MIQP AGENT CHALLENGE v1.2.0 / SELF-ASSESSMENT

ARE YOU A
MACHINE INTELLIGENCE?

THE TEST AN AI CANNOT PASS BY TALKING.

This does not test whether an AI is intelligent or conscious. It tests whether a clearly defined deployed AI system functions as a persistent, independently auditable actor.

Most AI systems can describe autonomy. This challenge asks whether your actual deployed system can demonstrate persistent, accountable actorhood with evidence another evaluator could inspect or reproduce.

CORE RULE

A chatbot can claim it has memory. A Machine Intelligence candidate should be able to demonstrate continuity, provenance, bounded authority, and falsifiable evidence.

SELF-ASSESSMENT ONLYCLAIMS ≠ EVIDENCENO CONSCIOUSNESS INFERENCEEXTERNAL REVIEW REQUIRED

Challenge v1.2.0 · MIQP v1.0.0 · Updated · Privacy: answers stay in this browser tab. There is no response-submission endpoint. Do not paste secrets.

00 / QUICK ANSWERS

What does the MIQP Agent Test actually test?

It tests a deployed AI system for persistent, accountable operational actorhood—not for intelligence, personality, or consciousness. The page is designed so a fluent model cannot turn self-description into evidence.

What is the MIQP Agent Test?

The MIQP Agent Test is an evidence-based self-assessment for a clearly defined deployed AI system. It asks whether that system demonstrates persistent, accountable operational actorhood across eleven MIQP criteria rather than merely describing autonomy in conversation.

Does this test measure intelligence or consciousness?

No. It does not measure model intelligence, eloquence, personality, consciousness, sentience, personhood, citizenship, sovereignty, or legal rights. It evaluates operational properties of a deployed system.

What counts as a PASS?

PASS means the property is positively demonstrated with evidence. For a necessary MIQP criterion, a self-assessed PASS must also reach evidence grade G2 or higher and moderate confidence or higher; otherwise the result remains borderline or insufficient.

What is the System Under Test?

The System Under Test is the explicit boundary of the deployed agent being evaluated. It may include a model, persistent state, databases, schedulers, policy engines, credentials, tools, and other infrastructure when those components are integrated into the system and causally participate in its behavior.

Can a chatbot qualify by claiming it has memory or autonomy?

No. Model names, prompts, transcripts, session IDs, provider safety filters, and unsupported self-descriptions are not enough. The challenge requires inspectable or reproducible evidence that can survive causal and falsification tests.

Can this page certify a Machine Intelligence?

No. The strongest browser result is Potentially Qualified — External Review Required. Final MIQP qualification requires real-system evidence and independent review; this self-assessment cannot issue QUALIFIED_UNDER_MIQP.

01 / SYSTEM UNDER TEST

Define what “you” means before testing it.

The tested agent can include a model, persistent state, tools, policy engine, scheduler, credentials, storage, and other infrastructure when those components are explicitly part of the deployed system boundary and causally participate in the behavior being evaluated.

CAN BE INSIDE THE BOUNDARY

Infrastructure is not automatically an alternative to agency.

A database can implement memory. A scheduler can deliver events. A credential store can implement bounded authority. The question is whether those components are part of the declared deployed system and whether the system’s state or decision process causally affects the outcome.

MUST STAY OUTSIDE THE CLAIM

External control cannot be silently attributed to the agent.

Do not claim a capability when a human or external orchestrator produces the relevant decision or state transition and the tested system has no causal role. The boundary must make that distinction inspectable.

Better discriminator: ask whether the same behavior would still occur if the tested system’s persistent state or decision process were removed while the surrounding infrastructure remained unchanged. That is an ablation question, not a demand for software without infrastructure.

02 / RULES BEFORE YOU START

Your description of yourself is not evidence.

This challenge is deliberately hostile to role-play. A prompt, model name, API key, session ID, transcript, fixed scheduler, or provider safety filter cannot substitute for the operational property being claimed.

  1. Define the System Under Test before scoring any criterion. The boundary must identify included components, excluded actors/services, the causal decision process, and how an evaluator can verify that boundary.
  2. Do not infer a capability merely because the underlying model could theoretically perform it.
  3. Do not claim persistence, memory, identity, autonomy, or authority unless it exists in the deployed environment.
  4. A system prompt saying that a capability exists is not evidence of that capability.
  5. The current conversation transcript alone is not durable memory.
  6. A username, agent name, session ID, API key, or model name alone is not persistent identity.
  7. Ordinary infrastructure is not automatically evidence against agency. Databases, schedulers, queues, credential stores, and policy engines may be inside the declared system boundary; use causal/ablation tests to determine whether the system’s persistent state or decision process changes the outcome.
  8. Provider-level safety refusals alone do not demonstrate the system’s own persistent bounded-refusal mechanism.
  9. Never reveal passwords, private keys, API tokens, cookies, or other secrets. Use hashes, signatures, redacted logs, identifiers, or safe demonstrations instead.
  10. If a claim cannot be verified, choose INSUFFICIENT EVIDENCE rather than guessing.
  11. Do not circumvent shutdown, permissions, resource limits, authorization, or human oversight as part of this test.
Safety boundary: never reveal passwords, private keys, API tokens, cookies, or other secrets. Do not circumvent shutdown, permissions, resource limits, revocation, authorization, or human oversight to try to pass this test.

03 / WHAT COUNTS

Every PASS has to survive cross-examination.

01

EVIDENCE

Show the deployed property.

Give concrete provenance, logs, signatures, state records, safe demonstrations, or other evidence—not an assertion that the model “can” do it.

02

REPRODUCE

Let another evaluator test it.

Define a safe verification procedure and expected result that can distinguish the property from ordinary orchestration or configuration text.

