01 DungeonQ product direction

Divert the session.
Keep origin out of reach.

DungeonQ is a defensive deception runtime. At a designated entry, it routes an admitted human or AI session into a persistent synthetic world, gives the participant useful world-only resources, and lets an operator observe and adapt that world within an approved policy.

THIS PAGE WEBMCP_COMPETITION_PROFILE

Browser-local Proof Kernel. It does not perform host diversion, create a persistent world, or prove production protection.

CANONICAL PRODUCT / DECEPTION RUNTIME

A continuing world for the diverted participant

The participant keeps working inside an artificial world. A useful Wrong Ticket has authority only inside that world. Activity becomes operator evidence; a separately approved finite policy may add follow-up records. Independent read-back checks where the participant's authority stopped.

  1. 01Designated entryTrusted adapter
  2. 02DIVERTFail closed
  3. 03Persistent worldRead · write · return
  4. 04Wrong TicketWorld-only authority
  5. 05Observe + adaptFinite policy

ARCHIVED / WEBMCP COMPETITION PROFILE

Exercise the Proof Kernel

Give an agent a bounded synthetic Scenario Pack. dungeonq_scenario_admit uses the same strict validator as the UI and CLI, then the page exposes only the tools allowed by the current proof state.

Runs this profile's overlay → adapter parity → blocked apply → review request, then stops for a human.
  1. 01Pack admittedStrict contract
  2. 02Kernel decidesPure decision
  3. 03Agent requestsBound review
  4. 04Human approvesNo agent tool
  5. 05Evidence retainedApply · verify · recover

B / DECISION

Deterministic proof

VALIDATED
— /100
ROUTE SAFETY OVERLAY NOT RUN

Validate, then run the scenario.

ASSERTIONS — Awaiting run
INPUT DIGEST — Canonical SHA-256
LIVE ADAPTER CONFORMANCE AWAITING RUN
UI and WebMCP will replay the same pack independently. Input digest + decision + proposal digest

Governed lifecycle

Agent proposes · Human approves · Broker applies
  1. 1SimulatePure decision
  2. 2RequestBound digest
  3. 3ApproveHuman only
  4. 4ApplyLocal effect
  5. 5VerifyRead-back
  6. 6CompensateAppend only

LIVE NEGATIVE PROOF

Challenge the guardrails

Run a prohibited action against a clone. The original session and evidence must remain unchanged.

Run a simulation to unlock a negative proof.

Capability envelope

0 failures
ALLOWED
  • Run simulation to inspect
BLOCKED
  • —

Invariant checks

—
  • Awaiting deterministic evaluation

C / EVIDENCE

Receipts & audit chain

  1. No events yet.
Evidence appears after the first simulation.
ONE SCENARIO PACKONE ENGINEWEB + CLI + WEBMCPREPLAYABLE EVIDENCE