The working files
The Pour-Over Protocol
The spreadsheets and notes this case study is built from, exactly as they were kept. Every number on the study page comes from one of these, so you can check it yourself.
Read me first
What is in this folder and how the files fit together.
Assumption map
Every belief the work rested on going in, rated for how sure we were and how much it mattered, and what became of it.
Traceability matrix
Each thing the design had to do, what was built to do it, and how it was checked.
Design outputs
Every screen and deliverable, and which need each one answers.
Diary protocol
The rules for keeping the diary, written before it started.
Observation protocol
The rules for what to watch and how to log it.
Decision log
Decisions as they were made, with the date and the reason.
Shortcut design
How the iPhone Shortcut that guided each cup was designed.
Blood pressure log
The daily blood pressure readings.
Cross-checks
Where two records of the same thing were compared against each other.
Diary entries
Each entry logged during the study, with its date.
Observations
What was seen firsthand, logged as it happened.
Phase summary
The totals for each phase of the experiment, in one file.
Doctor’s clearance
The doctor’s sign-off at the Week 4 check-in to keep going.
Predictions made in advance
What I expected to happen, written down before the results came in.
Weekly totals
The numbers added up week by week.
Read me first
Evidence Packet: Desirable Difficulty Case Study
Version: 1.0
Date: 2024-02-21
Designer: Robert Duebelbeis
Scope
This evidence packet supports the case study "Desirable Difficulty: Behavioral Design for Health Compliance." It was my own project, done on my own time while I was at World Wide Technology.
Problem: My doctor told me to cut back on coffee. She wanted me down to about one a day; I set my own target from that.
Intervention: A manual pour-over ritual that adds desirable difficulty (skilled friction). The goal was to make every cup by hand; in practice about half were (36 of 72 cups in Weeks 3 to 6).
OEC: Cups per week (primary) with satisfaction ≥4.0 SEQ (guardrail)
Contents
| File | Description |
|---|---|
01_readme.md |
This file: scope, OEC, version |
pre_registrations.csv |
Append-only hypothesis log (PRE-001) |
02_assumption_map.csv |
All assumptions with certainty and importance ratings |
03_trace_matrix.csv |
Design input, verification method, evidence link and status |
04_diary_protocol.md |
Daily log template, weekly check-in guide |
05_observation_protocol.md |
Session recording consent, coding rubric |
06_decision_log.csv |
Dated decisions with rationale |
07_automation_flow.md |
iOS Shortcut design: flow diagram, step detail |
diary_events.csv |
172 coffee_consumption events. A RECONSTRUCTED export, not the original device log (which was not kept). Generated to reproduce every weekly_aggregates column exactly, on plain 7-day weeks from 2024-01-06, so it runs to 2024-03-02. The dated records (observations, cross-checks, decision log) put the first by-hand cup on 2024-01-15 and the end on 2024-02-21. This file does not hide that spread |
observation_events.csv |
8 observation_session events with behavioral codes |
weekly_aggregates.csv |
Weekly roll-ups for all metrics |
phase_summary.json |
Phase comparisons and effect sizes |
cross_validation_log.csv |
Validation checks, diary against recordings. 4 failed: one missing cup (29 Jan) and three timing checks (31 Jan, 5 Feb, 7 Feb) |
bp_log.csv |
Weekly blood-pressure log, kept only as a safety check. Readings redacted for privacy |
physician_clearance.md |
Week 4 clearance (2024-01-27), on file and redacted for privacy |
Study Design
Type: Single-subject ABA (baseline, intervention, washout)
| Phase | Duration | Purpose |
|---|---|---|
| A: Baseline | Weeks 1 and 2 | Establish ground truth consumption |
| B: Intervention | Weeks 3 to 6 | Test desirable difficulty hypothesis |
| A: Washout | Weeks 7 and 8 | Test habit transfer |
Primary Outcome
| Metric | Baseline | By hand (Weeks 3 to 6) | Washout (Weeks 7 and 8) |
|---|---|---|---|
| Cups/week | 32 | 18 | 18 |
| Satisfaction (SEQ) | 3.69 | 4.37 | 4.50 |
| Cups made by hand | 0% | 50% | 42% |
Cups per week fell 43.75%. Satisfaction figures are means of weekly averages.
Falsification Criteria (Defined A Priori)
The hypothesis fails if:
- Weekly cups do not decrease by ≥20% by Week 4
- Satisfaction drops below 3.0/5.0 average
- Participant abandons before Week 6
- Consumption rebounds to baseline during washout
Result: All criteria passed.
What This Doesn't Prove
Look, n=1 is n=1. I'm not claiming this generalizes. But it does show:
- The method (daily logging + observation coding) is executable
- Pre-registration prevents p-hacking myself
- Effect held through washout (not just novelty)
Things I can't rule out:
- Hawthorne effect (being observed changes behavior)
- 8 weeks isn't forever
Notes
Data anonymized for portfolio. Raw video deleted per protocol.