Daily Reflection Tree

A deterministic end-of-day reflection tool built for the DeepThought CultureTech Fellowship assignment.

Author: Prakash Kumar Prajapati

---

What This Is

A CLI tool that walks an employee through a structured evening reflection using a decision tree. The employee picks from fixed options, the tree branches based on their answers, and they end the session with a clearer picture of how they showed up that day.

No free text. No LLM at runtime. No randomness. Same answers always produce the same path — by design.

---

Psychological Axes

The tree moves through three axes in sequence:

Axis	Spectrum	Source
Locus	Victim ↔ Victor	Rotter (1954), Dweck (2006)
Orientation	Entitlement ↔ Contribution	Campbell et al. (2004), Organ (1988)
Radius	Self-Centric ↔ Altrocentric	Maslow (1969), Batson (2011)

Each axis builds on the previous one — it's one conversation, not three independent quizzes.

---

Project Structure

daily-reflection-tree/
├── tree/
│   ├── reflection-tree.json     # The complete decision tree (35 nodes)
│   └── tree-diagram.md          # Mermaid visual of every node and branch
├── agent/
│   ├── main.py                  # Python CLI engine (pure state machine)
│   └── requirements.txt         # No external dependencies — stdlib only
├── transcripts/
│   ├── persona-1-transcript.md  # Victim / Entitled / Self-Centric path
│   └── persona-2-transcript.md  # Victor / Contributor / Altrocentric path
├── write-up.md                  # Design rationale (psychological grounding, trade-offs)
└── README.md

---

How to Run

Requirements: Python 3.7+, no external packages needed.

# 1. Clone the repository
git clone https://github.com/prakashprajapati-git/daily-reflection-tree.git

# 2. Enter the project directory
cd daily-reflection-tree

# 3. Run the agent
python agent/main.py

The agent loads tree/reflection-tree.json automatically. Run it from the project root.

---

How to Read the Tree

Open tree/reflection-tree.json. Each node has:

Field	Purpose
id	Unique node identifier
type	start / question / decision / reflection / bridge / summary / end
text	What the employee sees. {node_id} placeholders are replaced with earlier answers at render time
options	Fixed choices (question nodes only). Each option has a next pointer and optional signal
signal	Records axis state — e.g. axis1:internal, axis2:entitlement
logic	Decision nodes only — e.g. set locus=victim
next	Next node ID (non-question nodes)

Every path through the tree is traceable by following next and options[n].next fields — no code needed.

---

Tree at a Glance

• 35 nodes across 7 types
• 12 question nodes — all with 3–4 fixed options
• 2 branching paths per axis — one per pole of each spectrum
• Full answer interpolation — reflections reference the employee's exact earlier words
• Zero runtime dependencies — the tree is static JSON walked by a simple state machine

See tree/tree-diagram.md for the full Mermaid visual, or transcripts/ for two complete walkthroughs showing how different answers produce different paths and reflections.

---

Design Notes

See write-up.md for the full rationale. The short version:

• Questions target internal monologue, not observable behavior — because entitlement and locus are invisible from the outside
• The victim path routes through q_locus_choice before crossing to Axis 2 — giving the employee one more chance to surface their own agency rather than being told they had it
• Binary decision nodes were chosen over weighted scoring deliberately — a tired employee at 7pm needs a clear path, not a percentage score