Live config loaded — values shown in teal reflect what is currently deployed on this server.
AGS Documentation
Back to Dashboard

Adaptive Governance System

AGS sits alongside a participant during a guided online task, watches their interaction in real time, estimates how they are coping, and decides — at every interaction — whether to offer support, what kind to offer, and whether it helped.

Research context. AGS is deployed in Kenya to study adaptive digital support for citizens aged 60+ navigating government websites for the first time. The platform is a research instrument, not a commercial product. Every design decision exists to either protect participants or produce valid data.
Documentation status
Canonical human-readable system documentation. Last reviewed against the local source tree on 2026-06-12.
Implementation status
Core participant flow, signal capture, readiness, governance, adaptation, dashboard monitoring, exports, and audit logging are implemented. Some evaluation dashboards depend on study data volume.
Machine contract
Use /openapi.json as the API contract. This page explains intent, evidence, limits, and research operations.

The problem AGS solves

When an older adult tries to use eCitizen or Afya Yangu for the first time, the support available to them is almost nothing — a FAQ page, a helpline that puts them on hold. There is no system that watches what they are actually doing, notices when they are stuck, and responds in the moment.

AGS is that system. It observes interaction signals (typing speed, errors, hesitation, help-seeking, backtracking), builds a live estimate of the participant's readiness, runs that estimate through a 17-condition decision model, and delivers a hint, a prompt, or a walkthrough — or stays silent — based on the outcome.

What the system adapts

AGS does adapt
  • Intervention messages — hints, guided prompts, walkthroughs
  • Support intensity — light tooltip → guided assistance → full walkthrough
  • Language — help content in Kiswahili, Kiluhya, or English
  • Timing — waits, delays, or fires immediately based on state
  • Rest prompts — when fatigue signals cross a threshold
AGS does not adapt
  • The government websites themselves — untouched
  • Language preference — set once at onboarding
  • Task steps or instructions — fixed per template
  • Anything about the study condition — invisible to participants

What makes it a research instrument

Three study conditions run simultaneously: full adaptive support, a more liberal version, and a control that logs signals but shows no adaptation. Participants cannot tell which condition they are in — all overlays look identical. This is what allows the study to isolate the effect of adaptive support from other factors.

The governance system — the 17 rules — is what makes this rigorous. A simple threshold (e.g., "help if score < 40") would adapt but wouldn't be auditable or reproducible. Every decision AGS makes is logged with the reason code and all rules evaluated, including decisions to stay silent.