Explainables
Created with the help of AI
Category — Govern

Decision Timeline

audience adaptationprovenance
User question

What happened and when?

Consulting signal

Comes up when a client faces a challenge or complaint and can't reconstruct what happened: who reviewed the case, when, in what sequence, and with what information available at the time.

Overview

Why this pattern exists

Decisions are not moments: they are processes. A loan decision involves: data collection, model scoring, uncertainty assessment, caseworker review, override consideration, documentation, and communication. A content moderation decision involves: automated flagging, policy matching, human review (if any), and outcome. Each of these steps happens at a specific time, by a specific actor (human or AI), and with a specific rationale.

When users receive a decision, especially an adverse one, they often don't know any of this. Was the decision made by an AI or a human? Was there a review? Was additional information requested? Was there an appeal already in progress?

Decision Timeline makes this process visible. It answers not just "what was decided" but "how the decision was reached": a sequence of events that can be understood, audited, and, where needed, contested.

This is a governance pattern: its primary users are often auditors, regulators, and compliance teams who need to reconstruct a decision after the fact. But it is also meaningful for the people affected by decisions, who have a legitimate interest in understanding the process that affected them.

Design goal

Provide a clear, chronological view of the key events in a decision's lifecycle, including data collection, model assessment, human review, overrides, and communications, in a form that supports both user understanding and audit.

Usage guidance

When to use

  • Decisions are complex, multi-step processes involving both automated and human components
  • Users or affected persons have asked or have a right to understand "what happened"
  • Audit and compliance requirements demand a traceable decision record
  • Appeals or contestability workflows require a reviewable process history
  • Multiple actors (AI system, caseworkers, reviewers) contributed to the decision

When not to use

  • The decision is a simple, single-step automated process with no meaningful sequence of events
  • The timeline would expose confidential internal process details that should not be disclosed to the affected person
  • The decision is so recent that the timeline contains only one entry: a timeline with one step is not useful

Design

UI primitives

Data Visualization / Timeline

Vertical event timeline

A chronological list of events presented as a vertical timeline with: - Date and time - Event type (data received, model scored, caseworker reviewed, decision issued, etc.) - Actor (automated system, caseworker name or role, applicant) - Brief description of what happened

Inline Signal / Icon Set

Event type icons

Visual differentiation between event types: - 🤖 Automated step - 👤 Human action - 📄 Document event (submitted, received, requested) - ⚠️ Flag or alert - ✓ Decision or resolution

Contextual Overlay / Expansion

Expandable event detail

Each timeline entry can be expanded to show more detail: what data was available at that point, what the model score was, what the caseworker noted. Primary view stays compact; detail is accessible.

Inline Signal / Indicator

Current status indicator

A clear "where we are now" marker showing the current state of the case: useful when a decision is still in process (pending review, awaiting documents).

Data Visualization / Highlight

Decision point highlight

Visual emphasis on the specific moment a final decision was made and by whom: distinguishing the decision event from the surrounding process events.

Navigation & Flow / Export

Timeline export

For audit purposes: a downloadable record of the full decision timeline, including timestamps and actor references. Used by auditors and compliance teams.

How to use

Show the actor for every step.

A timeline that shows "assessment completed" without indicating whether this was automated or human is not auditable. Every event should have an identified actor: the system, the caseworker, the applicant.

Use plain language for event descriptions.

A timeline entry that says "POST /api/score v2.3 returned 0.71" is not useful to the affected person. "Automated assessment completed: risk score generated" is.

Distinguish automated from human steps visually.

The line between AI and human action is often the most important distinction in a decision process. Use consistent visual coding to make this clear throughout the timeline.

Include the applicant's own actions.

A timeline that only shows organizational events obscures the full picture. When the applicant submitted documents, requested review, or updated information: these are events too.

Show the timeline to affected persons in simplified form.

The full technical audit trail is for compliance teams. Affected persons need a simplified version: "Here's what happened with your application, in the order it happened."

Use cases

flow a

Applicant reviewing what happened

  1. 1. Applicant receives a declined decision and wants to understand the process.
  2. 2. They open "What happened with my application?"
  3. 3. Timeline shows: application submitted (by applicant) → automated assessment completed (system) → additional documents requested (caseworker) → documents received (applicant) → case reviewed (caseworker) → decision issued (caseworker).
  4. 4. Applicant can see that a human was involved and when. They may choose to request review if they have new information.
flow b

Auditor investigating a complaint

  1. 1. Regulator requests the full decision record for a specific case.
  2. 2. Auditor accesses the audit timeline view.
  3. 3. Full timeline shows all events with timestamps, system version references, data snapshots, and caseworker notes.
  4. 4. Auditor identifies that the caseworker override was not documented with a reason: a process compliance issue.
  5. 5. Auditor files a finding.
flow c

Caseworker reviewing a complex case

  1. 1. Caseworker picks up a case that has been in process for 2 weeks.
  2. 2. Timeline shows: the initial assessment, two document requests, one response from the applicant, a partial update, and a pending review step.
  3. 3. Caseworker understands the full history without reading through case notes: the timeline provides the context needed to pick up where the previous reviewer left off.

Design trade-offs

Completeness vs. comprehensibility

A fully complete audit timeline can contain dozens of technical events. The display must be designed for two different audiences: affected persons (simplified, plain-language) and auditors (complete, technical). Consider two different views.

Transparency vs. confidential process details

Some process details should not be disclosed to affected persons: reviewer notes, internal escalation discussions, business rules. Design the disclosure boundary carefully.

Real-time vs. retrospective

Timelines are most valuable for retrospective review. But for cases currently in progress, a real-time status view (where is my application now?) is what users actually need. Design for both use cases.

Connections

Relation to other patterns

Sources

Kroll et al. (2017) — Accountable Algorithms

argues that algorithmic accountability requires process transparency: the ability to trace how a decision was reached, not just what the decision was. The decision timeline is the design implementation of this accountability requirement

examines algorithmic accountability from a computational journalism perspective, distinguishing transparency about a system's existence, inputs, logic, and outcomes. Supports traceable, reconstructable decision processes as a precondition for accountability. Published in Communications of the ACM, 59(2)

European Union (2024) — EU AI Act, Article 12

requires record-keeping for high-risk AI systems, including logging of decisions, to enable post-hoc audit

Explainables
Created as a side project by Christian Laesser & AI