Error Forgiveness

Canonical
NormanClassic HCIEthical Design

Confidence

79%

Cognitive Load

Low

Evidence

production validated

Impact

component

Ethical Guardrail

Never blame the user. Never force permanent actions without undo. Never hide the recovery option.

Design Intent

Humans make mistakes. Good design anticipates and forgives them instead of punishing the user. Error Forgiveness prevents errors where possible, catches them early, and makes recovery effortless and reversible.

Psychology Principle

Humans make mistakes. Good design anticipates and forgives them instead of punishing the user.

Description

Prevent errors where possible, catch them early, and make recovery effortless with one-click undo and kind messaging.

When to use

Every action that could cause harm or data loss -- form submissions, deletions, sends, configuration changes.

Example

Gmail Undo Send: After sending, a clear banner appears for 30 seconds with one-tap Undo -- no extra steps required.

Autonomy Compatibility

SuggestConfirm

Behavioral Objective

Users recover from mistakes instantly and continue their task without frustration or abandonment.

  • Increased willingness to experiment
  • Higher overall completion rates
  • Stronger trust in the system

Target Actor

role

Everyday user

environment

Fast-paced, multi-tasking

emotional baseline

Fear of breaking things or losing work

ai familiarity

medium

risk tolerance

low

Execution Model

1

prevention

Stop errors before they happen.

Error occurs that could have been prevented.

2

detection

Catch errors the instant they occur.

User only discovers the error after many steps.

3

recovery

Make fixing the error trivial.

User must manually fix the mistake with multiple steps.

Failure Modes

Prevention feels restrictive

Use soft warnings instead of hard blocks

micro

Recovery path is hidden or complex

Always surface undo prominently

micro

Over-forgiveness encourages careless behavior

Reserve strong forgiveness for low-stakes actions

feature

Blaming language damages trust

Use neutral helpful tone only

micro

Irreversible actions slip through

Require explicit confirmation + audit trail

feature

Agent Decision Protocol

Triggers

  • Error rate is high in a flow
  • Users express fear of making mistakes
  • User says 'I accidentally did X and now I'm stuck'

Escalation Strategy

L1: Diagnose error type and recovery gap via behavioral_signals

L2: Nudge -- surface undo prominently and add inline validation

L3: Restructure -- add prevention layer with smart defaults and confirmation

L4: Constrain -- enforce auto-save and version history for all destructive actions

L5: Yield -- flag for human UX designer review

Example

User deletes a file -> agent immediately shows 'Undo delete' banner with one-tap restore.

Behavioral KPIs

Primary

  • Error recovery success rate
  • Time to recover from error
  • Flow abandonment after error

Risk

  • User frustration reports after mistakes
  • Permanent data-loss incidents

Trust

  • User confidence that 'I can't break anything'
  • Autonomy Dial usage when agent prevents or forgives errors

Behavioral Signals

error_occurred

error_detected=true AND error_recovered=false AND time_since_error > 30s

undo_used=true AND repeat_error_count > 2

recovery_success

error_detected=true AND undo_used=true AND time_to_recover < 5s

error_recovered=true AND flow_continued=true

prevention_working

validation_warning_shown=true AND error_prevented=true

confirmation_dialog_shown=true AND action_cancelled=true

Decay Monitoring

Revalidate when

  • New destructive actions are added to the product
  • User base includes more novice or high-stakes users
  • Platform undo capabilities evolve

Decay signals

  • Rising permanent-error incidents
  • User complaints about 'no undo'
  • Drop in willingness to experiment

Pattern Relationships

Related Patterns

Canonical Implementation

Gmail Undo Send: After sending, clear banner for 30 seconds with one-tap Undo -- no extra steps

Telemetry Hooks

error_detectederror_recoveredundo_used

Tags

hci-coretrustsafety