Recursive Field Theory
A conclusion built on five independent sources and one built on a single shaky chain can end up looking identical once compressed. RFT is our framework for compressing information without losing the trail that got you there.
Compression usually destroys the thing that made a conclusion trustworthy.
Most systems compress signals into lossy summaries: they discard ancestry, flatten material contradictions into a single stated truth, and drop the observer's context — who concluded what, and under what assumptions. Two summaries can read identically even when one rests on solid ground and the other doesn't.
Contract. Trace. Expand.
Contract
Minimise a problem down to its smallest useful primitives, target conditions and constraints.
Trace
Follow the relationships that actually fired through the system, preserving domain crossings, uncertainty and provenance along the way.
Two nodes with the same headline aren't the same node.
Every node in an RFT structure carries five things, not just a value: the value itself, its morphology (the shape of the evidence behind it — depth, redundancy, bottlenecks, source diversity), its ancestry (where it came from), its uncertainty (what's disputed or estimated), and its scope (whose frame it's valid in). A confident-looking answer built on a fragile support graph gets flagged for review even when its stated confidence is high.
The stated answer itself.
Deep enough to explain, high enough to stay relevant.
Rather than analysing every scale at once, RFT deliberately looks three levels down for root cause and three levels up for constraint — a fixed n−3 to n+3 window across causal substrate, the task's own operational boundary, and the systems above it.
The operational boundary RFT is reasoning inside.
The same discipline governs a swarm of agents.
Unstructured groups of AI agents degrade fast — contaminated context, loop failures, memory that quietly goes wrong. RFT's runtime gives every task a defined role, an owner and a verifiable evidence path. A Sender packages the minimum context a task needs; a Receiver runs its own checks before accepting work. Three separate decisions — whether to send it, whether to accept the result, and whether to save it into permanent memory — are never collapsed into one. An answer can be shown on screen and still be rejected from ever being saved.
WA-OFIS: tracking where money moves through an economy.
The clearest live test of RFT so far is the WA Opportunity Flow Intelligence platform — a statewide tool that tracks where money enters, accumulates and leaves the Western Australian economy, and surfaces hyper-local gaps between supply and demand. It follows capital through four stages — announced, committed, flowing, capturable — and keeps observed fact strictly separate from forecast, so a large "announced" number is never mistaken for money that's actually moving.