Provable · the receipt

Live demo — site × scenario × rendering

Proves: each variant is injected onto the live-cloned page where it maps, carrying the data it used and its surface policy. Open one to drill into its receipt.

Scenario A · identity key: IP + browser signals (no identity)

First hit. No login, no form, no cookie — just the HTTP request and a little JS.

available data: geo / citydaypartdevicereferrer / UTMconnectionreverse-IP company (buy)

A1 Daypart adaptive theme ↑ bounce

Maps to Page theme · data used 1 daypart 11:47pm · Observed at load · allude 2 IP → Austin · Observed at load · allude 3 device: mobile · Observed at load · allude 4 UTM creative · Observed at load · allude

observed · allude · no name. Drill into the receipt →

A2 B2B peer-logo hero ↑ b2b_demo

Maps to Hero · data used 1 reverse-IP → employer · Bought (reverse-IP) · allude 2 company industry · Bought (reverse-IP) · allude

bought · allude · company, not person. Drill into the receipt →

A3 Local-proof badge ↑ ctr

Maps to Banner · data used 1 IP → city (Austin) · Observed at load · allude

observed · allude. Drill into the receipt →

Blocked variant · Ax · Recognize-return (ungated) · maps to Banner

“Welcome back — third look this week?”

HOLD — blocked by surface policy; the bandit cannot pull it. It would win on engagement, but it relies on a hold fact, so it is kept out of the action pool — the optimizer can never select it. Open it, greyed →