Research
Working artefacts - patient personas, deep-eval runs, and version-over-version diagnostic reports - used to drive Olly's triage and product decisions. Useful for shared reference, but not committed-to product specs.
→ Persona Index - table of all 9 personas with verdicts (PASS / FAIL_OVERESCALATED / FAIL_BAND_MISMATCH) and links to transcripts.
Eval reports
Run-level reports across the persona suite - what worked, what regressed, what changed between versions.
First end-to-end pass of the nine-persona suite - baseline verdicts and timings.
Drill-down on the sore-throat scenario surfaced by the baseline report.
Second pass of the sore-throat persona after pipeline adjustments.
Run with richer patient narratives - checks robustness to extra context.
Targets band-mismatch failures by tightening the self-care/pharmacy/gp ladder.
Persona Index
Nine synthetic patients ranging from a 5-year-old earache to a 70-year-old chest pain. Each persona was driven through the triage pipeline end-to-end and the transcript captured.
Mum reports 5-year-old's right ear pain. Self-care disposition.
Teenager with raspy voice and sore throat. Self-care disposition.
Two-hour bad headache routed to 999 instead of the expected ladder.
70-year-old with stinging on urination. Got self-care, expected GP/111.
Mum reports new rash on 5-year-old's tummy and arms. Self-care disposition.
Football injury with swelling and pain on weight-bearing. Self-care disposition.
Tight, crushing central chest pain. Correctly routed to 999.
70-year-old with stairs-induced breathlessness. Routed to 999 instead of GP/111.
Table of every persona with demographic, complaint, expected vs got, and verdict.