Selection Engine
934 researched ideas, ranked before build.
The factory does not start with code. It starts by deciding where an AI-native product has a real chance to become a business.
934 Targets
4.7 Avg Score
24 Starred
3 Ready for Spec
What the Dashboard Scores
The goal is simple: build fewer random apps and spend more attention on opportunities with real unit economics.
Buyer
Who pays and how reachable that buyer is.
Max ARR
Can this become a meaningful product, not only a toy?
Market + Users
Market size, user volume, urgency, and frequency.
Build Days
How quickly the factory can ship the first useful version.
Blockers
Legal, platform, data, model, or trust risks that can stop the idea.
LTV/CAC
Whether paid acquisition can plausibly become profitable.
Dashboard View
The actual internal dashboard lets the founder browse, compare, annotate, shortlist, and move ideas through lifecycle stages.
| Rank | Buyer | Max ARR | Market | Users | Days | LTV/CAC | Name & Description |
|---|---|---|---|---|---|---|---|
| 7.31 | Prosumer | $540M | $7.4B | 50M | 5d | 9.0x | Monday.com alternative Simpler single-user project manager with strong incumbent pain. |
| 7.25 | B2B | $104M | $4.6B | 2M | 7d | 13.4x | Offline forms and PDF generator A simple tool for businesses frustrated by legacy form software. |
| 7.21 | B2B | $135M | $5B | 5M | 6d | 7.2x | Automatic project scheduler Modern scheduling workflow where AI can reduce manual planning. |
Selection Engine Flow
The selection block is a repeatable evidence system. It turns messy idea sources into ranked build decisions, then improves from every shipped product and paid test.
934 Candidate Ideas
▼
Phase 0
Intake & Target Record
Ideas, trends, user complaints, competitor changes, deck signals, and founder notes become one structured record: buyer, pain, job, incumbent, pricing hints, source links, and blocker hypotheses.
▼
Phase 1
Wide Evidence Scan
Specialized agents collect independent evidence before anything is built. The goal is to prove demand, pain, feasibility, and a clear GTM angle early.
Round 1 — Cast a Wide Net
Each lane checks one part of the opportunity. The system keeps sources, confidence, and contradictions instead of only writing a summary.
User Voices
pain, Reddit, reviews
Incumbent Pain
pricing, churn, gaps
Demand
keywords, trends, CPC
Pricing
ARPU, LTV/CAC
Build & Tech
days, APIs, AI fit
GTM Angle
ads, channels, hook
Risk & Blockers
legal, platform, trust
Financial Intel
TAM, users, Max ARR
Competitive Intel
positioning, wedge
Evidence Packets Collected
Round 2 — Fill Gaps & Find Contradictions
Follow-up agents go deeper only where the first scan is weak. They check missing data, conflicting claims, fatal blockers, and whether the opportunity fits the factory.
Missing Data
search gaps
Contradictions
conflicting evidence
Fatal Blockers
hard stop checks
Unit Economics
paid test logic
Founder Notes
taste and context
Clean Target Record
Phase 2
Scoring Kernel
The engine ranks buyer type, Max ARR, market size, user count, build days, blockers, LTV/CAC, and strategy fit. Fatal blockers can override the score.
▼
Phase 3
Founder Gate
Human taste stays in the loop before capital is committed. Eugene reviews the evidence, product wedge, AI-native fit, and whether the factory can sell it.
Stop
Defer
Star
Spec / Build
Phase 4
Self-Learning Feedback Loop
Ad performance, activation, support notes, retention, Deck Review outputs, shipped-product wins, and failures update the selection rules. The more products the factory tests, the better the next selection becomes.
Structured output
Every target has a score, evidence links, blocker flags, lifecycle stage, and next recommended action.
Human review
Founder taste stays in the loop before capital is committed to a build or paid test.
Self-learning loop
Outcomes from shipped products, Deck Review, ads, and support notes become new ranking evidence.
Productized Proof
Selection Engine is already becoming reusable product infrastructure.
Deck Review
A product built on this framework that researches a startup or deck and returns an investor-style brief. S16 tested it and confirmed the framework is useful.
BookScout
First consumer product proof point: vision pipeline, enrichment, pricing, and shareable AI interaction in one small app.
DoneCo
Internal AI project management layer for coordinating founder tasks, product tasks, agent reports, blockers, and metrics.