Smart Delegation

How it works

A verdict many validators agree on.

Choosing a validator mixes hard numbers with judgement calls. Smart Delegation puts both into one GenLayer Intelligent Contract, so the verdict is reproduced and agreed by independent validators instead of trusted from a single model or operator.

01 · Mandate

Rules, on-chain

A public list of up to 8 criteria, e.g. “stake share ≤ 10%” and “a security contact is published”.

02 · Leader evaluates

Two kinds of check

CodeObjective: computed from WatchLayer facts
LLMSubjective: judged from the validator's own website and GitHub
03 · Validators re-run

Independent replay

Every validator fetches the same sources and redoes the whole evaluation with its own LLM.

L✓✓✓✓
04 · Agreement

Verdict stored

Accepted only if eligibility matches; per-criterion results, reasons and evidence links are written on-chain.

The details

[02]
Objective · code

Numbers are never guessed

Criteria such as obj:stake_share_bps<=1000 are checked deterministically against facts from WatchLayer: stake, share of network stake, ban status and history, operator activity. A missing fact fails the criterion.

Subjective · LLM

Judged from identity pages

The model reads only the validator's registered website and GitHub and answers per criterion with a reason. A pass without an evidence URL from those pages is turned into a fail by code.

Consensus

What has to match

  • The eligible / not eligible outcome, exactly.
  • Every objective criterion, exactly.
  • At least n−1 of n criteria overall.
  • Free-text reasons are explanations and are not compared.

Then you delegate

[03]

The results page proposes a split of your GEN across the eligible validators, proportional to their current stake. You sign each delegation (Staking.delegatorJoin) from your own wallet; the app never holds funds. Verdicts are cached for 24 hours.