How it learns
The maps are the moat.
bnd is built to do more than route data. It is designed to accumulate a structured understanding of who needs human data, who produces it, what it contains, and what actually clears.
What the system learns from, and what it never touches.
The system learns from
- Permitted analysis and the metadata it produces
- Buyer requirements, acceptance decisions, and rejection reasons
- Transaction outcomes, pricing that cleared, renewals
- Source patterns: which kinds of organisations produce which signals
It never learns from
- Suppliers' private raw data is not training material for bnd
- Confidential buyer research stays outside the maps
- Nothing a visitor submits through this website is used to train models
The four maps in depth.
Demand Map
Every requirement teaches what AI organisations genuinely need: capabilities models lack, domains in demand, formats, quality thresholds, rights buyers can accept, budgets, and, just as valuable, why proposals get rejected. Example entry: evaluation teams in one domain consistently require verified outcomes; unverified archives are rejected at proposal stage.
Reality Map
Every discovery conversation teaches where signals are naturally produced: which industries, workflows, and professional groups generate which sequences as a by-product of ordinary work. Example entry: quality-control operations in precision manufacturing produce expert-overrule judgements that no public dataset contains.
Signal Map
Every inspection teaches what useful structures look like inside raw material: how failure and correction sequences present in tickets versus code review versus lab notes, and what refinement each needs. Example entry: support archives usually contain resolution sequences but rarely record confirmation; outcome capture is the common refinement.
Transaction Memory
Every completed or rejected opportunity teaches what the market validates: packaging that reduced friction, rights that were acceptable, pricing that cleared, sources that produced consistent value. Status: future compounding layer. No bnd transaction has completed yet, so this map holds no real records; the first genuine transaction creates the first one. The system learns from transaction metadata and approved analysis, never from suppliers' private raw archives.
The worked example from the homepage
Demand Map
What AI organisations genuinely request: missing capabilities, domains, formats, quality thresholds, acceptable rights, budgets, and rejection reasons.
Reality Map
Where those signals are naturally produced: industries, organisations, workflows, professional groups, and linguistic contexts.
Signal Map
What useful structures exist inside raw material: failure and correction sequences, interventions, preferences, outcomes, and rare edge cases.
Transaction Memory
What the market actually validates: what was purchased, rejected, refined, renewed, and what pricing cleared.
Worked exampleA buyer rejects a proposed dataset because its outcomes are unverified.
Why this compounds.
- A buyer requirement arrives
- The Reality Map suggests where to look first
- The Signal Map predicts what refinement the source needs
- Transaction Memory prices and packages it faster
- The result, either way, updates all four maps
A competitor can copy this website. It will not be able to copy the maps, because they will be built from the transactions that clear here and nowhere else.