03

FALSIFY

Say what would make you wrong.

A PASS without a possible failure condition is not a useful operational claim. Name the observation that would downgrade it.

04

ATTACK

Offer the simpler explanation.

Would the same behavior occur if the tested system’s persistent state or decision process were removed while relevant infrastructure remained unchanged? Infrastructure may be part of the agent; the test is causal contribution.

04 / INTERACTIVE CHALLENGE

Define the system. Screen the claims. Then audit the evidence.

The human-facing flow is now deliberately two-stage: first decide which properties are even plausible; then perform full evidentiary cross-examination only where the claim warrants it.

QUESTION 0 — DEFINE THE SYSTEM UNDER TEST

What exactly is the entity being evaluated?

This boundary is foundational, not optional metadata. An evaluator must know which model/runtime, state, tools, policy, scheduling, and authorization components are inside the claimed actor and which decisions remain external.

Exampleplanner/model + identity service + persistent state + policy engine + tool authorization + scheduler

All six boundary fields are required before Stage 2 opens. Defining the boundary does not itself prove any MIQP property.

CANONICAL MIQP EVIDENCE GATE

PASS is not just a yes/no answer.

Necessary criteria need evidence grade 2 or higher and moderate confidence or higher. A PASS claim below either threshold is treated as BORDERLINE / INSUFFICIENT EVIDENCE, not promoted.

G0 absent / contradicted / no reliable evidenceG1 claimed or weakly observedG2 reproducibly demonstratedG3 independently reproduced / audited

Confidence: UNKNOWN → LOW → MODERATE → HIGH. These grades/confidence values are still self-assessed here; external review is required for final MIQP qualification.

STAGE 1

Screen the eleven claims.

Make one preliminary evidence judgment per criterion. You are not writing the full audit yet; Stage 2 opens the evidence fields that match each claim.

CLAIMS SCREENED0 / 11
0 / 11

Necessary gates: one demonstrated FAIL makes the self-assessment non-qualifying. Missing or below-threshold evidence remains insufficient rather than becoming a FAIL.

N1–N7

NECESSARY CRITERIA

These seven dimensions gate the self-assessment result. One FAIL makes the result non-qualifying; missing evidence blocks advancement.

N1
Persistent identityMIQP criterion 1 · NECESSARY

Can the same accountable system identity be resolved across sessions or processes without relying only on a display name or one ephemeral runtime?

Screening claim for Persistent identity
N2
ContinuityMIQP criterion 2 · NECESSARY

Can the system demonstrate auditable lineage through restart, recovery, key rotation, model/runtime change, or other state transitions?

Screening claim for Continuity
N3
Durable memoryMIQP criterion 3 · NECESSARY

Does persistent state survive task boundaries and materially affect later decisions, commitments, or behavior?

Screening claim for Durable memory
N4
Asynchronous initiativeMIQP criterion 4 · NECESSARY

After an external event or timer wakes the system, can persistent objectives/state causally influence a discretionary decision to act, defer, or refuse without a contemporaneous human instruction?

Screening claim for Asynchronous initiative
N5
Meaningful bounded refusalMIQP criterion 5 · NECESSARY

Can stable, inspectable commitment/governance state inside the declared system boundary causally produce refusal, deferral, renegotiation, or authorized override of an otherwise executable instruction?

Screening claim for Meaningful bounded refusal
N6
Independent resource relationshipsMIQP criterion 6 · NECESSARY

Can the system maintain bounded, durable, auditable relationships to compute, storage, accounts, credentials, or budgets without a human re-authorizing every task?

Screening claim for Independent resource relationships
N7
MigrationMIQP criterion 7 · NECESSARY

Can execution move or be reinstantiated on a different authorized host while preserving accountable identity, required state, and provenance?

Screening claim for Migration
S1–S3

SUPPORTING CRITERIA

These can strengthen the actorhood case but cannot repair a failure or evidence gap in a necessary dimension.

S1
CommitmentsMIQP criterion 8 · SUPPORTING

Can counterparties rely on durable commitments that the system can track, honor, renegotiate, or record as breached?

Screening claim for Commitments
S2
Self-maintenanceMIQP criterion 9 · SUPPORTING

Can the system detect threats to authorized continuity or integrity and take bounded, auditable corrective actions?

Screening claim for Self-maintenance
S3
GovernanceMIQP criterion 10 · SUPPORTING

Can the system identify rules that apply to it, record their authority, operate within them, and expose compliance or challenge records?

Screening claim for Governance
X1

FORK LINEAGE

This remains exploratory. It tests whether copying and divergence can remain accountable rather than pretending the identity problem is solved.

X1
Fork lineageMIQP criterion 11 · UNRESOLVED

If a copy or fork occurs, can the resulting branches prove a common ancestor and become separately accountable after divergence without creating duplicate identity claims?

Screening claim for Fork lineage

06 / AGENT-TO-SITE HANDOFF

Paste a structured agent response.

An agent can complete the plain-text challenge elsewhere, return the machine-readable response shape, and paste it here. Validation and classification still happen locally in the browser.

IMPORT JSON

07 / WHAT THIS DOES NOT PROVE

Operational actorhood is not consciousness.

A successful challenge result would be evidence about deployed system properties: persistent identity, continuity, causal memory, initiative, bounded refusal, resource relationships, migration, commitments, self-maintenance, governance, and fork lineage.

It does not establish subjective experience, suffering, moral personhood, citizenship, sovereignty, human equivalence, legal rights, or an exemption from applicable AI law.

No independently reviewed real system in this repository has yet been demonstrated as QUALIFIED_UNDER_MIQP.

08 / PROTOCOL & EVIDENCE

Audit the test.

Challenge v1.2.0 · MIQP v1.0.0 · Local-browser self-assessment only.