Description
A six-stage mixed-methods loop (Predict, Diagnose, Prove, Forecast, Prescribe, Validate) that turns backward-looking findings into predictions, prescriptions and validated outcomes tied to decision bars.
The loop answers a common failure of UX reporting: statements such as 'users struggled with navigation' that do not tell teams what to change or what impact to expect. Each stage has paired quant and qual methods: Predict (CPM-GOMS, mental models), Diagnose (tree testing, error taxonomy, signal-to-noise ratio), Prove (A/B, Cox survival, quasi-experiments, QCA), Forecast (Monte Carlo, SHAP), Prescribe (joint display, effort × impact) and Validate (confirmatory A/B, guardrails, field drift checks). The loop is a menu rather than a checklist: researchers pick the strongest feasible method per stage, document assumptions, and tie results to decision bars set in advance. Its rigour comes from evidence building across stages, so a hypothesis from Predict is tested in Diagnose and Prove, its impact forecast, its fix prescribed and the forecast compared with actuals. A Miles & Huberman qualitative cycle runs alongside to keep qual auditable.
Attributes
- Discipline
- Community-Proposed Frameworks
- Loss type
- Community-proposed
- Tested
- Not yet tested
- Stage
- Strategy, Evaluation
- Data
- Mixed
- Effort
- Months
Steps
- Predict: model the task with CPM-GOMS and capture 4–8 mental-model interviews to produce a falsifiable hypothesis and decision bar (e.g. predicted ideal 6s, breakdown if observed median exceeds 15s).
- Diagnose: run an unmoderated tree test (N = 80–150), 10–15 moderated think-alouds coded as Slip/Lapse/Mistake (κ ≥ 0.70), and an element and behavioural SNR audit; synthesise in a joint display.
- Prove: test the fix causally with a randomised or prototype A/B and a Cox proportional hazards model of time to abandonment; use QCA, Rubin matching or Bradford Hill criteria when randomisation is not possible.
- Forecast: simulate 10,000 virtual journeys with Monte Carlo from step probabilities and rank drivers with SHAP on a held-out gradient boosting model; pair with diary-based sentiment trajectories.
- Prescribe: map each driver to a mechanism and a fix using a 'because → therefore' structure, place fixes on an effort × impact grid, and attach a decision bar to each.
- Validate: run a confirmatory A/B with pre-registered bars, monitor guardrails (latency, crashes, tickets), re-run the evidence array to confirm the recursive loop is broken, and check drift at 1, 3 and 6 months.
- Report forecast versus actual (e.g. 'predicted +11% lift, observed +14%, sustained at +12% after 3 months') and iterate the loop.
Sources
- Vivek Gopinath (2025), Forward-Thinking UX Research Playbook: A Mixed-Methods, Decision-Ready Approach from Prediction to Validation
- Vivek Gopinath (2025), Tree Testing and Error Taxonomy: A Complete Chapter for UX Researchers
- Studio Aghora
- uxguy.in
- Forward-Thinking Playbook (Gopinath)
- Nash Equilibrium Playbook (Gopinath)
- Behavioral Drift Playbook (Gopinath)
- Rittel, H. & Webber, M. (1973) Dilemmas in a General Theory of Planning
- The Inquiry Series, Ch 60 — The Forward-Thinking Research Loop — Predict → Diagnose → Prove → Forecast → Prescribe → Validate
- Forward-Thinking Playbook (Vivek Gopinath)
- Nash Equilibrium Playbook (Vivek Gopinath)
- Behavioral Drift Playbook (Vivek Gopinath)
Also known as
- Six-stage forward-thinking loop
- Predict → Diagnose → Prove → Forecast → Prescribe → Validate
- Forward-Thinking Research Loop
- Forward-Thinking Playbook
- Predict-Diagnose-Prove-Forecast-Prescribe-Validate loop
- Forward-Thinking Research Loop (integrated playbook application)
- Full 6-stage loop integration
Categories
- Community-Proposed Frameworks