​
What Can You Build?

Every scenario on Tacit is built from the same four parts:
  1. State - the facts of the situation, split into what is visible and what must be discovered
  2. A persona - the simulated person the situation revolves around
  3. Outputs - the decisions and artifacts the operator is expected to produce
  4. Criteria - the definition of what good looks like, written down before anyone runs the scenario
What changes between scenario types is how much of the situation is handed to the operator, and over how long. That gives you a ladder of three shapes. Each rung adds one dimension to the rung below it.
InputsDurationWhat it measures
Decision taskGiven up frontOne sittingJudgment
ConversationHeld by a persona, elicited turn by turnOne sessionJudgment + information gathering
JourneyElicited, and evolving over timeMany sessions across simulated weeks or monthsJudgment + information gathering + longitudinal management

Decision tasks

Case material in, a decision and its rationale out. The simplest shape, and the fastest to build.

Conversations

The facts are held by a persona. The operator has to earn them before they can decide anything.

Journeys

The same client over simulated weeks or months. Early choices change what later sessions look like.

​
Which shape fits your problem?

Your problem sounds likeBuild a
”Given this file, should we approve, deny, or escalate?”Decision task
”Are our people asking the right questions before they decide?”Conversation
”Does our chatbot handle a difficult customer correctly?”Conversation, with your agent in the operator seat
”Do our people manage a case well over its whole lifetime, not just the first call?”Journey
Start one rung lower than you think you need. A decision task that isolates the judgment is faster to build, faster to run, and easier to interpret than a conversation that wraps the same judgment in dialogue. Add the dialogue when you want to measure how the information gets gathered, not just what gets decided.

​
Who runs it is a separate question

Scenario shape says nothing about who sits in the operator seat. Any shape can be run by:
  • Your people - to capture how your experts handle it, or to measure a team at scale through cohorts
  • An AI agent - to benchmark a model, a prompt, or your own chatbot against the same criteria your people are measured on, in automated runs
Because both run the same scenario against the same criteria, the comparison between your people and your AI is direct: same situation, same definition of good, side by side.

​
Variants

A scenario can have variants: child scenarios that inherit everything from a parent and override specific fields. Use variants to create difficulty levels (the same case with a more guarded persona) or alternative angles (the same client, a different presenting problem) without rebuilding the scenario.

​
Next

Start with a decision task

The simplest shape, explained with a worked example

Scenario anatomy

The shared building blocks: personas, state, briefing, outputs, criteria