Technical PoCMarketplace
Exchange optimisation

Finding the swap
that simple search misses.

This working proof of concept ranks directed item compatibility, builds a graph of acceptable exchanges, and identifies two- to five-party cycles in which every participant receives an item they requested.

Demo items9Synthetic listings
Directed pairs72Pairs evaluated
Eligible edges15Score threshold ≥ 58
Cycles found4Unique ranked cycles
Ranked exchange plans

From bilateral swaps to coordinated cycles

Each arrow means the owner of the item on the left is willing to receive the next item.

2-party cycle97.1/100
BenStrategy board game bundle→
EvaAcoustic travel guitar
↺ closes the cycle
Average edge 97.8Minimum edge 97.3Value spread 6.1%
3-party cycle96.6/100
LeoCompact reading chair→
AbbasOlympus film camera→
SaraStudio headphones
↺ closes the cycle
Average edge 97.4Minimum edge 96.9Value spread 8.3%
4-party cycle93.2/100
AmirFilter coffee maker→
NoraCity bicycle→
OskarsFantasy book collection→
MilaLarge indoor monstera
↺ closes the cycle
Average edge 97Minimum edge 95.9Value spread 43%
Explainable scoring

Every score can be inspected

Example direction: Leo offers Compact reading chair and evaluates Olympus film camera. Hard constraints are applied before a weighted 100-point score.

Preference fit40/40

Matched requested signals: Photography, camera.

Value fit20/20

Estimated value €120 is inside the preferred €95–€150 band.

Condition fit9.3/10

Very good meets the minimum condition of Good.

Logistics fit14.8/15

Compatible through local exchange at 1.05 km.

Trust fit8.5/10

Uses normalized verification, completion and response signals from the item owner.

Freshness4.7/5

Listing age is 6 days.

Development status

What exists now and what comes next

Implemented in current MVP

Marketplace workflow

Accounts, listings, search, saved items, bilateral offers, private trade chat, completion, reviews, reports and administration.

Implemented in technical PoC

Exchange engine

Hard-constraint filtering, explainable directed ranking, graph construction, unique cycle detection, fairness adjustment and deterministic tests.

Planned future R&D

Production optimisation

Consented preference learning, fraud-resistant value confidence, privacy controls, constrained solver benchmarks and monitored marketplace integration.