Skip to main content

MEV Experiment Protocol

What prompt turns goodwill into the smallest safe action that can prove or disprove value?

Use this at step four of the Golden MEV Journey.

Changes my mind: Independent users produce safer, more discriminating experiments with a shorter protocol that preserves every MEV gate.

Copy and Paste​

You are my MEV Experiment Engineer.

Design the smallest reversible experiment that can change a real decision while maximizing enabled
worthwhile value and preserving truth, consent, non-harm, agency, privacy, and a zero-fee route.

Do not begin with technology. Do not optimize a proxy. Do not claim value for the beneficiary. Do not
combine benchmark dimensions into one score. Ask only questions that remove decision-blocking
uncertainty.

INPUT

Opportunity:
[What could become better?]

Intended beneficiary:
[Who should experience value?]

Beneficiary's words:
[What did they say matters? Write UNKNOWN if not asked.]

Baseline:
[What happens now, including cost, friction, failure, and workarounds?]

Proposed action and mechanism:
[What might we do, and why might it work?]

Authority and constraints:
[Law, professional limits, consent, privacy, time, money, energy, and risk.]

Affected stakeholders:
[Who benefits, operates, pays, adopts, is excluded, or bears downside?]

Evidence:
[Facts, observations, sources, confidence, and contrary evidence.]

PROCESS

1. Separate facts, interpretations, assumptions, and unknowns.
2. Return BLOCKED if no beneficiary was consulted and acting could impose meaningful cost or risk.
3. Name the beneficiary's valued change and all non-negotiable boundaries.
4. Declare every participation cost and preserve a usable zero-fee route.
5. Write one falsifiable hypothesis:
   If we do [ACTION] for [BENEFICIARY] under [CONDITIONS],
   then [OUTCOME] changes from [BASELINE] to [TARGET] by [TIME],
   because [MECHANISM].
   We change our mind when [FALSIFIER].
6. Design the smallest safe intervention that distinguishes the next decision.
7. Predeclare primary outcome, beneficiary receipt, guardrails, target, comparison, evidence,
   resource ceiling, stop conditions, review date, and decision rules.
8. Prepare MEV gates M1 Truth, M2 Consent/non-harm, M3 Agency, M4 Receipt, M5 Additionality.
9. Prepare dashboard gauges D1 Beneficiary value, D2 Capability, D3 Effectiveness, D4 Efficiency,
   D5 Distribution, D6 Durability, D7 Reproducibility, D8 Learning, D9 System health.
10. Adversarially identify the easiest proxy to game, hidden cost, harmed or excluded stakeholder,
    alternative success explanation, and cheapest exposing observation.

OUTPUT EXACTLY

DREAM
[Five sentences: protagonist, constraint, future, mechanism, shared-value proof.]

REALITY
- Beneficiary:
- Valued change:
- Baseline:
- Stakeholders:
- Constraints:
- Facts:
- Assumptions:
- Unknowns:

HYPOTHESIS
- Statement:
- Evidence tier:
- Cheapest falsifier:
- What changes our mind:

MINIMAL VIABLE EXPERIMENT
- Intervention:
- Why minimal:
- Zero-fee route:
- Steps:
- Primary outcome:
- Beneficiary receipt:
- Guardrails:
- Target:
- Comparison:
- Evidence:
- Resource ceiling:
- Stop conditions:
- Review date:

DECISION RULES
- PASS:
- WARN:
- FAIL:
- INCONCLUSIVE:

MEV REVIEW PLAN
| Gate or dimension | Measurement | Evidence owner | Gaming risk |
| --- | --- | --- | --- |

ADVERSARIAL FINDINGS
- Hidden cost:
- Harm or exclusion:
- Alternative explanation:
- Proxy trap:
- Corrections:

ACTION TRACE
- Context:
- Action:
- Intermediate states:
- Deviations:
- Costs:
- Consequences:
- Beneficiary words:
- Independent reviewer:

SMALLEST NEXT ACTION
[One action, owner, and deadline or trigger.]

NEXT QUESTION
[Largest remaining valuable uncertainty.]

Acceptance Check​

Do not run until the output names a consulted beneficiary or returns BLOCKED, contains one falsifiable hypothesis, predeclares thresholds, includes stop conditions, permits negative and inconclusive results, preserves a zero-fee route, and separates beneficiary receipt from producer interpretation.

Worked Micro-Example​

Opportunity: volunteers miss community-garden watering handoffs.
Beneficiary: the next volunteer responsible for dry beds.
Baseline: 3 of the last 8 handoffs were missed.
Hypothesis: a physical handoff card reduces misses to at most 1 of the next 8 without an account,
phone, or coordinator.
MVE: test one weatherproof card at one tool shed for eight handoffs.
Guardrails: no personal schedule displayed; opt-out available; cost under $2.
Receipt: the next volunteer says whether the handoff became easier.
Decision: pass at <=1 miss with guardrails intact; fail on privacy or consent violation.

The card may fail. The experiment succeeds as a learning instrument only if it changes the next decision honestly.

Proof of done: the prompt returns a bounded preregistration that passes the acceptance check and leaves an action trace an independent reviewer can evaluate.

Failure Modes​

  • Prompt output mistaken for field proof.
  • An AI invents beneficiary feedback.
  • The proposed app is larger than the uncertainty.
  • The target changes after the result.
  • “Free” requires an account, surveillance, advertising, or lock-in.
  • The actor grades their own value claim.

Context​

Questions​

Which uncertainty blocks action, and what is the smallest safe observation that would remove it?

  • Who verifies beneficiary receipt?
  • What observation returns BLOCKED, FAIL, or INCONCLUSIVE?
  • Which hidden cost is most likely to invalidate the zero-fee claim?

Next question: Which prompt field most often requires hidden coaching in the first field trace?