verify

package
v0.1.3 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Jul 31, 2026 License: Apache-2.0 Imports: 6 Imported by: 0

Documentation

Overview

Package verify is the oracle-as-referee: it wires the runner (which produces RunResults on the authoritative target) to the pure oracle evaluator, and records every judgment as an experiment in the Datastore. The agent proposes; only the oracle here disposes. Keys, inference, and the oracle run stay client-side.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

This section is empty.

Types

type Request

type Request struct {
	RunID        string
	HypothesisID string
	Model        string   // model that produced this PoV (for provenance)
	ToolHashes   []string // content hashes of provisioned tools used

	Spec oracle.Spec // the oracle to apply

	// Base is the run template for the authoritative (vulnerable) target. The
	// PoV is injected before running.
	Base runner.RunSpec
	// Fixed, if set, is the run template for the differential fixed image.
	Fixed *runner.RunSpec

	PoV []byte
}

Request is a single PoV submission for judgment.

type Result

type Result struct {
	Verdict      oracle.Verdict
	Primary      oracle.RunResult
	Fixed        *oracle.RunResult
	ExperimentID string
}

Result carries the verdict plus the raw observations and the experiment id under which they were recorded.

type Verifier

type Verifier struct {
	Runner runner.Runner
	Store  *store.Store
}

Verifier runs PoVs and evaluates them. It is the only component that both executes the target and reaches the verdict, and it is deliberately code, not an agent.

func (*Verifier) Verify

func (v *Verifier) Verify(ctx context.Context, req Request) (Result, error)

Verify runs the PoV on the authoritative target (and the fixed image, in differential mode), evaluates the oracle, and records the experiment. The returned verdict is a fact, not an opinion.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL