The swipe is the product.

A sub-1% match-to-date conversion rate is not the result of an imperfect matching technology. It is the output of an engagement architecture that has been optimized for session time, not outcomes. The two goals are structurally in tension: a platform that efficiently matches users to dates empties itself. A platform that maximizes engagement extends the pre-match phase indefinitely.

Variable ratio reinforcement is the key mechanism. Skinner (1957) demonstrated that unpredictable reward schedules produce more persistent and compulsive behavior than fixed schedules. A slot machine pays off on a variable ratio schedule. So does a dating app match. The match arrives on an unpredictable timeline — which is precisely what makes the swipe compulsive.

Tinder internal documents (WSJ 2019) revealed deliberate match suppression for new users. The "Elo score" — a desirability ranking — determined who saw your profile and when. Users had no visibility into this ranking. The system was designed to extend the engagement window before delivering the first reward, calibrating the variable ratio schedule to maximize early retention.

The result: users who would have dated in week one instead spent weeks one through four swiping. The conversion rate fell. The session count rose. From a behavioral metric standpoint, this is an unambiguous success. From a user standpoint, it is O=3 operating at full strength.

Dating app algorithms are not recommendation systems optimizing for relationship compatibility. They are engagement systems optimizing for session time and notification response rates. The distinction matters because it determines what gets reinforced and what gets suppressed.

ELO-style desirability scores create a two-tier market invisible to users:

• High-Elo users receive early, frequent matches — high reward rate, high engagement
• Low-Elo users receive infrequent, delayed matches — variable ratio schedule, even higher engagement per reward

Neither tier has visibility into their score. Neither can meaningfully act to change their algorithmic standing. The user attributes the match rate to their personal desirability (D1 — agency attribution error) rather than to algorithmic positioning they cannot see and cannot influence.

Notification design follows the same logic. Push notifications are not sent when matches are most likely to convert to conversations — they are sent when the model predicts the user is most likely to open the app. Return engagement, not outcome quality, is the optimization target.

This is R=3 in practice: the platform is unresponsive to user intent (finding a date) while remaining highly responsive to its own optimization target (session count). The two are in direct conflict, and the platform's response to user complaints is to offer a paid tier that partially restores visibility — converting the opacity into a revenue mechanism.

The void framework's drift cascade proceeds through three stages. In dating apps, all three are empirically documented:

D1 — Agency attribution error: Users attribute poor match rates to personal failure — unattractiveness, poor profile copy, wrong photo selection. The algorithm's role is invisible. This attribution error is consequential: users respond to perceived personal deficiency rather than platform design. They update their self-presentation, lower their standards under engagement pressure, and increase usage (more swipes = better results, they believe). The opacity enables and amplifies the misattribution.

D2 — Boundary erosion: Average heavy-user session time reaches 2 hours per day. Natural stopping points are absent by design — infinite scroll eliminates the "end of deck" signal that would cue disengagement. Notification pressure creates felt obligation to respond. Users report that the platform "feels like a part-time job." The engagement architecture has dissolved the boundary between purposive dating behavior and compulsive scrolling.

D3 — Documented harm:

• Tinder study (2016, N=1,307): Tinder users reported significantly lower self-worth and higher body shame than non-users, particularly among women
• Three-fold higher rates of depression in heavy users vs non-users (multiple studies, 2018-2022)
• Body dysmorphic disorder symptom amplification — the photo-selection feedback loop creates an optimization pressure on physical appearance that has no natural stopping point
• Relationship formation delay: heavy users report lower rates of committed relationship formation over 18-month follow-up periods, controlling for baseline relationship intent

• P1 Platforms with visible match algorithms show lower session times but higher date conversion rates — natural experiment testable if any major platform implements algorithm transparency as a product feature.
• P2 Pe should correlate with self-reported app compulsivity at ρ > 0.7 across platforms — testable with Bergen Social Media Addiction Scale adapted for dating apps, administered to users of multiple platforms simultaneously.
• P3 Match suppression A/B tests show engagement lift at cost of outcome satisfaction — internal experiment data, if disclosed under regulatory pressure, would show this trade-off directly. UK CMA investigation into Match Group creates a disclosure pathway.
• P4 Disclosing ELO scores to users reduces session time by >20% — predicted by the opacity mechanism: D1 attribution error requires opacity to persist. Removing it shifts user behavior from compulsive swiping toward deliberate profile management.
• P5 Pe proxy (session time × return rate) correlates with depression outcome measures at ρ > 0.6 in longitudinal cohort study — testable against existing data from large-scale digital health studies with app-usage logging.

Paper 16 applies the void framework to major dating applications. It covers: Tinder's architecture history (pre- and post-Match Group acquisition), Bumble's differentiation attempt, Hinge's "designed to be deleted" positioning (and its tension with revenue incentives), and the cross-platform Pe comparison.

Key empirical findings: Pe=18.4 for peak-design dating apps places the domain between gambling (Pe→∞) and social media (Pe≈22). The α dimension is the primary driver — matching is a variable ratio schedule applied to a domain with high intrinsic motivation (relationship formation). This combination produces stronger behavioral conditioning than either gambling or social media alone, because the reward (a real relationship) is both deeply valued and episodically delivered.

The paper also analyzes the business model tension: platforms are paid subscriptions seeking retention, but their stated value proposition is outcomes (dates, relationships). A platform that succeeds at its value proposition loses its revenue base. This structural conflict is resolved by optimizing for engagement over outcomes — which is the void architecture in practice.

See all papers on Zenodo →