Alt Personas

The people agents act for, as evidence, not empathy.

Each persona is a set of parameters enforced where the agent acts: patience, reading level, dexterity, input mode, device and network. Precision is published per persona, confirmed by paid testers from the groups we model.

first-timer-mobileAt launch

First-timer on mobile

Never visited, small screen, reads everything once.

patience 5 · techLiteracy 4 · device iPhone 13 · network 4G

careful-evaluatorAt launch

Careful evaluator

Reads policies before committing; abandons on ambiguity.

patience 8 · riskAversion 9 · readingLevel C1

impatient-shopperAt launch

Impatient shopper

Skims, taps the first plausible button, gives up after 2 dead ends.

patience 2 · goalFocus 9

elderly-low-dexterityAt launch

Older adult, low dexterity

Needs large targets and forgiving forms. Least valid simulation, human-confirmed.

dexterity 3 (min target 44px) · vision low · patience 6

non-native-b1At launch

Non-native reader (B1)

Reads English at CEFR B1; idioms and dense legal copy cause errors.

readingLevel B1 · language es-ES

keyboard-onlyAt launch

Keyboard-only

No pointer. Finds focus traps and unreachable controls.

inputMode keyboard-only

a11y-tree-probeQ1 2027

Accessibility-tree navigation probe

Navigates only through the accessibility tree. A probe, not a simulation of blind users.

inputMode a11y-tree · vision none

cvd-deutanQ1 2027

Colour-vision deficiency (deutan)

Colour-only signals are invisible.

vision cvd-deutan

low-bandwidth-3gQ1 2027

Low bandwidth (3G)

Slow network exposes layout shift and timeouts.

network 3G · device Moto G

distracted-multitaskerQ1 2027

Distracted multitasker

Leaves and returns mid-flow; tests session persistence.

patience 4 · priorState interrupted

power-userQ1 2027

Power user

Uses search, filters and shortcuts; expects speed.

techLiteracy 9 · patience 3

returning-customerQ1 2027

Returning customer

Signed-in state, saved details, reorder flows.

priorState signed-in

What the research says.

Simulated users are too polite and inflate success (CMU Sim2Real); older and non-native personas are the least valid (Lost in Simulation). That is why those personas are human-confirmed and flagged in every report.

Published precision.

Confirmed ÷ flagged, per persona, with n and a 95% confidence interval. First publication Q1 2027. How it is measured

See where agents fail on your funnel.

Run one free cell on a property you own. No sales call, no card.