synthetic-users

Build one markdown profile per persona covering overview, goals, functional/emotional/social jobs, grouped pains with intensity, gains, and behavioural motivations. Pair each profile with a behaviourally encoded journey map CSV, then assemble the facilitator config and run the panel.

Run a synthetic user panel against the brief: pose questions to 3-6 personas, collect in-character responses that lead with objections, and read the facilitator summary for deal-breakers, conflicts, and validation priorities to take into real user research.

No. Synthetic users produce observations, patterns, and hypotheses only, never insight or validation. Output must be labelled synthetic, kept out of insight ledgers, and used to aim real interviews and research where they matter most.

Each persona needs a profile markdown file and a journey map CSV encoding stage metadata, active jobs, pain intensity, emotional residue, identity tension, bias triggers, and decision thresholds. With thin data, a proto-profiles prompt generates starter archetypes flagged as grounded, inferred, or speculative.

LLMs default to sycophancy and idealism when roleplaying customers. The Anti-Idealism Protocol counters this by requiring 2-3 specific objections per response, banning filler praise, enforcing bounded rationality, and tagging speculative reasoning in an assumption trace.

Use 3 to 6 personas per panel. Fewer than 3 loses the disagreement that makes a panel useful, and more than 6 blurs the focus group into noise. Differentiate personas behaviourally rather than demographically.