Caveats for the engagement-mechanisms top-up: strong independent evidence sits at the MECHANISM level not the business-outcome level; nearly every effect is moderated

The honest summary of source quality across this brief:

• Where the strong independent evidence sits: at the MECHANISM level — cognition, memory, motivation, communication/HCI lab and field studies. Curiosity (Kang, Camerer, Loewenstein et al. (2009), Psychological Science 20(8) — "Wick in the Candle of Learning": fMRI shows curiosity → caudate (reward) activity; better recall 1-2 weeks later; people spend tokens to satisfy curiosity, Gruber, Gelman & Ranganath (2014), Neuron 84(2) — high-curiosity states enhanced midbrain (SN/VTA) + nucleus accumbens activity; improved memory for target AND incidental information); generation effect (d ≈ 0.40 — Bertsch, Pesta, Wiscott & McDaniel (2007), Memory & Cognition 35(2) — 86-study generation-effect meta: d ≈ 0.40 ("almost half a standard deviation"); LARGER at longer retention (d ≈ 0.64 for >1 day)); active learning (+0.47 SD — Freeman et al. (2014), PNAS 111(23) — 225-study meta on active learning: exam performance +0.47 SD; odds of failing 1.95× higher under passive lecturing; robust to publication-bias checks); choice → intrinsic motivation (Patall, Cooper & Robinson (2008), Psychological Bulletin (41 studies meta) — choice enhances intrinsic motivation, effort, performance, perceived competence; moderated (2-4 choices, no extrinsic reward, children > adults)); agency in customization (Sundar & Marathe (2010), Human Communication Research — customization (user acts) vs personalization (system acts): the appeal of customization is tied to the user's sense of agency); modality-interactivity absorption (Oh & Sundar (2015), Journal of Communication 65(2) — N=167 factorial experiment: modality interactivity (slider) produced more positive interface assessment, greater cognitive absorption, more favourable attitudes); motivating-uncertainty (Shen, Fishbach & Hsee (2015), JCR 41(5) — Motivating-Uncertainty Effect: people invest MORE effort for an uncertain reward (50% $2 / 50% $1) than for certain HIGHER-expected-value reward — but ONLY under PROCESS focus).
• Independent evidence for specific business outcomes of interactive tools (conversion lifts, time-on-tool) is weak and largely vendor-sourced — do not lean on it. The vendor stats (Mediafly / Demand Metric: "Interactive content shows 52.6% higher engagement than static; buyers spend 13 vs 8.5 minutes" — vendor sources, treat as marketing not fact, Outgrow's "interactive forms 47.3% vs static 2.8%, a 16.9× improvement" is the vendor's analysis of its own customers' 50,000+ forms — not an independent benchmark, Demand Metric 2014 Content & the Buyer's Journey Benchmark Study — vendor-sponsored online opinion survey of 185 marketers; the "2× engagement" headline rounds 70%/36%) remain as un-defended as the calculator brief found them.
• Convergence: Mechanisms 1-5 reinforce each other. This convergence, drawn from multiple independent literatures, is the robust core.
• Recurring limit: nearly every effect is moderated — inverted-U for curiosity (Curiosity follows an INVERTED-U over prior knowledge/confidence — peaks at MODERATE knowing, falls when one knows nearly nothing or nearly everything (Kang 2009, Dubey-Griffiths 2020, Lee 2024)); unstable challenge-skill for flow (Løvoll & Vittersø (2014), Social Indicators Research — neither flow indicator peaked at balance; supports an IMBALANCE model; Engeser-Rheinberg 2008 also found balance not always optimal); choice-overload and cultural caveats for SDT (Patall, Cooper & Robinson (2008), Psychological Bulletin (41 studies meta) — choice enhances intrinsic motivation, effort, performance, perceived competence; moderated (2-4 choices, no extrinsic reward, children > adults), SDT cultural-universality critique (Hagger et al. 2013) — autonomy's primacy may reflect Western individualism; collectivist participants sometimes show higher intrinsic motivation under authority direction; SDT defenders reply autonomy ≠ independence); material-complexity ceiling for generation (Text-generation meta-analysis (Educational Psychology Review 2023) — Hedges g ≈ .41; LARGEST for 301-600 word texts; NO EFFECT beyond ~900 words, CAVEAT — 2025 conceptual replication (Cognitive Research: Principles and Implications) — generation effect did NOT reliably transfer to learning from expository text; some experiments showed disadvantage); too-much-interactivity cost (COUNTER-finding: Oh & Sundar 2015 also showed modality interactivity REDUCED the number of message-related thoughts — absorption can come at the cost of deep elaboration; Sundar warns of "too much interactivity"); strong conditionality for variable reward (Shen, Fishbach & Hsee (2015), JCR 41(5) — Motivating-Uncertainty Effect: people invest MORE effort for an uncertain reward (50% $2 / 50% $1) than for certain HIGHER-expected-value reward — but ONLY under PROCESS focus).
• None is an unconditional "more = better" lever.
• Settled items NOT re-derived: Kivetz 2006 goal-gradient (Kivetz, Urminsky & Zheng (2006), Journal of Marketing Research — goal-gradient in consumer contexts: cafe loyalty stamps completed faster as customers neared reward; online raters persist longer near reward); Ainslie 1975 present bias (Ainslie (1975) + Phelps & Pollak (1968) — present bias / hyperbolic temporal discounting: people overvalue immediate rewards relative to delayed ones); discredited vendor stats (Mediafly / Demand Metric: "Interactive content shows 52.6% higher engagement than static; buyers spend 13 vs 8.5 minutes" — vendor sources, treat as marketing not fact, Demand Metric 2014 Content & the Buyer's Journey Benchmark Study — vendor-sponsored online opinion survey of 185 marketers; the "2× engagement" headline rounds 70%/36%).

Capture-layer status: these are sourced research notes only — Deliverable 1 only. The brief author explicitly declined to produce a Deliverable 2 (KB entry), Deliverable 3 (article), or Deliverable 4 (slug) for this top-up.

Cross-brief: sibling of Caveats for the interactive-tool-mechanisms brief: lead on mechanism evidence (peer-reviewed, independent); treat vendor outcome stats (52.6% / 88% / 47.3%) as marketing (Brief E), Caveats for the customer-facing-calculators brief: every conversion-lift figure is unproven; nearly all are vendor-self-reported (Brief A), Caveats for the data-driven-tools brief: vendor self-reporting on conversion; enterprise-scale benchmarks; named-user quotes; macro projections, Caveats for the client-portals brief: source-incentives are pervasive; the independent anchors are McKinsey and Gartner; market-size figures unreliable; the viral 42% stat is misattributed, Caveats for the dashboards brief: pervasive BI/embedded-analytics vendor sourcing; the viral "60-70%" stat is folklore; SMB data thin; retention claims unproven.