Strict
Use when the answer is a fact. Did the flight land? What was the closing price? Validators must return byte-identical JSON. No interpretation permitted.
gl.eq_principle.strict_eq(
"flight AA42 status"
)
Exhibit A — Interactive Demonstration
The Jury.
When code can't decide,
five AIs can.
Smart contracts move money. Intelligent Contracts decide who deserves it. Pose a subjective question, watch five LLMs deliberate as GenLayer validators, and see the Equivalence Principle deliver a verdict — live, in your browser.
Live on GenLayer testnet-bradbury — every verdict is real on-chain validator consensus, verifiable on the block explorer.
Filed · GenLayer Builder Program
Educational Content · v0.1.0
Exhibit B§02 / 07
How the tribunal works.
Three primitives, working in sequence. The contract proposes. The jury deliberates. The principle decides. Everything else is implementation.
/ 01
Intelligent Contracts
Python contracts that call LLMs and read live web pages. Same on-chain guarantees as Solidity, plus the ability to reason about subjective inputs no oracle can answer.
contract → llm.exec_prompt() → state
/ 02
Optimistic Democracy
A leader executes first. A small jury of validators re-runs and votes. Disagreement triggers an appeal; appeal doubles the jury. Finality is reached when the supermajority agrees.
leader → validators (5) → appeal (11) → finality
/ 03
Equivalence Principle
The rule the author writes that tells validators what "agreement" looks like. Strict equality for facts. Numeric tolerance for measurements. Rubric judgment for everything else.
strict · comparative · non-comparative
Exhibit C§03 / 07
Three modes of agreement.
The Equivalence Principle isn't one rule — it's three. Choosing the right one is the entire skill of writing an Intelligent Contract.
Comparative
Use when the answer is a measurement. How late was the delivery? What was the sentiment score? Validators must agree within a tolerance window.
gl.eq_principle.prompt_comparative(
tolerance=0.15
)
Non-comparative
Use when the answer is a judgment. Did the freelancer fulfill the contract? Is this proposal valid? Validators apply a rubric the author wrote and a sixth LLM judges agreement.
gl.eq_principle.prompt_non_comparative(
rubric="..."
)
Verified on-chain registry
Every verdict on The Jury is reached by GenLayer validators on these deployed intelligent contracts. Click any address to inspect the contract on-chain.
Live · Bradbury
Exhibit D — Live System State§04 / 07
The jury convenes.
Pose a question. Pick a mode. Preview how five model jurors might reason — then send the case to GenLayer, where the official verdict is decided on-chain.
Select a case
Equivalence Principle
Case · 24·001
Pattern: freelancer milestone
Question on Trial143 / 500
"A freelancer was paid $800 to deliver a 10-slide pitch deck by Friday. They delivered an 8-slide deck on Monday. Did they fulfill the contract?"
Reasoning PreviewEducational analysis · Not authoritative
This preview is generated locally for educational purposes. The official verdict is produced by GenLayer Intelligent Contracts on-chain.
GPT-OSS 120BSeat 01
AwaitingAwaiting summons.
Llama 3.3Seat 02
AwaitingAwaiting summons.
Gemma 4 31BSeat 03
AwaitingAwaiting summons.
Dolphin MistralSeat 04
AwaitingAwaiting summons.
Qwen 3 80BSeat 05
AwaitingAwaiting summons.
0
Accept0
Reject0
Undet.5
PendingThe jury awaits summons.
The Casebook§05 / 07
Five cases on file.
Each case shows the same path: the dispute, the jury's verdict, which mode fits, why Ethereum couldn't. Open one to leave with judgment, not just awe.
In Docket
05
№ 24·001Non-comparativeExpected · reject→№ 24·002StrictExpected · accept→№ 24·003Non-comparativeExpected · reject→№ 24·004Non-comparativeExpected · reject→№ 24·005Non-comparativeExpected · reject→
The freelancer's milestone on file.
A freelancer was paid $800 to deliver a 10-slide pitch deck by Friday. They delivered an 8-slide deck on Monday. Did they fulfill the contract?
The flight's delay on file.
AA42 was scheduled to land at 14:00. The flight tracker shows it landed at 16:47. Did it land more than 2 hours late?
The DAO's proposal on file.
A DAO's charter restricts treasury spending to 'protocol development and community growth.' A proposal allocates $50K to sponsor a music festival. Does this fall within the charter's intent?
The prediction market on file.
A prediction market asks: 'Will GPT-5 ship before July 2026?' On June 30, OpenAI releases 'GPT-5-mini' but not the full GPT-5. Should the market resolve YES?
The AI agent's SLA on file.
An autonomous research agent was contracted to deliver 'high-quality citations from peer-reviewed sources.' It returned 12 citations: 8 from arXiv preprints, 3 from Medium, 1 from a podcast. Did it fulfill the SLA?
Exhibit F — Appeal & Finality§06 / 07
Appeals turn doubt into procedure.
A no-consensus result is not a dead end. The contract can demand more validators, require a stronger majority, and make each vote economically accountable before final state changes.
Appeals double the court01
A split panel escalates.
The first panel starts small so most cases finish quickly. If the result is inconclusive, an appeal summons a larger validator set and raises the agreement threshold.
Stake makes votes costly02
Validators have skin in the verdict.
Validators earn fees when their judgment aligns with the supermajority. Bad or lazy votes risk stake, so the incentive is careful reasoning rather than automatic agreement.
Finality is a threshold03
Agreement writes the outcome.
Once the appellate jury reaches the required supermajority, the verdict becomes final and the contract can move state with an auditable trail of why it happened.
Field Guide - Author WorkflowDecision tree for mode fit
Choose the mode with intent.
Intelligent Contracts fail when authors treat equivalence modes as cosmetic. They succeed when mode choice matches the evidence type, dispute shape, and confidence requirement.
01
Start with evidence shape
Mode choice is not about model preference. It is about the shape of the evidence you expect validators to return.
02
Escalate only when needed
A small jury resolves most disputes quickly. Appeals exist for hard edges where interpretation or noisy data blocks confident agreement.
03
Write the rubric like law
For subjective disputes, phrase the rubric like a legal clause. Clear scope gives validators room to reason without drifting from intent.
Exhibit G — Closing
File your own case.
Take the playground further.
Read the guide, or fork it in Studio.