T21

See the scoped evidence_register and qualification_protocol. Structural plausibility, descriptive surveys and laboratory benchmarks do not alone identify a local commercial policy effect.

See the scoped evidence_register and qualification_protocol. Structural plausibility, descriptive surveys and laboratory benchmarks do not alone identify a local commercial policy effect.

Fit decomposition is useful when component observations are independently validated for the intended target. Retention reflects fit, contract terms, selection and switching; it is not a uniquely clean fit measure. The fit and business-kind layers already prevent some universal SaaS assumptions. Make their distinctions operational. An account can have good end-user fit but weak budget ownership, poor security acceptability, costly implementation, or negative supplier contribution. A PMF score should expose these observations and unknowns instead of absorbing them into one convenient multiplier.

• G2 — The Answer Economy — 15 April 2026. Primary report. Commercial survey; methods/findings inspected. March online sample of 1,076 B2B decision makers and 39 marketer interviews across regions. Reports AI research alongside continued Google use and frequent inaccuracies. Source: https://learn.g2.com/g2-2026-ai-search-insight-report
• Oded Netzer and Rajan Sambandam — Synthetic Data in Marketing Research How to Evaluate and When to Trust — 12 September, revised 19 September 2026. Record and PDF. Full methods inspected. Real 2025 US sample of 3,063 people, 18 sectors, 108 questions; grounded leave-one-question-out prediction and out-of-bag screening. Rich category histories improve agreement over demographics alone; 46 retained questions perform better than all questions. Source: https://arxiv.org/abs/2609.13995
• Kuric, Demcak and Krajcovic — Distorted Perspectives of LLM Simulated Preferences — 18 May 2026. Full preprint. Twenty-nine actual UX studies, 2,073 people, 78 tasks. GPT-4.1's top choice matches people in 53% of tasks; GPT-5.2 reaches 65% without an established general distribution-fit gain. Source: https://arxiv.org/html/2605.18311v1
• Leonard Kinzinger and Jochen Hartmann — Synthetic Personalities — 3 June 2026. Full preprint. Five hundred SOEP individuals, 183 held-out items, three models and 2.1 million generated responses. Grounding and preserving original responses help; reasoning improves ordering more reliably than closeness. Source: https://arxiv.org/html/2606.04592v1
• Stephen France and Pia Albinsson — Synthetic Consumer Insight Generation with Large Language Models — 7 July 2026. Record and full PDF. Full comparative methods inspected. Word associations and completions about five US cities, 173 students, six LLMs and varied prompting/seeding. Broad topics overlap while synthetic style differs. Source: https://arxiv.org/abs/2607.05761

• No fitted local coefficient or complete-policy deployment evidence is supplied.
• reported frequency and AI-attributed switching are not observed journey shares or causal switching. Sampling/weighting and question bases constrain universal adoption estimates. C1/C2/C3/C8.
• selected in-domain survey predictions, not novel procurement decisions. Question-level agreement differs from individual fidelity; generated replicas add no independent people. Commercial research affiliation disclosed. C1/C2/C6/C8.
• specific models/protocols and proprietary UX data, no procurement or transactions. Persona elaboration and reasoning do not consistently repair validity. C1/C2/C8.

Declare eligible population/business kind, target and denominator, pre-action information, policy/comparator/assignment, outcome horizon and coverage, cost/authority/configuration, uncertainty, causal assumptions, falsifier and expiry. Current coefficients/qualification remain unmeasured unless explicitly linked to local evidence.