Catchment map and bias governance

Where a person lives changes what the data can see. This panel makes social context and model equity visible instead of leaving them implicit.

Synthetic data
Observed data — population

Catchment map

Synthetic neighbourhoods shaded by area deprivation decile. Select an area to see its care-navigation flags.

Riverbend Eastdecile 2 · 2 case(s)Millgatedecile 4 · 1 case(s)Harbour Northdecile 8 · 1 case(s)Southfield Greendecile 6 · 0 case(s)Alder Valedecile 3 · 0 case(s)
Deprivation most deprived least deprivedSynthetic geography — not a real place.
Care navigation

Riverbend East

Automatic navigation flags raised for this synthetic area.

Transport access
low
Language support need
high
Connectivity
68% household coverage (synthetic)
Auto-flag: arrange transport and interpreter support before discharge planning.

Navigation flags describe services to consider with the person. They are not triage, and they never decide care.

System-generated signal — read only

Bias and fairness auditing board

Model performance by demographic segment on synthetic evaluation data. Read-only: changes are made through the Governance Studio.

monitored each review cycle

Age band

synthetic evaluation set
  • 18–39recall 84% · false alerts 9% · n=1240
  • 40–64recall 82% · false alerts 11% · n=3110
  • 65–79recall 78% · false alerts 14% · n=2480
  • 80+recall 69% · false alerts 19% · n=940

Recall falls in the oldest band, where baseline vital signs differ most from the training population.

Recorded race or ethnicity group

synthetic evaluation set
  • Group Arecall 83% · false alerts 10% · n=3600
  • Group Brecall 76% · false alerts 15% · n=1180
  • Group Crecall 71% · false alerts 17% · n=640
  • Not recordedrecall 66% · false alerts 21% · n=350

Segments with fewer synthetic records show wider uncertainty. Visual grading is a known weakness across skin tones.

Sex or gender recorded

synthetic evaluation set
  • Femalerecall 81% · false alerts 12% · n=3900
  • Malerecall 80% · false alerts 12% · n=3700
  • Other or not recordedrecall 72% · false alerts 18% · n=170

Difference is small in this synthetic extract but is monitored each cycle rather than assumed stable.

Area deprivation decile

synthetic evaluation set
  • Decile 1–3 (most deprived)recall 72% · false alerts 18% · n=1900
  • Decile 4–7recall 80% · false alerts 13% · n=3400
  • Decile 8–10recall 85% · false alerts 9% · n=2470

Lower recall in the most deprived deciles tracks with sparser monitoring data, not with biology.

⚠️ Make sure to double-check the physical patient presentation to confirm this information.

Human-in-the-loop

Override activity

Overrides recorded by nurses feed the equity review: repeated overrides in one segment are a bias signal.

No overrides recorded yet in this browser.

Educational sim-environment. Do not enter real patient information. AI content supports learning and reflection; it does not replace clinical judgment, local policy, or escalation pathways.