01Step 01
Machine recommendation
DO NOT PROGRESS
Reflective Intelligence Infrastructure · Designed for transferability
Govern the human judgement layer between information and action.
RILayer governs what happens between what a machine tells us and what an accountable human actually decides to do.
What that looks like in practice
An AI system flags a case: “do not proceed”. Before anyone acts, RILayer makes the basis visible — what evidence supports the flag, what is missing or uncertain, what needs escalation, and which authorised person makes the final decision and records why.
Decision environment
Before action proceeds
1
MACHINE / SYSTEM SIGNAL
2
EVIDENCE & UNCERTAINTY
3
HUMAN REVIEW / ESCALATION
4
ACCOUNTABLE ACTION
The machine signal remains visible. RILayer does not make the final consequential decision.
See the control layer working
What happens when a machine recommends “Do not progress”?
Chloe Taylor is a fictional adult early-career candidate in the Youth Transition Demonstrator YT-WM-001. RILayer does not simply accept or replace the simulated ATS recommendation. It makes the judgement process inspectable and keeps the original machine input visible.
This is a controlled proof-of-concept, not a live candidate decision or claimed client outcome.
Four-step proof
Machine signal → governed state → evidence change → accountable human action
- 02Step 02
Governance intervention
NOT READY FOR AN ADVERSE FINAL DECISION
- 03Step 03
Evidence change
CHRONOLOGY RECONCILED
- 04Step 04
Human action
PROCEED TO STRUCTURED INTERVIEW
The enterprise gap
Faster systems have not removed decision risk.
AI generates outputs at scale. Dashboards expose more signals. Managers face pressure to act. The exposed point remains the human judgement moment before action.
More output
AI, data and systems increase the volume and speed of signals.
More pressure
Teams are expected to act faster, often with incomplete context.
Same accountability
Humans and organisations remain responsible for what proceeds.
The decision-noise filter
Make the transition from noise to governed action visible.
Information, pressure, AI output and competing priorities can arrive together. RILayer creates a governed point where evidence, uncertainty, routing and ownership can be examined before action proceeds.
Decision noise
Pressure, assumptions and competing signals
RILayer
Governed interpretation
Governed execution
Clearer escalation and accountable action
01
Separate signal from noise
02
Apply evidence and escalation controls
03
Keep action human-owned and reviewable
Designed for transferability
One governance mechanism. Bounded proof now. Wider applications to be tested.
RILayer is currently demonstrated through a bounded Youth Transition proof-of-concept. Controlled live validation in an operating environment has not yet begun. Wider applications represent future validation pathways, not claims of completed deployment.
Future validation contexts
These contexts show where the architecture may later be adapted. They are intentionally subordinate to the current Youth Transition proof and should not be read as equal-status products.
Human–AI governance
Operational escalation
Financial risk
Health and care
Public and statutory-adjacent services
Education and workforce systems
4D implementation path
Discover → Discern → Develop → Dedicate
The current browser demonstrator focuses on DISCERN. The wider 4D path shows where the mechanism sits without claiming the full lifecycle has been validated through YT-WM-001.
- 01
Discover
Clarify the real decision environment and the human question before governed RILayer begins. - 02
Discern
Expose evidence, assumptions, readiness, governance risk, boundaries and authority before action. - 03
Develop
Build the capability, evidence or support needed where responsible action should not proceed yet. - 04
Dedicate
Embed accountable practice into routine workflows, review cycles and organisational ownership.
Next conversation
Start with a bounded decision problem.
An initial conversation can test whether the problem has been characterised clearly enough to justify a bounded discovery, validation or pilot conversation.