Branching Decision Tree

Platform Architecture

Two Phases. One Platform.

The BDT platform separates authoring from runtime — SMEs design games in the Design Studio; learners play them in the Player.

Phase 1

The Design Studio

Subject-matter experts use a two-phase authoring flow. First, a guided 7-step intake wizard collects everything the AI needs: course identity, learning objectives, cohort profile, session structure, active mechanics, and scenario seeds.

Once intake is complete, the studio transitions to a split-screen hybrid: a chat interface on the left (LLM-powered authoring conversation) and a live visual tree on the right that updates in real time as nodes are generated.

› 7-step guided intake — objectives, cohort profile, session structure, scenario seeds
› Split-screen hybrid mode — chat + live node graph updating in real time
› Quality flags on every node — review-required, review-recommended, approved
› Just-in-Time mode — AI auto-generates full configuration without step-by-step conversation

I've analyzed your intake. The scenario involves a field coordinator facing competing demands from local partners. Here's my proposed tree architecture — 3 branches from the initial crisis node…

Can we add a fourth branch where they escalate to HQ first?

Good instinct. Escalation as a first move is defensible — adds complexity around authority and timing. I'll add Node 1D: "Escalate to HQ before acting." Quality flag: approved.

Node 1 · Crisis Arrives

1A · Negotiate directly

1B · Buy time, gather data

1C · Defer to local lead

1D · Escalate to HQ ✦ new

Phase 2

The Runtime Player

Learners enter the player and are immediately placed in the scenario. No preamble. The situation is real; every choice has downstream consequences. The LLM responds dynamically to their decisions, adapting the consequence narrative to what they chose — and what that reveals about their judgment.

Option gating is active throughout: prior choices restrict or expand what's available later. Learners who make a naive move early find certain responses unavailable — not locked arbitrarily, but because the situation has evolved.

› LLM-powered dynamic consequences — narrative adapts to each learner's path
› Option gating — prior choices expand or restrict later options
› Structured debrief — AI-facilitated reflection embedded at the end of each session
› Session analytics — full path telemetry captured for facilitator review

Three days before the community workshop, your local partner calls. She has learned that two influential male community leaders plan to attend — not to support the program, but to observe and potentially object. Your female participants may not attend if the men are present.

How do you respond to this new information?

Hold a private meeting with the male leaders to understand their concerns before the event

Proceed as planned — their attendance is public space and cannot be restricted

Reframe the event as a mixed-gender dialogue, changing the participant list

Postpone and consult your organization's country director

Platform Capabilities

What the BDT Platform Does

🧠

No Obvious Correct Answers

Every option at every decision node must be defensible by a reasonable professional. The BDT reveals judgment, not knowledge recall.

🔀

Three Branching Structures

Tree (divergent), Diamond (convergent), and Network (non-linear) — each with different learning dynamics and cognitive demands.

🔗

Option Gating

Prior choices restrict or expand future options. The scenario evolves with the learner, creating real consequence across the session arc.

💬

AI-Powered Debrief

After each session, the AI facilitates a structured reflection using the learner's actual path — not generic feedback, but a response to what they specifically chose.

📊

Analytics Dashboard

Facilitators see cohort-level patterns: where groups diverge, which branches reveal systemic blind spots, and who chose what at which decision point.

🤖

Model-Agnostic LLM Layer

Supports Anthropic, OpenAI, and Gemini. Model assignment is configurable per phase — different models for generation, dynamic responses, and debrief.

Example Runtime Scenario

Afghan Women's Leadership Program

A humanitarian field coordination simulation designed for NGO program managers, country directors, and senior field staff. Learners take the role of a program coordinator navigating a women's leadership initiative in a complex, high-stakes environment.

Every decision node presents a situation that experienced practitioners would genuinely disagree on — balancing program integrity, community trust, organizational protocol, and participant safety across 5 consequence-rich cycles.

The scenario was built on the BDT Design Studio using real-world humanitarian programming contexts, and runs entirely on the BDT Player.

Play This Scenario ↗

Scenario Profile

Domain: Humanitarian field programming · NGO operations

Audience: Field coordinators, program managers, country directors with 3–10 years' experience

Stakes: Program continuation, community trust, participant safety, organizational accountability

Structure: 5 cycles · Network branching · AI-powered dynamic consequences · Structured debrief

Gender programming Community engagement Field coordination Stakeholder navigation Crisis response