Clarity under constraint
Current development
Fairlead
Progressive delivery of structured AI model payloads across constrained and intermittent links.
Threshold Signalworks is developing Fairlead for environments where a complete transfer may be slow, expensive or repeatedly interrupted. The receiver can move through authenticated, progressively useful states as more of the payload arrives.
The problem
Many model-delivery workflows make completion the point at which a new model state becomes usable. That becomes costly when contact time is short, bandwidth is limited, or a link may disappear before the transfer finishes.
Fairlead is built around the state of the receiver. It keeps track of what has been authenticated, which lineage and sequence the accepted payload belongs to, and the capability floor the receiver can rely on. Transfer can stop and resume without treating partial arrival, verified state and usable state as the same thing.
At the receiver
Progressive usefulness
The payload becomes useful in stages. Each accepted stage can advance what the receiving system is able to use.
Authenticated lineage
The receiver maintains explicit state about the payload lineage and candidate sequence it has accepted.
Capability floor
Accepted state has a defined floor. The receiver can establish the minimum payload state it is entitled to use.
Interruption expected
Delivery can pause and resume while preserving the distinction between received bytes and authenticated, usable state.
Where it fits
Satellite and orbital edge compute. Remote and robotic systems. Constrained edge devices. Other deployments where bandwidth, contact time or transfer cost shape what can be used at the far end.
Current status
A reference implementation and test harness are under active development, with the core receiver-state behaviour implemented and tested. The next phase is focused on evaluation in satellite, edge-device and related constrained-link environments.
The European patent application covers progressive transmission of structured AI model payloads. Detailed technical material can be shared for evaluation where appropriate.
Fairlead evaluation and partnership enquiries
Earlier work
Keel
Keel began as an internal tool for examining how a coding agent should be constrained at the point where it can act. It developed into a governance-native agent harness with a separate policy boundary and tamper-evident audit record, and was released publicly for others who might find it useful.
Keel remains available as an open-source reference and learning project. Current company development is Fairlead. Project on GitHub.
About
Threshold Signalworks Ltd is an independent Irish R&D company founded by Brian McCallion. Current company development is centred on Fairlead.
Brian's independent academic and experimental research is published separately at Threshold Systems.