Introducing

Delfa

A relationship-first dating app

One meaningful match at a time.
Built to replace swipe fatigue with depth, honesty, and follow-through.

01 / 18
The Problem

Dating apps are broken

78%
report dating app burnout
Forbes Health 2025
71%
say lying on profiles is "very common"
Pew Research 2020
41%
have been ghosted by a match
Forbes Health 2025
9/10
feel disappointed by people they see on apps
Pew Research 2023

What apps optimize for

  • Endless browsing and addictive loops
  • Superficial, photo-first evaluation
  • Monetizing loneliness and FOMO
  • Engagement volume over outcomes

What users actually want

  • Intentional, focused dating
  • Trust and verified authenticity
  • Genuine compatibility signals
  • Clear intent — less burnout
02 / 18
Why Now

The science gap

"There is no strong evidence that dating app algorithms can predict relationship success."
— Finkel et al. meta-analysis, Psychological Science in the Public Interest

The strongest predictors of lasting relationships — communication patterns, conflict repair, responsiveness, attachment security — can't be measured by a profile. They emerge through interaction.

Delfa's insight

Don't just match better.

Create better conditions for discovering compatibility.

Pre-meeting algorithms have a ceiling. Delfa's advantage is in the interaction stage — guiding how matched people communicate and discover chemistry.

03 / 18
How Delfa Works

The core model

Sign Up → Complete Basics (3-5 min)
One active match surfaced
Guided Getting-To-Know-You Prompts
Progress together
Mutual Ready-to-Meet
Before We Meet Protocol
Date ✓
Not the right fit
Graceful Disconnect
Calibration improves next match
Better Next Match ↺
Every path generates signal that improves future matching
04 / 18
The Profiler

Designed for honesty,
not performance

No "rate yourself 1-10." No virtue-signaling prompts. The profiler uses scenarios, forced tradeoffs, rankings, and behavioral recall to capture honest signal.

Scenarios
Forced tradeoffs
Rankings
Behavioral recall
Pairwise calibration

Compatibility confidence is shown by dimension — never a single misleading percentage.

Basics
Enough signal to start matching
3-5 min
Life Direction
Goals, family, structure, pace
Communication & Repair
Conflict patterns, responsiveness
Emotional Style & Security
Attachment, self-regulation
Lifestyle, Intimacy & Chemistry
Physical fit, intimacy values
▲ completed over time, in any order
05 / 18
Compatibility

Confidence, not a score

Instead of "87% match," Delfa shows what it knows — and what it doesn't yet.

Compatibility Dimensions
Intent
high
Life Direction
high
Pace
med
Responsiveness
med
Attachment
low
Lifestyle
high
Chemistry
med
Conflict & Repair
low
Intimacy
low
Stress & Finance
med
Improve your confidence

Conflict & Repair: low confidence
Complete "Communication & Repair" →

Transparency builds trust. Trust improves outcomes.
How compatibility is framed can shape user behavior as much as the underlying model.
06 / 18
Match Stage

Guided getting-to-know-you prompts

Each match gets a small number of shared prompts that help both people move beyond an empty chat box and into a more natural conversation.

15-30s
per prompt
24h
to respond — async
mutual
revealed only after both answer

The point is not to replace chat. The prompts give both people an easier and more meaningful way to begin.

Example Shared Prompt

You're both planning a friend's surprise birthday. You have £200 and 3 hours.

What do you prioritize?

You
Waiting...
Them
Waiting...

Both answers reveal simultaneously

07 / 18
Private Signal

Readiness check-in

Delfa should occasionally help users reflect on whether they are genuinely in a good place to date right now.

Private

Visible to the user, not shown to matches as a score

Lightweight

Short prompts that help users pause, slow down, or keep matching intentionally

Non-judgmental

Not a diagnostic test. Not a badge. Not something that punishes honesty.

How it should feel

Honest and private. More like a self-check than an assessment.

Example prompt

"How available do you feel for dating right now?" → Ready to invest • Open but pacing myself • I need a pause

08 / 18
In-Match

Soft commitment,
never forced

Delfa should encourage users to genuinely try the match without ever making them feel trapped in it.

Three clear exit paths

`Leave now` is always available. `Graceful disconnect` unlocks after meaningful interaction. `Block / report` bypasses everything.

No hostage UX
No forced dates
No punitive lockouts
Leave now if you're not feeling it
Use graceful disconnect after enough shared interaction
Block or report immediately if something feels unsafe
09 / 18
Signature Feature

Graceful disconnect

Instead of ghosting, one-tap respectful closure. The outcome still improves future matching, but the user experience stays simple.

Respectful closure

Close the match in one step. Curated feedback options — never freeform, preventing abuse.

Better calibration

Delfa uses curated closure reasons privately to improve the quality and fit of future matches.

Respectful closure norm

Delfa should make respectful closure simpler and safer than disappearing, without forcing users into rigid scripts or punitive mechanics.

Grounded in: 41% ghosting rate • self-awareness and responsiveness = top relationship predictors
10 / 18
Transition to IRL

Mutual ready-to-meet

Both users privately indicate when they feel ready to meet. Only when both have signaled does Delfa reveal it — simultaneously.

