Double Diamond (Deep)
The full design-thinking process as three linked diamonds: problem space (Discover broad -> Define key problems), solution exploration (Ideate broad + lo-fi -> down-select), solution refinement (mid-fi build -> Deliver tested solution presentation). Broad empathy, affinity clustering, and a lo-fi -> mid-fi fidelity ladder with real prototype tests.
When you want depth: many personas, broad problem exploration, clustered key problems, several lo-fi prototypes tested and down-selected, a refined mid-fi prototype tested, and a buildable solution presentation.
Phasen
Discover
Empathize broadly across DISTINCT ANGLES (e.g. the trigger moment, the mental model / language barrier, money reality, trust, a premise-check, the provider/insider view). Run a FEW real MULTI-persona councils — several segment-diverse personas reacting to each other — NOT one council per persona. Surface real lived problems, feelings, frictions; no solutions; anti-steering is strongest here (include a disconfirmation angle that tests whether the premise even holds).
Define
Affinity-cluster the many problem explorations into themes; name the KEY PROBLEM(S) and a sharp Point-of-View. Author payload.clusters [{label, member_node_ids, insight}] + payload.key_problems.
Ideate (Lo-Fi)
Generate many distinct solution concepts — push for genuine NON-OBVIOUS innovation, not incremental UX tweaks. Apply the creativity lenses in next_action.act.ideation_lenses (cross-domain analogy, make-the-invisible-EXPERIENCEABLE via an interactive model/simulation, provocation/reversal, extreme-user, mechanism-transfer, hybrid, subtract-the-obvious, honest anti-goal); at least ONE concept must carry a mechanism you did not expect at the start. Then build SEVERAL lo-fi prototypes DIVERSE IN KIND (next_action.act.artifact_palette: a guided flow, a comparison, an overview, a card interface, an interactive MODEL the persona steers — not N forms), including >=1 deliberately extreme 'dark-horse'. Go broad before judging.
Lo-Fi Test & Down-Select
Have personas USE the lo-fi prototypes (real sessions), then down-select. Author payload.ranking [{prototype_id, score_rationale}] + payload.shortlist (prototype_ids to carry to mid-fi).
Refine (Mid-Fi)
Build mid-fi prototype(s) of the shortlist and explore concrete refinements/variants. One exploration per refinement angle.
Deliver (Solution Presentation)
Personas USE the mid-fi prototype (real sessions); synthesize into the final SOLUTION PRESENTATION: the winning concept, who it wins (and deliberate non-targets), validated pain-solvers, evidence trail, open risks, and the buildable hand-off spec.