How an Implementation Is Built

Every such marketplace is assembled from three roles — two reusable, one the sponsor’s. Read the four diagrams in order: each zooms in one level, then the last pulls back to show why the model scales.

A working thin-market exchange is not one monolithic product. It is a composition of three distinct responsibilities. The trick that makes the economics work is that two of the three are generic — built once and reused across every market vertical — while only the third is bespoke to a particular market and its sponsor.

The diagrams below use a single color throughout so the same components are recognizable as they reappear at each level of zoom. Learn the three colors once; they carry the whole story.

The DeeperPoint thesis: the frictions that keep markets thin are common across thousands of markets — so the engine that neutralizes them is built once and reused. A sponsor adds only their market’s business logic, instead of rediscovering the same problems alone.

Curatorial & feeder systems

Domain knowledge wiki, reference library, and a synthetic population. Feeds real-world data into the marketplace. Owned by curators / a specialist setup service.

Matching engine

Schema compiler, candidate matching + scoring, deal-brief assembly, precompute. The reusable engine — built once, replicated per vertical. Drawn white with a dotted edge in the diagrams to mark it as the shared, reused element.

Market-vertical layer

Branded front-end UX plus auth, ads, payments, piggybacked services. Owned by the market sponsor.

Figure 1 · 30,000 ft

The anatomy of an implementation

Three roles, left to right. The feeder systems supply domain knowledge and a synthetic population; the matching engine turns that into matches, deal briefs and a cache; the market-vertical layer presents it to the people in that specific market. The middle block is identical from one vertical to the next. The feeder↔engine link is two-way: matching results feed gaps back so the knowledge, population, and weights keep improving.

1 · FEEDER SYSTEMS Domain knowledge wiki Reference knowledge library Synthetic population + synthetic-data watermark Curators / forging service 2 · MATCHING ENGINE Schema + vocab compiler Matching engine Deal brief + precompute match story · static cache Open-source engine — built once, replicated 3 · MARKET-VERTICAL LAYER Branded front-end UX Auth · ads · payments Piggybacked services market-specific logic Market sponsor data + twin feedback matches + briefs

The third role is the only one that changes between markets. A sponsor may build it themselves or buy it from a paid market-forging service — either way, roles 1 and 2 are reused unchanged.

Figure 2 · 10,000 ft

How the three roles operate together

The same three colors, now as swimlanes showing the operational flow for one vertical. Domain knowledge and a watermarked synthetic population feed the engine; the engine matches, scores, assembles deals and writes a static cache; the sponsor’s front end reads that cache. Cross-lane arrows show where one role hands off to the next.

CURATORIAL / FEEDER MATCHING ENGINE VERTICAL LAYER Domain wiki knowledge capture Reference library curated knowledge Synthetic data population + watermark Schema + vocab config · field weights Ingest + embed canonical + raw Match funnel gate → rank → weighted score Deal assembly facilitator sub-deals Match story generative narrative Precompute → static cache synthetic profiles grounds the story ⇅ data + feedback SPONSOR FRONT END (reads the cache) Branded front-end UX Auth Ads Payments Piggybacked services reads cache

Everything the engine does is deterministic, which is what lets the whole run be precomputed once into a static cache — a fully precomputed deployment runs at zero per-visitor cost. (See Figure 3.)

Figure 3 · ground level

Inside the matching engine

Zooming into the matching engine. The engine ingests the population, runs a two-stage match funnel (a cheap deterministic gate, then semantic ranking and weighted per-role scoring), assembles deals with facilitator sub-deals, and generates the match story. Because every step is deterministic, the run can be materialized — computed once on write, then read cheap — into two projections: a match cache (the deal/story experience; Mode 1 serves it, Modes 2–3 compute live) and a public profile / gallery cache (a public-only view that powers live galleries and search).

Synthetic population + watermark verify Schema + vocab fields · weights Reference library domain grounding MATCHING ENGINE A · Ingest validate → embed (canonical + raw) → index participants B · Match funnel B1 · structured gate deterministic, cacheable B2 · semantic rank vector similarity weighted score per-role + slots C · Deal assembly facilitator injection → sub-deals → deal brief D · Match story generative narrative (privacy gate) ⇅ feedback loop match gaps refine each feeder MATERIALIZATION LAYER A · Match cache full scenario JSON Mode 1 serves this cache Mode 2/3 compute live B · Profile / gallery cache PUBLIC-ONLY projection regenerated on write → galleries + search Vertical front end deal/story · galleries + search

Both projections share one mechanism but differ in contract: the match cache may use private fields and (in the demo) is frozen synthetic data; the profile / gallery cache is a public-only projection of real profiles, regenerated per record on write. Rule of thumb: precompute the gallery, compute the match. The feeder↔engine links are two-way: match gaps feed back to improve the reference library (knowledge), the synthetic population (coverage), and the matching weights — a continuous-improvement loop.

Figure 4 · the payoff

Why it scales: the same engine, replicated per vertical

Pull all the way back. Three different markets, side by side — and each is a complete, independent stack: its own knowledge feed (violet), its own copy of the matching engine (white), and its own storefront (amber). The engines look identical because they are the same open-source build, replicated — not because they share one running instance. That is the economic argument: the expensive engine is written once, then cloned per vertical, so launching the next market means authoring only a feed and a front end.

≡ all three engines are one build, replicated per vertical INDEPENDENT DEPLOYMENT #1 Market vertical A domain wiki · reference library + synthetic population Matching engine schema · matching + scoring deal brief · precompute open source Vertical A front-end UX + auth · ads · services market sponsor #1 👥 vertical A participants INDEPENDENT DEPLOYMENT #2 Market vertical B domain wiki · reference library + synthetic population Matching engine schema · matching + scoring deal brief · precompute open source Vertical B front-end UX + auth · ads · services market sponsor #2 👥 vertical B participants INDEPENDENT DEPLOYMENT #3 Market vertical C domain wiki · reference library + synthetic population Matching engine schema · matching + scoring deal brief · precompute open source Vertical C front-end UX + auth · ads · services market sponsor #3 👥 vertical C participants

Each vertical deploys its own engine today — the same open-source components, separate instances. A single multi-tenant engine serving a “constellation” of marketplaces is a possible future, not the starting model. The reuse is in the build, not a shared runtime: you clone the reference implementation rather than reinvent it. Each deployment also runs its own feeder↔engine feedback loop, so every vertical’s knowledge, population, and matching improve independently.

The choice

Two ways to build a market

Strip everything else away and this is the decision a market builder faces — rediscover the same frictions alone, or start from a solution already grounded in the theory of why markets stay thin. The reusable engine carries any pairing as far as the Deal Brief — the last step common to every market — and the sponsor’s own layer (contracting, payment, fulfillment) takes it from there.

Build it alone · trial and error
  • Pick a vertical, build a platform from scratch.
  • Discover each friction only after you hit it.
  • See it through your own buyers and sellers — miss the pattern.
  • Reinvent the same wheels every other builder is reinventing.
  • Succeed or fail — alone.
Start from the theory · systematic
  • The frictions are already named — common and Nobel-defined.
  • Deploy the open-source engine that neutralizes them.
  • Build only your market’s business logic on top.
  • Improve continuously through the feeder↔engine loop.
  • Replicate a proven solution — you are not alone.

Where to go next

Figures 1–4 are the build-side view. To see what a participant experiences inside one of these markets — from anonymous discovery to a Deal Brief — read How It Works, or open the live demo.

Try the Interactive Demo The Participant Journey → Back to the Project Suite
The DeeperPoint story · Step 5 of 5
That’s the logic, end to end. Now see it run.
See the demo →