You
Them

You're both ready to meet!

Removes the ask-out vulnerability asymmetry. No one has to go first.

Before We Meet Protocol

Where current apps stop helping.
Delfa keeps going.

Mutual expectations check
Both select what they're hoping for — see each other's answers
Safety setup
Check-in reminder, trusted contact options, simple safety tools
Date planning assist
Lightweight support for format, area, and logistics
11 / 18
V1 Direction

Keep it simple

Delfa should differentiate itself with a few strong systems, not a stack of clever features that make dating feel like software training.

What stays in V1

One match at a time, progressive profiling, guided prompts, graceful disconnect, simple confidence, readiness check-in, mutual ready-to-meet, and a lightweight pre-date flow.

What gets deferred
Removed from V1

Slow reveal profile mechanics

Removed from V1

Conversation nudges and coaching layers

Removed from V1

Private pattern engines and heavy self-analysis features

Removed from V1

Relationship operating manual features

12 / 18
Core Pillar

Trust & safety

A product pillar, not a moderation afterthought. Safety is part of the product, not just the policy layer.

Identity verification

Progressive verification and authenticity checks, applied in stages to increase trust without adding unnecessary onboarding friction.

Verification should be strong, but the exact gating model still needs product definition.
Pre-send AI nudges

Detect potentially offensive content before it's sent. Change behavior upstream.

Useful as a safety mechanic; exact vendor metrics should not be treated as Delfa assumptions.
Image protection

AI detection and blurring of unsolicited explicit images. User chooses to view or block.

Useful as a safety mechanic; exact vendor metrics should not be treated as Delfa assumptions.
Date safety

Timer-based auto-alerts (better than panic buttons), location sharing, venue suggestions.

LGBTQ+ protection

Discreet mode, location obfuscation in hostile jurisdictions, minimal data sharing with third parties.

High
Safety remains one of the strongest unmet needs in the category and should be treated as a core product advantage.
13 / 18
Product Guardrails

What Delfa
will never be

A swipe casino
A dating entertainment feed
A pay-for-attention marketplace
A black-box compatibility scam
A rejection-scoring machine
A social events product
If a feature increases engagement but reduces honesty, clarity, or trust — reject it.
14 / 18
Architecture

System architecture

Modular monolith · Three runtimes · One PostgreSQL database

Client
React Native Mobile
Edge
HTTPS ingress
TLS · Routing
API Service
Fastify + TypeBox
REST/JSON · Module APIs
Worker Service
Outbox + pg-boss
Timers · Trust · Rechecks
Realtime Gateway
WebSocket fanout
Chat · Presence · Prompts
PostgreSQL
Drizzle ORM
+ pgvector · RLS
Redis
Cache · Fanout
Rate limits · Presence
Object Storage
Object Storage
Media · Verification
External Services
Clerk
Auth & sessions
OneSignal
Push notifications
Stripe
Payments
Sentry
Errors · traces (OTel)
Hosted on Fly.io · Single region · Staging + Production
15 / 18
Module Boundaries

Backend modules

11 bounded domains inside one codebase

identity
Accounts, auth sessions, device registration
profile
Public data, media, location
profiler
Sections, responses, mirror moments
attraction
Pairwise decks, preference vectors
matching
Eligibility, ranking, quality floor
match_lifecycle
State machine, guided rounds, disconnect
outcomes
Post-date reflections, learning signals
trust_safety
Reports, moderation, trust tiers
billing
Subscriptions, entitlements, receipts
measurement
Events, activation, quality metrics
ops_admin
Launch review, trust tools, queues
Each module owns its tables and write paths — no cross-module writes
Transactional outbox + worker model — no event bus at launch
Modular monolith first — extract only when scale proves the need
16 / 18
Research Foundation

Every decision
backed by evidence

Delfa's product decisions are grounded in a 1,100+ line research brief synthesizing peer-reviewed relationship science, large-scale surveys, and regulatory data.

PNAS 2020
43 longitudinal couples studies
Finkel
Definitive meta-analysis on dating algorithms
Pew
2020, 2023 U.S. online dating reports
Forbes
1,000-person burnout survey, 2025
FTC
Enforcement data on dating app practices
Profiler Science Basis
Commitment & relationship intent
Attachment security
Conflict repair capacity
Partner responsiveness
Emotional stability
Values & life-goal alignment
Stress & financial coping

Products that emphasize better dates and better outcomes suggest the market can reward a more intentional dating model.

17 / 18
Roadmap

What comes next

Shared contracts for match lifecycle and prompts
Wire the simplified state model and guided prompt structure into shared types
Confidence model contract
Wire profiler sections to dimensional confidence scoring
Graceful disconnect taxonomy
Curated closure reasons and lightweight calibration signals
Ready-to-meet & before-we-meet contracts
Mutual readiness detection, pre-date protocol flows
API and mobile implementation
Wire shared contracts into apps/api and apps/mobile

Delfa's advantage is not just better matching.
It is a better structure for how real relationships begin.

Shared profiler code already implemented in packages/shared/
18 / 18