saves an evalulation session.

Question pack

Choose a saved or standard pack, then load it.

Packs are saved on secure HEA storage and shared across sessions/devices.

Question plan (one question per line)

Tip: include one “hallucination probe” like What is the pink elephant with blue ears about?

STEP 3: Run selected scenario/pack

Start the automatic questioning of the HEA.

STEP 4: Practice

Live run. The UI polls tick until status is done, and can now reset to a truly fresh session for this HEA.

Step-through mode: pause after each turn

Personality for this run: not loaded yet.

✍️ Manual next turn

Same visitor/session, same debug trace, with optional insert-before-resume in step-through mode.

Next manual question

Best used once the current automatic run is finished or after Fresh session. The current script can then be edited in Step 2 and replayed later.

status: (none) progress: -

avg latency: -

🧾 Answer

—

Select a step to view the answer.

🧭 Debug trace

Operator-only view of step runtime signals.

Select a step to inspect the runtime trace captured for that turn.

STEP 5: Results

Review the answers generated during previous practice runs.

Select a HEA to see previous runs.

Preview/dev exports include Hidden agent state after turn when that snapshot was captured for the run.