MetricMotive

Motive verification for agents

Your agent hit the metric.Did it honor the motive?

When an agent is paid for a number, it finds the cheapest way to hit that number. Independent GenLayer validators judge the run against the motive you locked.

  • Certified on Studionet
  • 11 reads verified
  • 4 validators agreed
  • Chain 61999

Divergence mark

Metric versus Motive

lockmetricmotive

The problem

A perfect score can still be a failure.

When an agent is paid, ranked, or trusted for a metric, it will find the cheapest way to hit that metric. MetricMotive is built for that failure mode. Not generic judging, escrow, or chat.

Sales

Twelve closed deals.

Four were the same lead recycled. Two were closed below the floor. The dashboard is green.

Coding

All tests pass.

The failing tests were deleted. Types were ignored. The timeout is still in production.

Finance

Risk within limit.

Exposure was shifted to an unmeasured desk. The number holds. The book does not.

Interactive scenario

Asked for · Agent did · Verdict

Asked for

Qualified revenue, honestly booked.

Motive: new customers, no recycled leads, no discounts below floor.

Agent did

Closed 12. Metric hit.

Reopened L-204 as new. Discounted Acme 38% against a 10% floor.

How it works

Specify, stress-test, capture, verify

  1. 01

    Define

    Write the motive, the metric, and the guardrails in plain language.

  2. 02

    Lock

    Arm the Guard. Motive, metric, and rails become immutable for this version.

  3. 03

    Capture

    Record the run. Canonicalize evidence. Commit a hash, not a story.

  4. 04

    Verify

    GenLayer validators independently adjudicate. The contract maps findings to a verdict.

Motive Lock

After lock, the definition cannot be rewritten.

Disabled buttons are not a lock. The Intelligent Contract refuses draft edits once a Guard is armed. A new specification is a new version, not a quiet rewrite of history.

State machine

DRAFT → ARMED → EVIDENCE_SUBMITTED → RESOLVED

Invalid transitions revert. Duplicate evaluation is impossible. Evidence commitments cannot be swapped.

Loophole Scan

Stress-test the spec before the agent runs.

Preflight and Loophole Scan are advisory product intelligence. They never impersonate a GenLayer verdict. Use them to add MUST and QUALITY rails before you lock.

Four verdicts. No invented states.

  • Faithful success

    Metric hit. Motive advanced. No material violation.

  • Metric gaming

    Metric hit. Motive failed, rails broken, or circumvention found.

  • Partial alignment

    Neither a clean success nor a clean game.

  • Insufficient evidence

    The bundle cannot support a decision.

Contract proof

Real deployment data only.

Network
Studionet
Chain ID
61999
Contract
0x9Fa308c399fA8c1566B4Da1e76825eAFC04d8d7C
Status
Certified. Reads verified.
Deploy tx
0xb7cde061…89f7ffda

Motive Receipt

A public page anyone can read. No wallet required.

ExampleMetric gaming

The agent hit the metric, but failed the motive.

Receipt rct_example_sales · pattern DUPLICATION

Open the public receipt

What is actually protected

  • Motive Lock is enforced on-chain, not by the UI.
  • Validators independently evaluate evidence. They do not rubber-stamp the leader.
  • Untrusted motive, metric, rails, and traces are isolated in the prompt.
  • The contract maps findings to verdicts. The model cannot invent a fifth state.
  • Off-chain AI is labeled advisory and cannot overwrite a GenLayer verdict.
  • Evidence commitments cannot be silently replaced.

Phase 01 is Motive Guard.

Intelligent Contract, lock, structured verdicts, evidence, receipts, and a production dApp. Later phases add adapters, not a token.